One more translation update

Reject Index and Request messages with unexpected flags

.gitignore

@@ -4,9 +4,14 @@ syncthing.exe
 *.zip
 *.asc
 *.sublime*
+.idea/
 .jshintrc
 coverage.out
 files/pidx
 bin
 perfstats*.csv
 coverage.xml
+!gui/scripts/syncthing
+.DS_Store
+syncthing.md5
+syncthing.exe.md5

.mailmap

@@ -0,0 +1 @@
+NICKS

AUTHORS

@@ -1,25 +1,39 @@
+# This is the official list of Syncthing authors for copyright purposes.
+
 Aaron Bieber <qbit@deftly.net>
 Alexander Graf <register-github@alex-graf.de>
 Andrew Dunham <andrew@du.nham.ca>
 Audrius Butkevicius <audrius.butkevicius@gmail.com>
 Arthur Axel fREW Schmidt <frew@afoolishmanifesto.com> <frioux@gmail.com>
+Ben Schulz <ueomkail@gmail.com> <uok@users.noreply.github.com>
 Ben Sidhom <bsidhom@gmail.com>
 Brandon Philips <brandon@ifup.org>
+Brendan Long <self@brendanlong.com>
 Caleb Callaway <enlightened.despot@gmail.com>
+Cathryne Linenweaver <cathryne.linenweaver@gmail.com> <Cathryne@users.noreply.github.com>
 Chris Joel <chris@scriptolo.gy>
 Daniel Martí <mvdan@mvdan.cc>
+Dennis Wilson <dw@risu.io>
+Dominik Heidler <dominik@heidler.eu>
+Emil Hessman <emil@hessman.se>
+Federico Castagnini <federico.castagnini@gmail.com>
 Felix Ableitner <me@nutomic.com>
 Felix Unterpaintner <bigbear2nd@gmail.com>
 Gilli Sigurdsson <gilli@vx.is>
+Jakob Borg <jakob@nym.se>
 James Patterson <jamespatterson@operamail.com> <jpjp@users.noreply.github.com>
 Jens Diemer <github.com@jensdiemer.de> <git@jensdiemer.de>
 Jochen Voss <voss@seehuhn.de>
 Lode Hoste <zillode@zillode.be>
 Marcin Dziadus <dziadus.marcin@gmail.com>
 Michael Tilli <pyfisch@gmail.com>
+Peter Hoeg <peter@speartail.com>
 Philippe Schommers <philippe@schommers.be>
 Phill Luby <phill.luby@newredo.com>
+Piotr Bejda <piotrb10@gmail.com>
 Ryan Sullivan <kayoticsully@gmail.com>
+Tim Abell <tim@timwise.co.uk>
+Tomas Cerveny <kozec@kozec.com>
 Tully Robinson <tully@tojr.org>
 Veeti Paananen <veeti.paananen@rojekti.fi>
 Vil Brekin <vilbrekin@gmail.com>

CONDUCT.md

@@ -0,0 +1,92 @@
+## Conduct
+
+* We are committed to providing a friendly, safe and welcoming
+  environment for all, regardless of gender, sexual orientation,
+  disability, ethnicity, religion, or similar personal characteristic.
+
+* On IRC, please avoid using overtly sexual nicknames or other nicknames
+  that might detract from a friendly, safe and welcoming environment for
+  all.
+
+* Please be kind and courteous. There's no need to be mean or rude.
+
+* Respect that people have differences of opinion and that every design
+  or implementation choice carries a trade-off and numerous costs. There
+  is seldom a right answer.
+
+* Please keep unstructured critique to a minimum. If you have solid
+  ideas you want to experiment with, make a fork and see how it works.
+
+* We will exclude you from interaction if you insult, demean or harass
+  anyone. That is not welcome behaviour. We interpret the term
+  "harassment" as including the definition in the <a
+  href="http://citizencodeofconduct.org/">Citizen Code of Conduct</a>;
+  if you have any lack of clarity about what might be included in that
+  concept, please read their definition. In particular, we don't
+  tolerate behavior that excludes people in socially marginalized
+  groups.
+
+* Private harassment is also unacceptable. No matter who you are, if you
+  feel you have been or are being harassed or made uncomfortable by a
+  community member, please contact one of the channel ops or any of the
+  Syncthing core team immediately. Whether you're a regular  contributor
+  or a newcomer, we care about making this community a safe place for
+  you and we've got your back.
+
+* Likewise any spamming, trolling, flaming, baiting or other
+  attention-stealing behaviour is not welcome.
+
+## Moderation
+
+These are the policies for upholding our community's standards of
+conduct in our communication channels, most notably in Syncthing-related
+IRC channels and on the web forum.
+
+1. Remarks that violate the Syncthing standards of conduct, including
+   hateful, hurtful, oppressive, or exclusionary remarks, are not
+   allowed. (Cursing is allowed, but never targeting another user, and
+   never in a hateful manner.)
+
+2. Remarks that moderators find inappropriate, whether listed in the
+   code of conduct or not, are also not allowed.
+
+3. Moderators will first respond to such remarks with a warning.
+
+4. If the warning is unheeded, the user will be "kicked," i.e., kicked
+   out of the communication channel to cool off.
+
+5. If the user comes back and continues to make trouble, they will be
+   banned, i.e., indefinitely excluded.
+
+6. Moderators may choose at their discretion to un-ban the user if it
+   was a first offense and they offer the offended party a genuine
+   apology.
+
+7. If a moderator bans someone and you think it was unjustified, please
+   take it up with that moderator, or with a different moderator, **in
+   private**. Complaints about bans in-channel are not allowed.
+
+8. Moderators are held to a higher standard than other community
+   members. If a moderator creates an inappropriate situation, they
+   should expect less leeway than others.
+
+In the Syncthing community we strive to go the extra step to look out
+for each other. Don't just aim to be technically unimpeachable, try to
+be your best self. In particular, avoid flirting with offensive or
+sensitive issues, particularly if they're off-topic; this all too
+often leads to unnecessary fights, hurt feelings, and damaged trust;
+worse, it can drive people away from the community entirely.
+
+And if someone takes issue with something you said or did, resist the
+urge to be defensive. Just stop doing what it was they complained about
+and apologize. Even if you feel you were misinterpreted or unfairly
+accused, chances are good there was something you could've communicated
+better — remember that it's your responsibility to make your fellow
+community members comfortable. Everyone wants to get along and we are
+all here first and foremost because we want to talk about cool
+technology. You will find that people will be eager to assume good
+intent and forgive as long as you earn their trust.
+
+*Adapted from the [Rust Code of Conduct](https://github.com/rust-lang/rust/wiki/Note-development-policy#conduct)*
+
+*Adapted from the [Node.js Policy on Trolling](http://blog.izs.me/post/30036893703/policy-on-trolling)*

CONTRIBUTING.md

@@ -31,24 +31,65 @@ latest info on Transifex.
 
 ## Contributing Code
 
-Please do contribute! If you want to contribute but are unsure where to
-start, the [Contributions Needed
-topic](http://discourse.syncthing.net/t/49) lists areas in need of
-attention. In general, any open issues are fair game! Be prepared for a
-[certain amount of
-review](https://discourse.syncthing.net/t/733); it's all in the name of
-quality. :)
+Every contribution is welcome. If you want to contribute but are unsure
+where to start, any open issues are fair game! Be prepared for a
+[certain amount of review](https://discourse.syncthing.net/t/733); it's
+all in the name of quality. :) Following the points below will make this
+a smoother process.
 
-You may be asked to "rebase" or "squash" your pull request before it's
-accepted. This means to make sure that the pull request:
+Individuals making significant and valuable contributions are given
+commit-access to the project. If you make a significant contribution and
+are not considered for commit-access, please contact any of the
+Syncthing core team members.
 
- - Is based on the latest commit on `master`.
+All nontrivial contributions should go through the pull request
+mechanism for internal review. Determining what is "nontrivial" is left
+at the discretion of the contributor.
 
- - Only contains a single commit (exceptions to be made where
-   appropriate).
+### Authorship
 
- - Has a descriptive commit message, ending in `(fixes #123)` if it
-   resolves an existing issue.
+All code authors are listed in the AUTHORS file. Commits must be made
+with the same name and email as listed in the AUTHORS file. To
+accomplish this, ensure that your git configuration is set correctly
+prior to making your first commit;
+
+    $ git config --global user.name "Jane Doe"
+    $ git config --global user.email janedoe@example.com
+
+You must be reachable on the given email address. If you do not wish to
+use your real name for whatever reason, using a nickname or pseudonym is
+perfectly acceptable.
+
+### Core Team
+
+The Syncthing core team currently consists of the following members;
+
+ - Jakob Borg (@calmh)
+ - Audrius Butkevicius (@AudriusButkevicius)
+
+## Coding Style
+
+- Follow the conventions laid out in [Effective Go](https://golang.org/doc/effective_go.html)
+  as much as makes sense.
+
+- All text files use Unix line endings.
+
+- Each commit should be `go fmt` clean.
+
+- The commit message subject should be a single short sentence
+  describing the change, starting with a capital letter.
+
+- Commits that resolve an existing issue must include the issue number
+  as `(fixes #123)` at the end of the commit message subject.
+
+- Imports are grouped per `goimports` standard; that is, standard
+  library first, then third party libraries after a blank line.
+
+- A contribution solving a single issue or introducing a single new
+  feature should probably be a single commit based on the current
+  `master` branch. You may be asked to "rebase" or "squash" your pull
+  request to make sure this is the case, especially if there have been
+  amendments during review.
 
 ## Licensing
 
@@ -59,8 +100,8 @@ International License. You retain the copyright to code you have
 written.
 
 When accepting your first contribution, the maintainer of the project
-will ensure that you are added to the CONTRIBUTORS file. You are welcome
-to add yourself as a separate commit in your first pull request.
+will ensure that you are added to the AUTHORS file. You are welcome to
+add yourself as a separate commit in your first pull request.
 
 ## Building
 
@@ -90,12 +131,6 @@ signed by GPG key BCE524C7.
 
 Yes please!
 
-## Style
-
- - `go fmt`
-
- - Unix line breaks
-
 ## Documentation
 
 [Over here!](http://discourse.syncthing.net/category/documentation)

Godeps/Godeps.json

@@ -1,35 +1,10 @@
 {
 	"ImportPath": "github.com/syncthing/syncthing",
-	"GoVersion": "go1.3.3",
+	"GoVersion": "go1.4",
 	"Packages": [
 		"./cmd/..."
 	],
 	"Deps": [
-		{
-			"ImportPath": "code.google.com/p/go.crypto/bcrypt",
-			"Comment": "null-216",
-			"Rev": "41cd4647fccc72b0b79ef1bd1fe6735e718257cd"
-		},
-		{
-			"ImportPath": "code.google.com/p/go.crypto/blowfish",
-			"Comment": "null-216",
-			"Rev": "41cd4647fccc72b0b79ef1bd1fe6735e718257cd"
-		},
-		{
-			"ImportPath": "code.google.com/p/go.text/transform",
-			"Comment": "null-90",
-			"Rev": "d65bffbc88a153d23a6d2a864531e6e7c2cde59b"
-		},
-		{
-			"ImportPath": "code.google.com/p/go.text/unicode/norm",
-			"Comment": "null-90",
-			"Rev": "d65bffbc88a153d23a6d2a864531e6e7c2cde59b"
-		},
-		{
-			"ImportPath": "code.google.com/p/snappy-go/snappy",
-			"Comment": "null-15",
-			"Rev": "12e4b4183793ac4b061921e7980845e750679fd0"
-		},
 		{
 			"ImportPath": "github.com/AudriusButkevicius/lfu-go",
 			"Rev": "164bcecceb92fd6037f4d18a8d97b495ec6ef669"
@@ -48,7 +23,7 @@
 		},
 		{
 			"ImportPath": "github.com/calmh/xdr",
-			"Rev": "ec3d404f43731551258977b38dd72cf557d00398"
+			"Rev": "214788d8fedfc310c18eca9ed12be408a5054cd5"
 		},
 		{
 			"ImportPath": "github.com/juju/ratelimit",
@@ -56,7 +31,11 @@
 		},
 		{
 			"ImportPath": "github.com/syndtr/goleveldb/leveldb",
-			"Rev": "0d8857b7ec571b0a6c9677d8e6c0a4ceeabd1d71"
+			"Rev": "63c9e642efad852f49e20a6f90194cae112fd2ac"
+		},
+		{
+			"ImportPath": "github.com/syndtr/gosnappy/snappy",
+			"Rev": "ce8acff4829e0c2458a67ead32390ac0a381c862"
 		},
 		{
 			"ImportPath": "github.com/vitrun/qart/coding",
@@ -69,6 +48,22 @@
 		{
 			"ImportPath": "github.com/vitrun/qart/qr",
 			"Rev": "ccb109cf25f0cd24474da73b9fee4e7a3e8a8ce0"
+		},
+		{
+			"ImportPath": "golang.org/x/crypto/bcrypt",
+			"Rev": "731db29863ea7213d9556d0170afb38987f401d4"
+		},
+		{
+			"ImportPath": "golang.org/x/crypto/blowfish",
+			"Rev": "731db29863ea7213d9556d0170afb38987f401d4"
+		},
+		{
+			"ImportPath": "golang.org/x/text/transform",
+			"Rev": "985ee5acfaf1ff6712c7c99438752f8e09416ccb"
+		},
+		{
+			"ImportPath": "golang.org/x/text/unicode/norm",
+			"Rev": "985ee5acfaf1ff6712c7c99438752f8e09416ccb"
 		}
 	]
 }

Godeps/_workspace/src/code.google.com/p/go.text/unicode/norm/maketesttables.go

@@ -1,45 +0,0 @@
-// Copyright 2011 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-// +build ignore
-
-// Generate test data for trie code.
-
-package main
-
-import (
-	"fmt"
-)
-
-func main() {
-	printTestTables()
-}
-
-// We take the smallest, largest and an arbitrary value for each
-// of the UTF-8 sequence lengths.
-var testRunes = []rune{
-	0x01, 0x0C, 0x7F, // 1-byte sequences
-	0x80, 0x100, 0x7FF, // 2-byte sequences
-	0x800, 0x999, 0xFFFF, // 3-byte sequences
-	0x10000, 0x10101, 0x10FFFF, // 4-byte sequences
-	0x200, 0x201, 0x202, 0x210, 0x215, // five entries in one sparse block
-}
-
-const fileHeader = `// Generated by running
-//	maketesttables
-// DO NOT EDIT
-
-package norm
-
-`
-
-func printTestTables() {
-	fmt.Print(fileHeader)
-	fmt.Printf("var testRunes = %#v\n\n", testRunes)
-	t := newNode()
-	for i, r := range testRunes {
-		t.insert(r, uint16(i))
-	}
-	t.printTables("testdata")
-}

Godeps/_workspace/src/code.google.com/p/go.text/unicode/norm/trie.go

@@ -1,232 +0,0 @@
-// Copyright 2011 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-package norm
-
-type valueRange struct {
-	value  uint16 // header: value:stride
-	lo, hi byte   // header: lo:n
-}
-
-type trie struct {
-	index        []uint8
-	values       []uint16
-	sparse       []valueRange
-	sparseOffset []uint16
-	cutoff       uint8 // indices >= cutoff are sparse
-}
-
-// lookupValue determines the type of block n and looks up the value for b.
-// For n < t.cutoff, the block is a simple lookup table. Otherwise, the block
-// is a list of ranges with an accompanying value. Given a matching range r,
-// the value for b is by r.value + (b - r.lo) * stride.
-func (t *trie) lookupValue(n uint8, b byte) uint16 {
-	if n < t.cutoff {
-		return t.values[uint16(n)<<6+uint16(b)]
-	}
-	offset := t.sparseOffset[n-t.cutoff]
-	header := t.sparse[offset]
-	lo := offset + 1
-	hi := lo + uint16(header.lo)
-	for lo < hi {
-		m := lo + (hi-lo)/2
-		r := t.sparse[m]
-		if r.lo <= b && b <= r.hi {
-			return r.value + uint16(b-r.lo)*header.value
-		}
-		if b < r.lo {
-			hi = m
-		} else {
-			lo = m + 1
-		}
-	}
-	return 0
-}
-
-const (
-	t1 = 0x00 // 0000 0000
-	tx = 0x80 // 1000 0000
-	t2 = 0xC0 // 1100 0000
-	t3 = 0xE0 // 1110 0000
-	t4 = 0xF0 // 1111 0000
-	t5 = 0xF8 // 1111 1000
-	t6 = 0xFC // 1111 1100
-	te = 0xFE // 1111 1110
-)
-
-// lookup returns the trie value for the first UTF-8 encoding in s and
-// the width in bytes of this encoding. The size will be 0 if s does not
-// hold enough bytes to complete the encoding. len(s) must be greater than 0.
-func (t *trie) lookup(s []byte) (v uint16, sz int) {
-	c0 := s[0]
-	switch {
-	case c0 < tx:
-		return t.values[c0], 1
-	case c0 < t2:
-		return 0, 1
-	case c0 < t3:
-		if len(s) < 2 {
-			return 0, 0
-		}
-		i := t.index[c0]
-		c1 := s[1]
-		if c1 < tx || t2 <= c1 {
-			return 0, 1
-		}
-		return t.lookupValue(i, c1), 2
-	case c0 < t4:
-		if len(s) < 3 {
-			return 0, 0
-		}
-		i := t.index[c0]
-		c1 := s[1]
-		if c1 < tx || t2 <= c1 {
-			return 0, 1
-		}
-		o := uint16(i)<<6 + uint16(c1)
-		i = t.index[o]
-		c2 := s[2]
-		if c2 < tx || t2 <= c2 {
-			return 0, 2
-		}
-		return t.lookupValue(i, c2), 3
-	case c0 < t5:
-		if len(s) < 4 {
-			return 0, 0
-		}
-		i := t.index[c0]
-		c1 := s[1]
-		if c1 < tx || t2 <= c1 {
-			return 0, 1
-		}
-		o := uint16(i)<<6 + uint16(c1)
-		i = t.index[o]
-		c2 := s[2]
-		if c2 < tx || t2 <= c2 {
-			return 0, 2
-		}
-		o = uint16(i)<<6 + uint16(c2)
-		i = t.index[o]
-		c3 := s[3]
-		if c3 < tx || t2 <= c3 {
-			return 0, 3
-		}
-		return t.lookupValue(i, c3), 4
-	}
-	// Illegal rune
-	return 0, 1
-}
-
-// lookupString returns the trie value for the first UTF-8 encoding in s and
-// the width in bytes of this encoding. The size will be 0 if s does not
-// hold enough bytes to complete the encoding. len(s) must be greater than 0.
-func (t *trie) lookupString(s string) (v uint16, sz int) {
-	c0 := s[0]
-	switch {
-	case c0 < tx:
-		return t.values[c0], 1
-	case c0 < t2:
-		return 0, 1
-	case c0 < t3:
-		if len(s) < 2 {
-			return 0, 0
-		}
-		i := t.index[c0]
-		c1 := s[1]
-		if c1 < tx || t2 <= c1 {
-			return 0, 1
-		}
-		return t.lookupValue(i, c1), 2
-	case c0 < t4:
-		if len(s) < 3 {
-			return 0, 0
-		}
-		i := t.index[c0]
-		c1 := s[1]
-		if c1 < tx || t2 <= c1 {
-			return 0, 1
-		}
-		o := uint16(i)<<6 + uint16(c1)
-		i = t.index[o]
-		c2 := s[2]
-		if c2 < tx || t2 <= c2 {
-			return 0, 2
-		}
-		return t.lookupValue(i, c2), 3
-	case c0 < t5:
-		if len(s) < 4 {
-			return 0, 0
-		}
-		i := t.index[c0]
-		c1 := s[1]
-		if c1 < tx || t2 <= c1 {
-			return 0, 1
-		}
-		o := uint16(i)<<6 + uint16(c1)
-		i = t.index[o]
-		c2 := s[2]
-		if c2 < tx || t2 <= c2 {
-			return 0, 2
-		}
-		o = uint16(i)<<6 + uint16(c2)
-		i = t.index[o]
-		c3 := s[3]
-		if c3 < tx || t2 <= c3 {
-			return 0, 3
-		}
-		return t.lookupValue(i, c3), 4
-	}
-	// Illegal rune
-	return 0, 1
-}
-
-// lookupUnsafe returns the trie value for the first UTF-8 encoding in s.
-// s must hold a full encoding.
-func (t *trie) lookupUnsafe(s []byte) uint16 {
-	c0 := s[0]
-	if c0 < tx {
-		return t.values[c0]
-	}
-	if c0 < t2 {
-		return 0
-	}
-	i := t.index[c0]
-	if c0 < t3 {
-		return t.lookupValue(i, s[1])
-	}
-	i = t.index[uint16(i)<<6+uint16(s[1])]
-	if c0 < t4 {
-		return t.lookupValue(i, s[2])
-	}
-	i = t.index[uint16(i)<<6+uint16(s[2])]
-	if c0 < t5 {
-		return t.lookupValue(i, s[3])
-	}
-	return 0
-}
-
-// lookupStringUnsafe returns the trie value for the first UTF-8 encoding in s.
-// s must hold a full encoding.
-func (t *trie) lookupStringUnsafe(s string) uint16 {
-	c0 := s[0]
-	if c0 < tx {
-		return t.values[c0]
-	}
-	if c0 < t2 {
-		return 0
-	}
-	i := t.index[c0]
-	if c0 < t3 {
-		return t.lookupValue(i, s[1])
-	}
-	i = t.index[uint16(i)<<6+uint16(s[1])]
-	if c0 < t4 {
-		return t.lookupValue(i, s[2])
-	}
-	i = t.index[uint16(i)<<6+uint16(s[2])]
-	if c0 < t5 {
-		return t.lookupValue(i, s[3])
-	}
-	return 0
-}

Godeps/_workspace/src/code.google.com/p/go.text/unicode/norm/trie_test.go

@@ -1,152 +0,0 @@
-// Copyright 2011 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-package norm
-
-import (
-	"testing"
-	"unicode/utf8"
-)
-
-// Test data is located in triedata_test.go; generated by maketesttables.
-var testdata = testdataTrie
-
-type rangeTest struct {
-	block   uint8
-	lookup  byte
-	result  uint16
-	table   []valueRange
-	offsets []uint16
-}
-
-var range1Off = []uint16{0, 2}
-var range1 = []valueRange{
-	{0, 1, 0},
-	{1, 0x80, 0x80},
-	{0, 2, 0},
-	{1, 0x80, 0x80},
-	{9, 0xff, 0xff},
-}
-
-var rangeTests = []rangeTest{
-	{10, 0x80, 1, range1, range1Off},
-	{10, 0x00, 0, range1, range1Off},
-	{11, 0x80, 1, range1, range1Off},
-	{11, 0xff, 9, range1, range1Off},
-	{11, 0x00, 0, range1, range1Off},
-}
-
-func TestLookupSparse(t *testing.T) {
-	for i, test := range rangeTests {
-		n := trie{sparse: test.table, sparseOffset: test.offsets, cutoff: 10}
-		v := n.lookupValue(test.block, test.lookup)
-		if v != test.result {
-			t.Errorf("LookupSparse:%d: found %X; want %X", i, v, test.result)
-		}
-	}
-}
-
-// Test cases for illegal runes.
-type trietest struct {
-	size  int
-	bytes []byte
-}
-
-var tests = []trietest{
-	// illegal runes
-	{1, []byte{0x80}},
-	{1, []byte{0xFF}},
-	{1, []byte{t2, tx - 1}},
-	{1, []byte{t2, t2}},
-	{2, []byte{t3, tx, tx - 1}},
-	{2, []byte{t3, tx, t2}},
-	{1, []byte{t3, tx - 1, tx}},
-	{3, []byte{t4, tx, tx, tx - 1}},
-	{3, []byte{t4, tx, tx, t2}},
-	{1, []byte{t4, t2, tx, tx - 1}},
-	{2, []byte{t4, tx, t2, tx - 1}},
-
-	// short runes
-	{0, []byte{t2}},
-	{0, []byte{t3, tx}},
-	{0, []byte{t4, tx, tx}},
-
-	// we only support UTF-8 up to utf8.UTFMax bytes (4 bytes)
-	{1, []byte{t5, tx, tx, tx, tx}},
-	{1, []byte{t6, tx, tx, tx, tx, tx}},
-}
-
-func mkUTF8(r rune) ([]byte, int) {
-	var b [utf8.UTFMax]byte
-	sz := utf8.EncodeRune(b[:], r)
-	return b[:sz], sz
-}
-
-func TestLookup(t *testing.T) {
-	for i, tt := range testRunes {
-		b, szg := mkUTF8(tt)
-		v, szt := testdata.lookup(b)
-		if int(v) != i {
-			t.Errorf("lookup(%U): found value %#x, expected %#x", tt, v, i)
-		}
-		if szt != szg {
-			t.Errorf("lookup(%U): found size %d, expected %d", tt, szt, szg)
-		}
-	}
-	for i, tt := range tests {
-		v, sz := testdata.lookup(tt.bytes)
-		if v != 0 {
-			t.Errorf("lookup of illegal rune, case %d: found value %#x, expected 0", i, v)
-		}
-		if sz != tt.size {
-			t.Errorf("lookup of illegal rune, case %d: found size %d, expected %d", i, sz, tt.size)
-		}
-	}
-	// Verify defaults.
-	if v, _ := testdata.lookup([]byte{0xC1, 0x8C}); v != 0 {
-		t.Errorf("lookup of non-existing rune should be 0; found %X", v)
-	}
-}
-
-func TestLookupUnsafe(t *testing.T) {
-	for i, tt := range testRunes {
-		b, _ := mkUTF8(tt)
-		v := testdata.lookupUnsafe(b)
-		if int(v) != i {
-			t.Errorf("lookupUnsafe(%U): found value %#x, expected %#x", i, v, i)
-		}
-	}
-}
-
-func TestLookupString(t *testing.T) {
-	for i, tt := range testRunes {
-		b, szg := mkUTF8(tt)
-		v, szt := testdata.lookupString(string(b))
-		if int(v) != i {
-			t.Errorf("lookup(%U): found value %#x, expected %#x", i, v, i)
-		}
-		if szt != szg {
-			t.Errorf("lookup(%U): found size %d, expected %d", i, szt, szg)
-		}
-	}
-	for i, tt := range tests {
-		v, sz := testdata.lookupString(string(tt.bytes))
-		if int(v) != 0 {
-			t.Errorf("lookup of illegal rune, case %d: found value %#x, expected 0", i, v)
-		}
-		if sz != tt.size {
-			t.Errorf("lookup of illegal rune, case %d: found size %d, expected %d", i, sz, tt.size)
-		}
-	}
-}
-
-func TestLookupStringUnsafe(t *testing.T) {
-	for i, tt := range testRunes {
-		b, _ := mkUTF8(tt)
-		v := testdata.lookupStringUnsafe(string(b))
-		if int(v) != i {
-			t.Errorf("lookupUnsafe(%U): found value %#x, expected %#x", i, v, i)
-		}
-	}
-}

Godeps/_workspace/src/code.google.com/p/go.text/unicode/norm/triedata_test.go

@@ -1,85 +0,0 @@
-// Generated by running
-//	maketesttables
-// DO NOT EDIT
-
-package norm
-
-var testRunes = []int32{1, 12, 127, 128, 256, 2047, 2048, 2457, 65535, 65536, 65793, 1114111, 512, 513, 514, 528, 533}
-
-// testdataValues: 192 entries, 384 bytes
-// Block 2 is the null block.
-var testdataValues = [192]uint16{
-	// Block 0x0, offset 0x0
-	0x000c: 0x0001,
-	// Block 0x1, offset 0x40
-	0x007f: 0x0002,
-	// Block 0x2, offset 0x80
-}
-
-// testdataSparseOffset: 10 entries, 20 bytes
-var testdataSparseOffset = []uint16{0x0, 0x2, 0x4, 0x8, 0xa, 0xc, 0xe, 0x10, 0x12, 0x14}
-
-// testdataSparseValues: 22 entries, 88 bytes
-var testdataSparseValues = [22]valueRange{
-	// Block 0x0, offset 0x1
-	{value: 0x0000, lo: 0x01},
-	{value: 0x0003, lo: 0x80, hi: 0x80},
-	// Block 0x1, offset 0x2
-	{value: 0x0000, lo: 0x01},
-	{value: 0x0004, lo: 0x80, hi: 0x80},
-	// Block 0x2, offset 0x3
-	{value: 0x0001, lo: 0x03},
-	{value: 0x000c, lo: 0x80, hi: 0x82},
-	{value: 0x000f, lo: 0x90, hi: 0x90},
-	{value: 0x0010, lo: 0x95, hi: 0x95},
-	// Block 0x3, offset 0x4
-	{value: 0x0000, lo: 0x01},
-	{value: 0x0005, lo: 0xbf, hi: 0xbf},
-	// Block 0x4, offset 0x5
-	{value: 0x0000, lo: 0x01},
-	{value: 0x0006, lo: 0x80, hi: 0x80},
-	// Block 0x5, offset 0x6
-	{value: 0x0000, lo: 0x01},
-	{value: 0x0007, lo: 0x99, hi: 0x99},
-	// Block 0x6, offset 0x7
-	{value: 0x0000, lo: 0x01},
-	{value: 0x0008, lo: 0xbf, hi: 0xbf},
-	// Block 0x7, offset 0x8
-	{value: 0x0000, lo: 0x01},
-	{value: 0x0009, lo: 0x80, hi: 0x80},
-	// Block 0x8, offset 0x9
-	{value: 0x0000, lo: 0x01},
-	{value: 0x000a, lo: 0x81, hi: 0x81},
-	// Block 0x9, offset 0xa
-	{value: 0x0000, lo: 0x01},
-	{value: 0x000b, lo: 0xbf, hi: 0xbf},
-}
-
-// testdataLookup: 640 bytes
-// Block 0 is the null block.
-var testdataLookup = [640]uint8{
-	// Block 0x0, offset 0x0
-	// Block 0x1, offset 0x40
-	// Block 0x2, offset 0x80
-	// Block 0x3, offset 0xc0
-	0x0c2: 0x01, 0x0c4: 0x02,
-	0x0c8: 0x03,
-	0x0df: 0x04,
-	0x0e0: 0x02,
-	0x0ef: 0x03,
-	0x0f0: 0x05, 0x0f4: 0x07,
-	// Block 0x4, offset 0x100
-	0x120: 0x05, 0x126: 0x06,
-	// Block 0x5, offset 0x140
-	0x17f: 0x07,
-	// Block 0x6, offset 0x180
-	0x180: 0x08, 0x184: 0x09,
-	// Block 0x7, offset 0x1c0
-	0x1d0: 0x04,
-	// Block 0x8, offset 0x200
-	0x23f: 0x0a,
-	// Block 0x9, offset 0x240
-	0x24f: 0x06,
-}
-
-var testdataTrie = trie{testdataLookup[:], testdataValues[:], testdataSparseValues[:], testdataSparseOffset[:], 1}

Godeps/_workspace/src/code.google.com/p/go.text/unicode/norm/triegen.go

@@ -1,317 +0,0 @@
-// Copyright 2011 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-// +build ignore
-
-// Trie table generator.
-// Used by make*tables tools to generate a go file with trie data structures
-// for mapping UTF-8 to a 16-bit value. All but the last byte in a UTF-8 byte
-// sequence are used to lookup offsets in the index table to be used for the
-// next byte. The last byte is used to index into a table with 16-bit values.
-
-package main
-
-import (
-	"fmt"
-	"hash/crc32"
-	"log"
-	"unicode/utf8"
-)
-
-const (
-	blockSize        = 64
-	blockOffset      = 2 // Subtract two blocks to compensate for the 0x80 added to continuation bytes.
-	maxSparseEntries = 16
-)
-
-// Intermediate trie structure
-type trieNode struct {
-	table [256]*trieNode
-	value int
-	b     byte
-	leaf  bool
-}
-
-func newNode() *trieNode {
-	return new(trieNode)
-}
-
-func (n trieNode) String() string {
-	s := fmt.Sprint("trieNode{table: { non-nil at index: ")
-	for i, v := range n.table {
-		if v != nil {
-			s += fmt.Sprintf("%d, ", i)
-		}
-	}
-	s += fmt.Sprintf("}, value:%#x, b:%#x leaf:%v}", n.value, n.b, n.leaf)
-	return s
-}
-
-func (n trieNode) isInternal() bool {
-	internal := true
-	for i := 0; i < 256; i++ {
-		if nn := n.table[i]; nn != nil {
-			if !internal && !nn.leaf {
-				log.Fatalf("triegen: isInternal: node contains both leaf and non-leaf children (%v)", n)
-			}
-			internal = internal && !nn.leaf
-		}
-	}
-	return internal
-}
-
-func (n trieNode) mostFrequentStride() int {
-	counts := make(map[int]int)
-	v := 0
-	for _, t := range n.table[0x80 : 0x80+blockSize] {
-		if t != nil {
-			if stride := t.value - v; v != 0 && stride >= 0 {
-				counts[stride]++
-			}
-			v = t.value
-		} else {
-			v = 0
-		}
-	}
-	var maxs, maxc int
-	for stride, cnt := range counts {
-		if cnt > maxc || (cnt == maxc && stride < maxs) {
-			maxs, maxc = stride, cnt
-		}
-	}
-	return maxs
-}
-
-func (n trieNode) countSparseEntries() int {
-	stride := n.mostFrequentStride()
-	var count, v int
-	for _, t := range n.table[0x80 : 0x80+blockSize] {
-		tv := 0
-		if t != nil {
-			tv = t.value
-		}
-		if tv-v != stride {
-			if tv != 0 {
-				count++
-			}
-		}
-		v = tv
-	}
-	return count
-}
-
-func (n *trieNode) insert(r rune, value uint16) {
-	var p [utf8.UTFMax]byte
-	sz := utf8.EncodeRune(p[:], r)
-
-	for i := 0; i < sz; i++ {
-		if n.leaf {
-			log.Fatalf("triegen: insert: node (%#v) should not be a leaf", n)
-		}
-		nn := n.table[p[i]]
-		if nn == nil {
-			nn = newNode()
-			nn.b = p[i]
-			n.table[p[i]] = nn
-		}
-		n = nn
-	}
-	n.value = int(value)
-	n.leaf = true
-}
-
-type nodeIndex struct {
-	lookupBlocks []*trieNode
-	valueBlocks  []*trieNode
-	sparseBlocks []*trieNode
-	sparseOffset []uint16
-	sparseCount  int
-
-	lookupBlockIdx map[uint32]int
-	valueBlockIdx  map[uint32]int
-}
-
-func newIndex() *nodeIndex {
-	index := &nodeIndex{}
-	index.lookupBlocks = make([]*trieNode, 0)
-	index.valueBlocks = make([]*trieNode, 0)
-	index.sparseBlocks = make([]*trieNode, 0)
-	index.sparseOffset = make([]uint16, 1)
-	index.lookupBlockIdx = make(map[uint32]int)
-	index.valueBlockIdx = make(map[uint32]int)
-	return index
-}
-
-func computeOffsets(index *nodeIndex, n *trieNode) int {
-	if n.leaf {
-		return n.value
-	}
-	hasher := crc32.New(crc32.MakeTable(crc32.IEEE))
-	// We only index continuation bytes.
-	for i := 0; i < blockSize; i++ {
-		v := 0
-		if nn := n.table[0x80+i]; nn != nil {
-			v = computeOffsets(index, nn)
-		}
-		hasher.Write([]byte{uint8(v >> 8), uint8(v)})
-	}
-	h := hasher.Sum32()
-	if n.isInternal() {
-		v, ok := index.lookupBlockIdx[h]
-		if !ok {
-			v = len(index.lookupBlocks) - blockOffset
-			index.lookupBlocks = append(index.lookupBlocks, n)
-			index.lookupBlockIdx[h] = v
-		}
-		n.value = v
-	} else {
-		v, ok := index.valueBlockIdx[h]
-		if !ok {
-			if c := n.countSparseEntries(); c > maxSparseEntries {
-				v = len(index.valueBlocks) - blockOffset
-				index.valueBlocks = append(index.valueBlocks, n)
-				index.valueBlockIdx[h] = v
-			} else {
-				v = -len(index.sparseOffset)
-				index.sparseBlocks = append(index.sparseBlocks, n)
-				index.sparseOffset = append(index.sparseOffset, uint16(index.sparseCount))
-				index.sparseCount += c + 1
-				index.valueBlockIdx[h] = v
-			}
-		}
-		n.value = v
-	}
-	return n.value
-}
-
-func printValueBlock(nr int, n *trieNode, offset int) {
-	boff := nr * blockSize
-	fmt.Printf("\n// Block %#x, offset %#x", nr, boff)
-	var printnewline bool
-	for i := 0; i < blockSize; i++ {
-		if i%6 == 0 {
-			printnewline = true
-		}
-		v := 0
-		if nn := n.table[i+offset]; nn != nil {
-			v = nn.value
-		}
-		if v != 0 {
-			if printnewline {
-				fmt.Printf("\n")
-				printnewline = false
-			}
-			fmt.Printf("%#04x:%#04x, ", boff+i, v)
-		}
-	}
-}
-
-func printSparseBlock(nr int, n *trieNode) {
-	boff := -n.value
-	fmt.Printf("\n// Block %#x, offset %#x", nr, boff)
-	v := 0
-	//stride := f(n)
-	stride := n.mostFrequentStride()
-	c := n.countSparseEntries()
-	fmt.Printf("\n{value:%#04x,lo:%#02x},", stride, uint8(c))
-	for i, nn := range n.table[0x80 : 0x80+blockSize] {
-		nv := 0
-		if nn != nil {
-			nv = nn.value
-		}
-		if nv-v != stride {
-			if v != 0 {
-				fmt.Printf(",hi:%#02x},", 0x80+i-1)
-			}
-			if nv != 0 {
-				fmt.Printf("\n{value:%#04x,lo:%#02x", nv, nn.b)
-			}
-		}
-		v = nv
-	}
-	if v != 0 {
-		fmt.Printf(",hi:%#02x},", 0x80+blockSize-1)
-	}
-}
-
-func printLookupBlock(nr int, n *trieNode, offset, cutoff int) {
-	boff := nr * blockSize
-	fmt.Printf("\n// Block %#x, offset %#x", nr, boff)
-	var printnewline bool
-	for i := 0; i < blockSize; i++ {
-		if i%8 == 0 {
-			printnewline = true
-		}
-		v := 0
-		if nn := n.table[i+offset]; nn != nil {
-			v = nn.value
-		}
-		if v != 0 {
-			if v < 0 {
-				v = -v - 1 + cutoff
-			}
-			if printnewline {
-				fmt.Printf("\n")
-				printnewline = false
-			}
-			fmt.Printf("%#03x:%#02x, ", boff+i, v)
-		}
-	}
-}
-
-// printTables returns the size in bytes of the generated tables.
-func (t *trieNode) printTables(name string) int {
-	index := newIndex()
-	// Values for 7-bit ASCII are stored in first two block, followed by nil block.
-	index.valueBlocks = append(index.valueBlocks, nil, nil, nil)
-	// First byte of multi-byte UTF-8 codepoints are indexed in 4th block.
-	index.lookupBlocks = append(index.lookupBlocks, nil, nil, nil, nil)
-	// Index starter bytes of multi-byte UTF-8.
-	for i := 0xC0; i < 0x100; i++ {
-		if t.table[i] != nil {
-			computeOffsets(index, t.table[i])
-		}
-	}
-
-	nv := len(index.valueBlocks) * blockSize
-	fmt.Printf("// %sValues: %d entries, %d bytes\n", name, nv, nv*2)
-	fmt.Printf("// Block 2 is the null block.\n")
-	fmt.Printf("var %sValues = [%d]uint16 {", name, nv)
-	printValueBlock(0, t, 0)
-	printValueBlock(1, t, 64)
-	printValueBlock(2, newNode(), 0)
-	for i := 3; i < len(index.valueBlocks); i++ {
-		printValueBlock(i, index.valueBlocks[i], 0x80)
-	}
-	fmt.Print("\n}\n\n")
-
-	ls := len(index.sparseBlocks)
-	fmt.Printf("// %sSparseOffset: %d entries, %d bytes\n", name, ls, ls*2)
-	fmt.Printf("var %sSparseOffset = %#v\n\n", name, index.sparseOffset[1:])
-
-	ns := index.sparseCount
-	fmt.Printf("// %sSparseValues: %d entries, %d bytes\n", name, ns, ns*4)
-	fmt.Printf("var %sSparseValues = [%d]valueRange {", name, ns)
-	for i, n := range index.sparseBlocks {
-		printSparseBlock(i, n)
-	}
-	fmt.Print("\n}\n\n")
-
-	cutoff := len(index.valueBlocks) - blockOffset
-	ni := len(index.lookupBlocks) * blockSize
-	fmt.Printf("// %sLookup: %d bytes\n", name, ni)
-	fmt.Printf("// Block 0 is the null block.\n")
-	fmt.Printf("var %sLookup = [%d]uint8 {", name, ni)
-	printLookupBlock(0, newNode(), 0, cutoff)
-	printLookupBlock(1, newNode(), 0, cutoff)
-	printLookupBlock(2, newNode(), 0, cutoff)
-	printLookupBlock(3, t, 0xC0, cutoff)
-	for i := 4; i < len(index.lookupBlocks); i++ {
-		printLookupBlock(i, index.lookupBlocks[i], 0x80, cutoff)
-	}
-	fmt.Print("\n}\n\n")
-	fmt.Printf("var %sTrie = trie{ %sLookup[:], %sValues[:], %sSparseValues[:], %sSparseOffset[:], %d}\n\n",
-		name, name, name, name, name, cutoff)
-	return nv*2 + ns*4 + ni + ls*2
-}

Godeps/_workspace/src/github.com/calmh/logger/README.md

@@ -12,3 +12,4 @@ License
 -------
 
 MIT
+

Godeps/_workspace/src/github.com/calmh/xdr/cmd/genxdr/main.go

@@ -11,6 +11,8 @@ import (
 	"go/format"
 	"go/parser"
 	"go/token"
+	"io"
+	"log"
 	"os"
 	"regexp"
 	"strconv"
@@ -269,7 +271,7 @@ func handleStruct(t *ast.StructType) []fieldInfo {
 	return fs
 }
 
-func generateCode(s structInfo) {
+func generateCode(output io.Writer, s structInfo) {
 	name := s.Name
 	fs := s.Fields
 
@@ -286,7 +288,7 @@ func generateCode(s structInfo) {
 	if err != nil {
 		panic(err)
 	}
-	fmt.Println(string(bs))
+	fmt.Fprintln(output, string(bs))
 }
 
 func uncamelize(s string) string {
@@ -295,16 +297,16 @@ func uncamelize(s string) string {
 	})
 }
 
-func generateDiagram(s structInfo) {
+func generateDiagram(output io.Writer, s structInfo) {
 	sn := s.Name
 	fs := s.Fields
 
-	fmt.Println(sn + " Structure:")
-	fmt.Println()
-	fmt.Println(" 0                   1                   2                   3")
-	fmt.Println(" 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1")
+	fmt.Fprintln(output, sn+" Structure:")
+	fmt.Fprintln(output)
+	fmt.Fprintln(output, " 0                   1                   2                   3")
+	fmt.Fprintln(output, " 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1")
 	line := "+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+"
-	fmt.Println(line)
+	fmt.Fprintln(output, line)
 
 	for _, f := range fs {
 		tn := f.FieldType
@@ -312,52 +314,52 @@ func generateDiagram(s structInfo) {
 		name := uncamelize(f.Name)
 
 		if sl {
-			fmt.Printf("| %s |\n", center("Number of "+name, 61))
-			fmt.Println(line)
+			fmt.Fprintf(output, "| %s |\n", center("Number of "+name, 61))
+			fmt.Fprintln(output, line)
 		}
 		switch tn {
 		case "bool":
-			fmt.Printf("| %s |V|\n", center(name+" (V=0 or 1)", 59))
-			fmt.Println(line)
+			fmt.Fprintf(output, "| %s |V|\n", center(name+" (V=0 or 1)", 59))
+			fmt.Fprintln(output, line)
 		case "uint16":
-			fmt.Printf("| %s | %s |\n", center("0x0000", 29), center(name, 29))
-			fmt.Println(line)
+			fmt.Fprintf(output, "| %s | %s |\n", center("0x0000", 29), center(name, 29))
+			fmt.Fprintln(output, line)
 		case "uint32":
-			fmt.Printf("| %s |\n", center(name, 61))
-			fmt.Println(line)
+			fmt.Fprintf(output, "| %s |\n", center(name, 61))
+			fmt.Fprintln(output, line)
 		case "int64", "uint64":
-			fmt.Printf("| %-61s |\n", "")
-			fmt.Printf("+ %s +\n", center(name+" (64 bits)", 61))
-			fmt.Printf("| %-61s |\n", "")
-			fmt.Println(line)
+			fmt.Fprintf(output, "| %-61s |\n", "")
+			fmt.Fprintf(output, "+ %s +\n", center(name+" (64 bits)", 61))
+			fmt.Fprintf(output, "| %-61s |\n", "")
+			fmt.Fprintln(output, line)
 		case "string", "byte": // XXX We assume slice of byte!
-			fmt.Printf("| %s |\n", center("Length of "+name, 61))
-			fmt.Println(line)
-			fmt.Printf("/ %61s /\n", "")
-			fmt.Printf("\\ %s \\\n", center(name+" (variable length)", 61))
-			fmt.Printf("/ %61s /\n", "")
-			fmt.Println(line)
+			fmt.Fprintf(output, "| %s |\n", center("Length of "+name, 61))
+			fmt.Fprintln(output, line)
+			fmt.Fprintf(output, "/ %61s /\n", "")
+			fmt.Fprintf(output, "\\ %s \\\n", center(name+" (variable length)", 61))
+			fmt.Fprintf(output, "/ %61s /\n", "")
+			fmt.Fprintln(output, line)
 		default:
 			if sl {
 				tn = "Zero or more " + tn + " Structures"
-				fmt.Printf("/ %s /\n", center("", 61))
-				fmt.Printf("\\ %s \\\n", center(tn, 61))
-				fmt.Printf("/ %s /\n", center("", 61))
+				fmt.Fprintf(output, "/ %s /\n", center("", 61))
+				fmt.Fprintf(output, "\\ %s \\\n", center(tn, 61))
+				fmt.Fprintf(output, "/ %s /\n", center("", 61))
 			} else {
-				fmt.Printf("| %s |\n", center(tn, 61))
+				fmt.Fprintf(output, "| %s |\n", center(tn, 61))
 			}
-			fmt.Println(line)
+			fmt.Fprintln(output, line)
 		}
 	}
-	fmt.Println()
-	fmt.Println()
+	fmt.Fprintln(output)
+	fmt.Fprintln(output)
 }
 
-func generateXdr(s structInfo) {
+func generateXdr(output io.Writer, s structInfo) {
 	sn := s.Name
 	fs := s.Fields
 
-	fmt.Printf("struct %s {\n", sn)
+	fmt.Fprintf(output, "struct %s {\n", sn)
 
 	for _, f := range fs {
 		tn := f.FieldType
@@ -373,21 +375,21 @@ func generateXdr(s structInfo) {
 
 		switch tn {
 		case "uint16", "uint32":
-			fmt.Printf("\tunsigned int %s%s;\n", fn, suf)
+			fmt.Fprintf(output, "\tunsigned int %s%s;\n", fn, suf)
 		case "int64":
-			fmt.Printf("\thyper %s%s;\n", fn, suf)
+			fmt.Fprintf(output, "\thyper %s%s;\n", fn, suf)
 		case "uint64":
-			fmt.Printf("\tunsigned hyper %s%s;\n", fn, suf)
+			fmt.Fprintf(output, "\tunsigned hyper %s%s;\n", fn, suf)
 		case "string":
-			fmt.Printf("\tstring %s<%s>;\n", fn, l)
+			fmt.Fprintf(output, "\tstring %s<%s>;\n", fn, l)
 		case "byte":
-			fmt.Printf("\topaque %s<%s>;\n", fn, l)
+			fmt.Fprintf(output, "\topaque %s<%s>;\n", fn, l)
 		default:
-			fmt.Printf("\t%s %s%s;\n", tn, fn, suf)
+			fmt.Fprintf(output, "\t%s %s%s;\n", tn, fn, suf)
 		}
 	}
-	fmt.Println("}")
-	fmt.Println()
+	fmt.Fprintln(output, "}")
+	fmt.Fprintln(output)
 }
 
 func center(s string, w int) string {
@@ -418,25 +420,35 @@ func inspector(structs *[]structInfo) func(ast.Node) bool {
 }
 
 func main() {
+	outputFile := flag.String("o", "", "Output file, blank for stdout")
 	flag.Parse()
 	fname := flag.Arg(0)
 
 	fset := token.NewFileSet()
 	f, err := parser.ParseFile(fset, fname, nil, parser.ParseComments)
 	if err != nil {
-		panic(err)
+		log.Fatal(err)
 	}
 
 	var structs []structInfo
 	i := inspector(&structs)
 	ast.Inspect(f, i)
 
-	headerTpl.Execute(os.Stdout, map[string]string{"Package": f.Name.Name})
+	var output io.Writer = os.Stdout
+	if *outputFile != "" {
+		fd, err := os.Create(*outputFile)
+		if err != nil {
+			log.Fatal(err)
+		}
+		output = fd
+	}
+
+	headerTpl.Execute(output, map[string]string{"Package": f.Name.Name})
 	for _, s := range structs {
-		fmt.Printf("\n/*\n\n")
-		generateDiagram(s)
-		generateXdr(s)
-		fmt.Printf("*/\n")
-		generateCode(s)
+		fmt.Fprintf(output, "\n/*\n\n")
+		generateDiagram(output, s)
+		generateXdr(output, s)
+		fmt.Fprintf(output, "*/\n")
+		generateCode(output, s)
 	}
 }

Godeps/_workspace/src/github.com/calmh/xdr/reader.go

@@ -154,6 +154,10 @@ func (e XDRError) Error() string {
 	return "xdr " + e.op + ": " + e.err.Error()
 }
 
+func (e XDRError) IsEOF() bool {
+	return e.err == io.EOF
+}
+
 func (r *Reader) Error() error {
 	if r.err == nil {
 		return nil

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/batch.go

@@ -8,65 +8,84 @@ package leveldb
 
 import (
 	"encoding/binary"
-	"errors"
+	"fmt"
 
+	"github.com/syndtr/goleveldb/leveldb/errors"
 	"github.com/syndtr/goleveldb/leveldb/memdb"
 )
 
-var (
-	errBatchTooShort  = errors.New("leveldb: batch is too short")
-	errBatchBadRecord = errors.New("leveldb: bad record in batch")
+type ErrBatchCorrupted struct {
+	Reason string
+}
+
+func (e *ErrBatchCorrupted) Error() string {
+	return fmt.Sprintf("leveldb: batch corrupted: %s", e.Reason)
+}
+
+func newErrBatchCorrupted(reason string) error {
+	return errors.NewErrCorrupted(nil, &ErrBatchCorrupted{reason})
+}
+
+const (
+	batchHdrLen  = 8 + 4
+	batchGrowRec = 3000
 )
 
-const kBatchHdrLen = 8 + 4
-
-type batchReplay interface {
-	put(key, value []byte, seq uint64)
-	delete(key []byte, seq uint64)
+type BatchReplay interface {
+	Put(key, value []byte)
+	Delete(key []byte)
 }
 
 // Batch is a write batch.
 type Batch struct {
-	buf        []byte
+	data       []byte
 	rLen, bLen int
 	seq        uint64
 	sync       bool
 }
 
 func (b *Batch) grow(n int) {
-	off := len(b.buf)
+	off := len(b.data)
 	if off == 0 {
-		// include headers
-		off = kBatchHdrLen
-		n += off
+		off = batchHdrLen
+		if b.data != nil {
+			b.data = b.data[:off]
+		}
 	}
-	if cap(b.buf)-off >= n {
-		return
+	if cap(b.data)-off < n {
+		if b.data == nil {
+			b.data = make([]byte, off, off+n)
+		} else {
+			odata := b.data
+			div := 1
+			if b.rLen > batchGrowRec {
+				div = b.rLen / batchGrowRec
+			}
+			b.data = make([]byte, off, off+n+(off-batchHdrLen)/div)
+			copy(b.data, odata)
+		}
 	}
-	buf := make([]byte, 2*cap(b.buf)+n)
-	copy(buf, b.buf)
-	b.buf = buf[:off]
 }
 
-func (b *Batch) appendRec(t vType, key, value []byte) {
+func (b *Batch) appendRec(kt kType, key, value []byte) {
 	n := 1 + binary.MaxVarintLen32 + len(key)
-	if t == tVal {
+	if kt == ktVal {
 		n += binary.MaxVarintLen32 + len(value)
 	}
 	b.grow(n)
-	off := len(b.buf)
-	buf := b.buf[:off+n]
-	buf[off] = byte(t)
+	off := len(b.data)
+	data := b.data[:off+n]
+	data[off] = byte(kt)
 	off += 1
-	off += binary.PutUvarint(buf[off:], uint64(len(key)))
-	copy(buf[off:], key)
+	off += binary.PutUvarint(data[off:], uint64(len(key)))
+	copy(data[off:], key)
 	off += len(key)
-	if t == tVal {
-		off += binary.PutUvarint(buf[off:], uint64(len(value)))
-		copy(buf[off:], value)
+	if kt == ktVal {
+		off += binary.PutUvarint(data[off:], uint64(len(value)))
+		copy(data[off:], value)
 		off += len(value)
 	}
-	b.buf = buf[:off]
+	b.data = data[:off]
 	b.rLen++
 	//  Include 8-byte ikey header
 	b.bLen += len(key) + len(value) + 8
@@ -75,18 +94,51 @@ func (b *Batch) appendRec(t vType, key, value []byte) {
 // Put appends 'put operation' of the given key/value pair to the batch.
 // It is safe to modify the contents of the argument after Put returns.
 func (b *Batch) Put(key, value []byte) {
-	b.appendRec(tVal, key, value)
+	b.appendRec(ktVal, key, value)
 }
 
 // Delete appends 'delete operation' of the given key to the batch.
 // It is safe to modify the contents of the argument after Delete returns.
 func (b *Batch) Delete(key []byte) {
-	b.appendRec(tDel, key, nil)
+	b.appendRec(ktDel, key, nil)
+}
+
+// Dump dumps batch contents. The returned slice can be loaded into the
+// batch using Load method.
+// The returned slice is not its own copy, so the contents should not be
+// modified.
+func (b *Batch) Dump() []byte {
+	return b.encode()
+}
+
+// Load loads given slice into the batch. Previous contents of the batch
+// will be discarded.
+// The given slice will not be copied and will be used as batch buffer, so
+// it is not safe to modify the contents of the slice.
+func (b *Batch) Load(data []byte) error {
+	return b.decode(0, data)
+}
+
+// Replay replays batch contents.
+func (b *Batch) Replay(r BatchReplay) error {
+	return b.decodeRec(func(i int, kt kType, key, value []byte) {
+		switch kt {
+		case ktVal:
+			r.Put(key, value)
+		case ktDel:
+			r.Delete(key)
+		}
+	})
+}
+
+// Len returns number of records in the batch.
+func (b *Batch) Len() int {
+	return b.rLen
 }
 
 // Reset resets the batch.
 func (b *Batch) Reset() {
-	b.buf = nil
+	b.data = b.data[:0]
 	b.seq = 0
 	b.rLen = 0
 	b.bLen = 0
@@ -97,24 +149,10 @@ func (b *Batch) init(sync bool) {
 	b.sync = sync
 }
 
-func (b *Batch) put(key, value []byte, seq uint64) {
-	if b.rLen == 0 {
-		b.seq = seq
-	}
-	b.Put(key, value)
-}
-
-func (b *Batch) delete(key []byte, seq uint64) {
-	if b.rLen == 0 {
-		b.seq = seq
-	}
-	b.Delete(key)
-}
-
 func (b *Batch) append(p *Batch) {
 	if p.rLen > 0 {
-		b.grow(len(p.buf) - kBatchHdrLen)
-		b.buf = append(b.buf, p.buf[kBatchHdrLen:]...)
+		b.grow(len(p.data) - batchHdrLen)
+		b.data = append(b.data, p.data[batchHdrLen:]...)
 		b.rLen += p.rLen
 	}
 	if p.sync {
@@ -122,95 +160,93 @@ func (b *Batch) append(p *Batch) {
 	}
 }
 
-func (b *Batch) len() int {
-	return b.rLen
-}
-
+// size returns sums of key/value pair length plus 8-bytes ikey.
 func (b *Batch) size() int {
 	return b.bLen
 }
 
 func (b *Batch) encode() []byte {
 	b.grow(0)
-	binary.LittleEndian.PutUint64(b.buf, b.seq)
-	binary.LittleEndian.PutUint32(b.buf[8:], uint32(b.rLen))
+	binary.LittleEndian.PutUint64(b.data, b.seq)
+	binary.LittleEndian.PutUint32(b.data[8:], uint32(b.rLen))
 
-	return b.buf
+	return b.data
 }
 
-func (b *Batch) decode(buf []byte) error {
-	if len(buf) < kBatchHdrLen {
-		return errBatchTooShort
+func (b *Batch) decode(prevSeq uint64, data []byte) error {
+	if len(data) < batchHdrLen {
+		return newErrBatchCorrupted("too short")
 	}
 
-	b.seq = binary.LittleEndian.Uint64(buf)
-	b.rLen = int(binary.LittleEndian.Uint32(buf[8:]))
+	b.seq = binary.LittleEndian.Uint64(data)
+	if b.seq < prevSeq {
+		return newErrBatchCorrupted("invalid sequence number")
+	}
+	b.rLen = int(binary.LittleEndian.Uint32(data[8:]))
+	if b.rLen < 0 {
+		return newErrBatchCorrupted("invalid records length")
+	}
 	// No need to be precise at this point, it won't be used anyway
-	b.bLen = len(buf) - kBatchHdrLen
-	b.buf = buf
+	b.bLen = len(data) - batchHdrLen
+	b.data = data
 
 	return nil
 }
 
-func (b *Batch) decodeRec(f func(i int, t vType, key, value []byte)) error {
-	off := kBatchHdrLen
+func (b *Batch) decodeRec(f func(i int, kt kType, key, value []byte)) (err error) {
+	off := batchHdrLen
 	for i := 0; i < b.rLen; i++ {
-		if off >= len(b.buf) {
-			return errors.New("leveldb: invalid batch record length")
+		if off >= len(b.data) {
+			return newErrBatchCorrupted("invalid records length")
 		}
 
-		t := vType(b.buf[off])
-		if t > tVal {
-			return errors.New("leveldb: invalid batch record type in batch")
+		kt := kType(b.data[off])
+		if kt > ktVal {
+			return newErrBatchCorrupted("bad record: invalid type")
 		}
 		off += 1
 
-		x, n := binary.Uvarint(b.buf[off:])
+		x, n := binary.Uvarint(b.data[off:])
 		off += n
-		if n <= 0 || off+int(x) > len(b.buf) {
-			return errBatchBadRecord
+		if n <= 0 || off+int(x) > len(b.data) {
+			return newErrBatchCorrupted("bad record: invalid key length")
 		}
-		key := b.buf[off : off+int(x)]
+		key := b.data[off : off+int(x)]
 		off += int(x)
-
 		var value []byte
-		if t == tVal {
-			x, n := binary.Uvarint(b.buf[off:])
+		if kt == ktVal {
+			x, n := binary.Uvarint(b.data[off:])
 			off += n
-			if n <= 0 || off+int(x) > len(b.buf) {
-				return errBatchBadRecord
+			if n <= 0 || off+int(x) > len(b.data) {
+				return newErrBatchCorrupted("bad record: invalid value length")
 			}
-			value = b.buf[off : off+int(x)]
+			value = b.data[off : off+int(x)]
 			off += int(x)
 		}
 
-		f(i, t, key, value)
+		f(i, kt, key, value)
 	}
 
 	return nil
 }
 
-func (b *Batch) replay(to batchReplay) error {
-	return b.decodeRec(func(i int, t vType, key, value []byte) {
-		switch t {
-		case tVal:
-			to.put(key, value, b.seq+uint64(i))
-		case tDel:
-			to.delete(key, b.seq+uint64(i))
-		}
-	})
-}
-
 func (b *Batch) memReplay(to *memdb.DB) error {
-	return b.decodeRec(func(i int, t vType, key, value []byte) {
-		ikey := newIKey(key, b.seq+uint64(i), t)
+	return b.decodeRec(func(i int, kt kType, key, value []byte) {
+		ikey := newIkey(key, b.seq+uint64(i), kt)
 		to.Put(ikey, value)
 	})
 }
 
+func (b *Batch) memDecodeAndReplay(prevSeq uint64, data []byte, to *memdb.DB) error {
+	if err := b.decode(prevSeq, data); err != nil {
+		return err
+	}
+	return b.memReplay(to)
+}
+
 func (b *Batch) revertMemReplay(to *memdb.DB) error {
-	return b.decodeRec(func(i int, t vType, key, value []byte) {
-		ikey := newIKey(key, b.seq+uint64(i), t)
+	return b.decodeRec(func(i int, kt kType, key, value []byte) {
+		ikey := newIkey(key, b.seq+uint64(i), kt)
 		to.Delete(ikey)
 	})
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/batch_test.go

@@ -15,7 +15,7 @@ import (
 )
 
 type tbRec struct {
-	t          vType
+	kt         kType
 	key, value []byte
 }
 
@@ -23,39 +23,39 @@ type testBatch struct {
 	rec []*tbRec
 }
 
-func (p *testBatch) put(key, value []byte, seq uint64) {
-	p.rec = append(p.rec, &tbRec{tVal, key, value})
+func (p *testBatch) Put(key, value []byte) {
+	p.rec = append(p.rec, &tbRec{ktVal, key, value})
 }
 
-func (p *testBatch) delete(key []byte, seq uint64) {
-	p.rec = append(p.rec, &tbRec{tDel, key, nil})
+func (p *testBatch) Delete(key []byte) {
+	p.rec = append(p.rec, &tbRec{ktDel, key, nil})
 }
 
 func compareBatch(t *testing.T, b1, b2 *Batch) {
 	if b1.seq != b2.seq {
 		t.Errorf("invalid seq number want %d, got %d", b1.seq, b2.seq)
 	}
-	if b1.len() != b2.len() {
-		t.Fatalf("invalid record length want %d, got %d", b1.len(), b2.len())
+	if b1.Len() != b2.Len() {
+		t.Fatalf("invalid record length want %d, got %d", b1.Len(), b2.Len())
 	}
 	p1, p2 := new(testBatch), new(testBatch)
-	err := b1.replay(p1)
+	err := b1.Replay(p1)
 	if err != nil {
 		t.Fatal("error when replaying batch 1: ", err)
 	}
-	err = b2.replay(p2)
+	err = b2.Replay(p2)
 	if err != nil {
 		t.Fatal("error when replaying batch 2: ", err)
 	}
 	for i := range p1.rec {
 		r1, r2 := p1.rec[i], p2.rec[i]
-		if r1.t != r2.t {
-			t.Errorf("invalid type on record '%d' want %d, got %d", i, r1.t, r2.t)
+		if r1.kt != r2.kt {
+			t.Errorf("invalid type on record '%d' want %d, got %d", i, r1.kt, r2.kt)
 		}
 		if !bytes.Equal(r1.key, r2.key) {
 			t.Errorf("invalid key on record '%d' want %s, got %s", i, string(r1.key), string(r2.key))
 		}
-		if r1.t == tVal {
+		if r1.kt == ktVal {
 			if !bytes.Equal(r1.value, r2.value) {
 				t.Errorf("invalid value on record '%d' want %s, got %s", i, string(r1.value), string(r2.value))
 			}
@@ -75,7 +75,7 @@ func TestBatch_EncodeDecode(t *testing.T) {
 	b1.Delete([]byte("k"))
 	buf := b1.encode()
 	b2 := new(Batch)
-	err := b2.decode(buf)
+	err := b2.decode(0, buf)
 	if err != nil {
 		t.Error("error when decoding batch: ", err)
 	}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/cache/cache.go

@@ -8,152 +8,669 @@
 package cache
 
 import (
+	"sync"
 	"sync/atomic"
+	"unsafe"
+
+	"github.com/syndtr/goleveldb/leveldb/util"
 )
 
-// SetFunc is the function that will be called by Namespace.Get to create
-// a cache object, if charge is less than one than the cache object will
-// not be registered to cache tree, if value is nil then the cache object
-// will not be created.
-type SetFunc func() (charge int, value interface{})
-
-// DelFin is the function that will be called as the result of a delete operation.
-// Exist == true is indication that the object is exist, and pending == true is
-// indication of deletion already happen but haven't done yet (wait for all handles
-// to be released). And exist == false means the object doesn't exist.
-type DelFin func(exist, pending bool)
-
-// PurgeFin is the function that will be called as the result of a purge operation.
-type PurgeFin func(ns, key uint64)
-
-// Cache is a cache tree. A cache instance must be goroutine-safe.
-type Cache interface {
-	// SetCapacity sets cache tree capacity.
-	SetCapacity(capacity int)
-
-	// Capacity returns cache tree capacity.
+// Cacher provides interface to implements a caching functionality.
+// An implementation must be goroutine-safe.
+type Cacher interface {
+	// Capacity returns cache capacity.
 	Capacity() int
 
-	// Used returns used cache tree capacity.
-	Used() int
+	// SetCapacity sets cache capacity.
+	SetCapacity(capacity int)
 
-	// Size returns entire alive cache objects size.
-	Size() int
+	// Promote promotes the 'cache node'.
+	Promote(n *Node)
 
-	// NumObjects returns number of alive objects.
-	NumObjects() int
+	// Ban evicts the 'cache node' and prevent subsequent 'promote'.
+	Ban(n *Node)
 
-	// GetNamespace gets cache namespace with the given id.
-	// GetNamespace is never return nil.
-	GetNamespace(id uint64) Namespace
+	// Evict evicts the 'cache node'.
+	Evict(n *Node)
 
-	// PurgeNamespace purges cache namespace with the given id from this cache tree.
-	// Also read Namespace.Purge.
-	PurgeNamespace(id uint64, fin PurgeFin)
+	// EvictNS evicts 'cache node' with the given namespace.
+	EvictNS(ns uint64)
 
-	// ZapNamespace detaches cache namespace with the given id from this cache tree.
-	// Also read Namespace.Zap.
-	ZapNamespace(id uint64)
+	// EvictAll evicts all 'cache node'.
+	EvictAll()
 
-	// Purge purges all cache namespace from this cache tree.
-	// This is behave the same as calling Namespace.Purge method on all cache namespace.
-	Purge(fin PurgeFin)
-
-	// Zap detaches all cache namespace from this cache tree.
-	// This is behave the same as calling Namespace.Zap method on all cache namespace.
-	Zap()
+	// Close closes the 'cache tree'
+	Close() error
 }
 
-// Namespace is a cache namespace. A namespace instance must be goroutine-safe.
-type Namespace interface {
-	// Get gets cache object with the given key.
-	// If cache object is not found and setf is not nil, Get will atomically creates
-	// the cache object by calling setf. Otherwise Get will returns nil.
-	//
-	// The returned cache handle should be released after use by calling Release
-	// method.
-	Get(key uint64, setf SetFunc) Handle
+// Value is a 'cacheable object'. It may implements util.Releaser, if
+// so the the Release method will be called once object is released.
+type Value interface{}
 
-	// Delete removes cache object with the given key from cache tree.
-	// A deleted cache object will be released as soon as all of its handles have
-	// been released.
-	// Delete only happen once, subsequent delete will consider cache object doesn't
-	// exist, even if the cache object ins't released yet.
-	//
-	// If not nil, fin will be called if the cache object doesn't exist or when
-	// finally be released.
-	//
-	// Delete returns true if such cache object exist and never been deleted.
-	Delete(key uint64, fin DelFin) bool
-
-	// Purge removes all cache objects within this namespace from cache tree.
-	// This is the same as doing delete on all cache objects.
-	//
-	// If not nil, fin will be called on all cache objects when its finally be
-	// released.
-	Purge(fin PurgeFin)
-
-	// Zap detaches namespace from cache tree and release all its cache objects.
-	// A zapped namespace can never be filled again.
-	// Calling Get on zapped namespace will always return nil.
-	Zap()
+type CacheGetter struct {
+	Cache *Cache
+	NS    uint64
 }
 
-// Handle is a cache handle.
-type Handle interface {
-	// Release releases this cache handle. This method can be safely called mutiple
-	// times.
-	Release()
-
-	// Value returns value of this cache handle.
-	// Value will returns nil after this cache handle have be released.
-	Value() interface{}
+func (g *CacheGetter) Get(key uint64, setFunc func() (size int, value Value)) *Handle {
+	return g.Cache.Get(g.NS, key, setFunc)
 }
 
+// The hash tables implementation is based on:
+// "Dynamic-Sized Nonblocking Hash Tables", by Yujie Liu, Kunlong Zhang, and Michael Spear. ACM Symposium on Principles of Distributed Computing, Jul 2014.
+
 const (
-	DelNotExist = iota
-	DelExist
-	DelPendig
+	mInitialSize           = 1 << 4
+	mOverflowThreshold     = 1 << 5
+	mOverflowGrowThreshold = 1 << 7
 )
 
-// Namespace state.
-type nsState int
-
-const (
-	nsEffective nsState = iota
-	nsZapped
-)
-
-// Node state.
-type nodeState int
-
-const (
-	nodeZero nodeState = iota
-	nodeEffective
-	nodeEvicted
-	nodeDeleted
-)
-
-// Fake handle.
-type fakeHandle struct {
-	value interface{}
-	fin   func()
-	once  uint32
+type mBucket struct {
+	mu     sync.Mutex
+	node   []*Node
+	frozen bool
 }
 
-func (h *fakeHandle) Value() interface{} {
-	if atomic.LoadUint32(&h.once) == 0 {
-		return h.value
+func (b *mBucket) freeze() []*Node {
+	b.mu.Lock()
+	defer b.mu.Unlock()
+	if !b.frozen {
+		b.frozen = true
+	}
+	return b.node
+}
+
+func (b *mBucket) get(r *Cache, h *mNode, hash uint32, ns, key uint64, noset bool) (done, added bool, n *Node) {
+	b.mu.Lock()
+
+	if b.frozen {
+		b.mu.Unlock()
+		return
+	}
+
+	// Scan the node.
+	for _, n := range b.node {
+		if n.hash == hash && n.ns == ns && n.key == key {
+			atomic.AddInt32(&n.ref, 1)
+			b.mu.Unlock()
+			return true, false, n
+		}
+	}
+
+	// Get only.
+	if noset {
+		b.mu.Unlock()
+		return true, false, nil
+	}
+
+	// Create node.
+	n = &Node{
+		r:    r,
+		hash: hash,
+		ns:   ns,
+		key:  key,
+		ref:  1,
+	}
+	// Add node to bucket.
+	b.node = append(b.node, n)
+	bLen := len(b.node)
+	b.mu.Unlock()
+
+	// Update counter.
+	grow := atomic.AddInt32(&r.nodes, 1) >= h.growThreshold
+	if bLen > mOverflowThreshold {
+		grow = grow || atomic.AddInt32(&h.overflow, 1) >= mOverflowGrowThreshold
+	}
+
+	// Grow.
+	if grow && atomic.CompareAndSwapInt32(&h.resizeInProgess, 0, 1) {
+		nhLen := len(h.buckets) << 1
+		nh := &mNode{
+			buckets:         make([]unsafe.Pointer, nhLen),
+			mask:            uint32(nhLen) - 1,
+			pred:            unsafe.Pointer(h),
+			growThreshold:   int32(nhLen * mOverflowThreshold),
+			shrinkThreshold: int32(nhLen >> 1),
+		}
+		ok := atomic.CompareAndSwapPointer(&r.mHead, unsafe.Pointer(h), unsafe.Pointer(nh))
+		if !ok {
+			panic("BUG: failed swapping head")
+		}
+		go nh.initBuckets()
+	}
+
+	return true, true, n
+}
+
+func (b *mBucket) delete(r *Cache, h *mNode, hash uint32, ns, key uint64) (done, deleted bool) {
+	b.mu.Lock()
+
+	if b.frozen {
+		b.mu.Unlock()
+		return
+	}
+
+	// Scan the node.
+	var (
+		n    *Node
+		bLen int
+	)
+	for i := range b.node {
+		n = b.node[i]
+		if n.ns == ns && n.key == key {
+			if atomic.LoadInt32(&n.ref) == 0 {
+				deleted = true
+
+				// Call releaser.
+				if n.value != nil {
+					if r, ok := n.value.(util.Releaser); ok {
+						r.Release()
+					}
+					n.value = nil
+				}
+
+				// Remove node from bucket.
+				b.node = append(b.node[:i], b.node[i+1:]...)
+				bLen = len(b.node)
+			}
+			break
+		}
+	}
+	b.mu.Unlock()
+
+	if deleted {
+		// Call OnDel.
+		for _, f := range n.onDel {
+			f()
+		}
+
+		// Update counter.
+		atomic.AddInt32(&r.size, int32(n.size)*-1)
+		shrink := atomic.AddInt32(&r.nodes, -1) < h.shrinkThreshold
+		if bLen >= mOverflowThreshold {
+			atomic.AddInt32(&h.overflow, -1)
+		}
+
+		// Shrink.
+		if shrink && len(h.buckets) > mInitialSize && atomic.CompareAndSwapInt32(&h.resizeInProgess, 0, 1) {
+			nhLen := len(h.buckets) >> 1
+			nh := &mNode{
+				buckets:         make([]unsafe.Pointer, nhLen),
+				mask:            uint32(nhLen) - 1,
+				pred:            unsafe.Pointer(h),
+				growThreshold:   int32(nhLen * mOverflowThreshold),
+				shrinkThreshold: int32(nhLen >> 1),
+			}
+			ok := atomic.CompareAndSwapPointer(&r.mHead, unsafe.Pointer(h), unsafe.Pointer(nh))
+			if !ok {
+				panic("BUG: failed swapping head")
+			}
+			go nh.initBuckets()
+		}
+	}
+
+	return true, deleted
+}
+
+type mNode struct {
+	buckets         []unsafe.Pointer // []*mBucket
+	mask            uint32
+	pred            unsafe.Pointer // *mNode
+	resizeInProgess int32
+
+	overflow        int32
+	growThreshold   int32
+	shrinkThreshold int32
+}
+
+func (n *mNode) initBucket(i uint32) *mBucket {
+	if b := (*mBucket)(atomic.LoadPointer(&n.buckets[i])); b != nil {
+		return b
+	}
+
+	p := (*mNode)(atomic.LoadPointer(&n.pred))
+	if p != nil {
+		var node []*Node
+		if n.mask > p.mask {
+			// Grow.
+			pb := (*mBucket)(atomic.LoadPointer(&p.buckets[i&p.mask]))
+			if pb == nil {
+				pb = p.initBucket(i & p.mask)
+			}
+			m := pb.freeze()
+			// Split nodes.
+			for _, x := range m {
+				if x.hash&n.mask == i {
+					node = append(node, x)
+				}
+			}
+		} else {
+			// Shrink.
+			pb0 := (*mBucket)(atomic.LoadPointer(&p.buckets[i]))
+			if pb0 == nil {
+				pb0 = p.initBucket(i)
+			}
+			pb1 := (*mBucket)(atomic.LoadPointer(&p.buckets[i+uint32(len(n.buckets))]))
+			if pb1 == nil {
+				pb1 = p.initBucket(i + uint32(len(n.buckets)))
+			}
+			m0 := pb0.freeze()
+			m1 := pb1.freeze()
+			// Merge nodes.
+			node = make([]*Node, 0, len(m0)+len(m1))
+			node = append(node, m0...)
+			node = append(node, m1...)
+		}
+		b := &mBucket{node: node}
+		if atomic.CompareAndSwapPointer(&n.buckets[i], nil, unsafe.Pointer(b)) {
+			if len(node) > mOverflowThreshold {
+				atomic.AddInt32(&n.overflow, int32(len(node)-mOverflowThreshold))
+			}
+			return b
+		}
+	}
+
+	return (*mBucket)(atomic.LoadPointer(&n.buckets[i]))
+}
+
+func (n *mNode) initBuckets() {
+	for i := range n.buckets {
+		n.initBucket(uint32(i))
+	}
+	atomic.StorePointer(&n.pred, nil)
+}
+
+// Cache is a 'cache map'.
+type Cache struct {
+	mu     sync.RWMutex
+	mHead  unsafe.Pointer // *mNode
+	nodes  int32
+	size   int32
+	cacher Cacher
+	closed bool
+}
+
+// NewCache creates a new 'cache map'. The cacher is optional and
+// may be nil.
+func NewCache(cacher Cacher) *Cache {
+	h := &mNode{
+		buckets:         make([]unsafe.Pointer, mInitialSize),
+		mask:            mInitialSize - 1,
+		growThreshold:   int32(mInitialSize * mOverflowThreshold),
+		shrinkThreshold: 0,
+	}
+	for i := range h.buckets {
+		h.buckets[i] = unsafe.Pointer(&mBucket{})
+	}
+	r := &Cache{
+		mHead:  unsafe.Pointer(h),
+		cacher: cacher,
+	}
+	return r
+}
+
+func (r *Cache) getBucket(hash uint32) (*mNode, *mBucket) {
+	h := (*mNode)(atomic.LoadPointer(&r.mHead))
+	i := hash & h.mask
+	b := (*mBucket)(atomic.LoadPointer(&h.buckets[i]))
+	if b == nil {
+		b = h.initBucket(i)
+	}
+	return h, b
+}
+
+func (r *Cache) delete(n *Node) bool {
+	for {
+		h, b := r.getBucket(n.hash)
+		done, deleted := b.delete(r, h, n.hash, n.ns, n.key)
+		if done {
+			return deleted
+		}
+	}
+	return false
+}
+
+// Nodes returns number of 'cache node' in the map.
+func (r *Cache) Nodes() int {
+	return int(atomic.LoadInt32(&r.nodes))
+}
+
+// Size returns sums of 'cache node' size in the map.
+func (r *Cache) Size() int {
+	return int(atomic.LoadInt32(&r.size))
+}
+
+// Capacity returns cache capacity.
+func (r *Cache) Capacity() int {
+	if r.cacher == nil {
+		return 0
+	}
+	return r.cacher.Capacity()
+}
+
+// SetCapacity sets cache capacity.
+func (r *Cache) SetCapacity(capacity int) {
+	if r.cacher != nil {
+		r.cacher.SetCapacity(capacity)
+	}
+}
+
+// Get gets 'cache node' with the given namespace and key.
+// If cache node is not found and setFunc is not nil, Get will atomically creates
+// the 'cache node' by calling setFunc. Otherwise Get will returns nil.
+//
+// The returned 'cache handle' should be released after use by calling Release
+// method.
+func (r *Cache) Get(ns, key uint64, setFunc func() (size int, value Value)) *Handle {
+	r.mu.RLock()
+	defer r.mu.RUnlock()
+	if r.closed {
+		return nil
+	}
+
+	hash := murmur32(ns, key, 0xf00)
+	for {
+		h, b := r.getBucket(hash)
+		done, _, n := b.get(r, h, hash, ns, key, setFunc == nil)
+		if done {
+			if n != nil {
+				n.mu.Lock()
+				if n.value == nil {
+					if setFunc == nil {
+						n.mu.Unlock()
+						n.unref()
+						return nil
+					}
+
+					n.size, n.value = setFunc()
+					if n.value == nil {
+						n.size = 0
+						n.mu.Unlock()
+						n.unref()
+						return nil
+					}
+					atomic.AddInt32(&r.size, int32(n.size))
+				}
+				n.mu.Unlock()
+				if r.cacher != nil {
+					r.cacher.Promote(n)
+				}
+				return &Handle{unsafe.Pointer(n)}
+			}
+
+			break
+		}
 	}
 	return nil
 }
 
-func (h *fakeHandle) Release() {
-	if !atomic.CompareAndSwapUint32(&h.once, 0, 1) {
+// Delete removes and ban 'cache node' with the given namespace and key.
+// A banned 'cache node' will never inserted into the 'cache tree'. Ban
+// only attributed to the particular 'cache node', so when a 'cache node'
+// is recreated it will not be banned.
+//
+// If onDel is not nil, then it will be executed if such 'cache node'
+// doesn't exist or once the 'cache node' is released.
+//
+// Delete return true is such 'cache node' exist.
+func (r *Cache) Delete(ns, key uint64, onDel func()) bool {
+	r.mu.RLock()
+	defer r.mu.RUnlock()
+	if r.closed {
+		return false
+	}
+
+	hash := murmur32(ns, key, 0xf00)
+	for {
+		h, b := r.getBucket(hash)
+		done, _, n := b.get(r, h, hash, ns, key, true)
+		if done {
+			if n != nil {
+				if onDel != nil {
+					n.mu.Lock()
+					n.onDel = append(n.onDel, onDel)
+					n.mu.Unlock()
+				}
+				if r.cacher != nil {
+					r.cacher.Ban(n)
+				}
+				n.unref()
+				return true
+			}
+
+			break
+		}
+	}
+
+	if onDel != nil {
+		onDel()
+	}
+
+	return false
+}
+
+// Evict evicts 'cache node' with the given namespace and key. This will
+// simply call Cacher.Evict.
+//
+// Evict return true is such 'cache node' exist.
+func (r *Cache) Evict(ns, key uint64) bool {
+	r.mu.RLock()
+	defer r.mu.RUnlock()
+	if r.closed {
+		return false
+	}
+
+	hash := murmur32(ns, key, 0xf00)
+	for {
+		h, b := r.getBucket(hash)
+		done, _, n := b.get(r, h, hash, ns, key, true)
+		if done {
+			if n != nil {
+				if r.cacher != nil {
+					r.cacher.Evict(n)
+				}
+				n.unref()
+				return true
+			}
+
+			break
+		}
+	}
+
+	return false
+}
+
+// EvictNS evicts 'cache node' with the given namespace. This will
+// simply call Cacher.EvictNS.
+func (r *Cache) EvictNS(ns uint64) {
+	r.mu.RLock()
+	defer r.mu.RUnlock()
+	if r.closed {
 		return
 	}
-	if h.fin != nil {
-		h.fin()
-		h.fin = nil
+
+	if r.cacher != nil {
+		r.cacher.EvictNS(ns)
 	}
 }
+
+// EvictAll evicts all 'cache node'. This will simply call Cacher.EvictAll.
+func (r *Cache) EvictAll() {
+	r.mu.RLock()
+	defer r.mu.RUnlock()
+	if r.closed {
+		return
+	}
+
+	if r.cacher != nil {
+		r.cacher.EvictAll()
+	}
+}
+
+// Close closes the 'cache map' and releases all 'cache node'.
+func (r *Cache) Close() error {
+	r.mu.Lock()
+	if !r.closed {
+		r.closed = true
+
+		if r.cacher != nil {
+			if err := r.cacher.Close(); err != nil {
+				return err
+			}
+		}
+
+		h := (*mNode)(r.mHead)
+		h.initBuckets()
+
+		for i := range h.buckets {
+			b := (*mBucket)(h.buckets[i])
+			for _, n := range b.node {
+				// Call releaser.
+				if n.value != nil {
+					if r, ok := n.value.(util.Releaser); ok {
+						r.Release()
+					}
+					n.value = nil
+				}
+
+				// Call OnDel.
+				for _, f := range n.onDel {
+					f()
+				}
+			}
+		}
+	}
+	r.mu.Unlock()
+	return nil
+}
+
+// Node is a 'cache node'.
+type Node struct {
+	r *Cache
+
+	hash    uint32
+	ns, key uint64
+
+	mu    sync.Mutex
+	size  int
+	value Value
+
+	ref   int32
+	onDel []func()
+
+	CacheData unsafe.Pointer
+}
+
+// NS returns this 'cache node' namespace.
+func (n *Node) NS() uint64 {
+	return n.ns
+}
+
+// Key returns this 'cache node' key.
+func (n *Node) Key() uint64 {
+	return n.key
+}
+
+// Size returns this 'cache node' size.
+func (n *Node) Size() int {
+	return n.size
+}
+
+// Value returns this 'cache node' value.
+func (n *Node) Value() Value {
+	return n.value
+}
+
+// Ref returns this 'cache node' ref counter.
+func (n *Node) Ref() int32 {
+	return atomic.LoadInt32(&n.ref)
+}
+
+// GetHandle returns an handle for this 'cache node'.
+func (n *Node) GetHandle() *Handle {
+	if atomic.AddInt32(&n.ref, 1) <= 1 {
+		panic("BUG: Node.GetHandle on zero ref")
+	}
+	return &Handle{unsafe.Pointer(n)}
+}
+
+func (n *Node) unref() {
+	if atomic.AddInt32(&n.ref, -1) == 0 {
+		n.r.delete(n)
+	}
+}
+
+func (n *Node) unrefLocked() {
+	if atomic.AddInt32(&n.ref, -1) == 0 {
+		n.r.mu.RLock()
+		if !n.r.closed {
+			n.r.delete(n)
+		}
+		n.r.mu.RUnlock()
+	}
+}
+
+type Handle struct {
+	n unsafe.Pointer // *Node
+}
+
+func (h *Handle) Value() Value {
+	n := (*Node)(atomic.LoadPointer(&h.n))
+	if n != nil {
+		return n.value
+	}
+	return nil
+}
+
+func (h *Handle) Release() {
+	nPtr := atomic.LoadPointer(&h.n)
+	if nPtr != nil && atomic.CompareAndSwapPointer(&h.n, nPtr, nil) {
+		n := (*Node)(nPtr)
+		n.unrefLocked()
+	}
+}
+
+func murmur32(ns, key uint64, seed uint32) uint32 {
+	const (
+		m = uint32(0x5bd1e995)
+		r = 24
+	)
+
+	k1 := uint32(ns >> 32)
+	k2 := uint32(ns)
+	k3 := uint32(key >> 32)
+	k4 := uint32(key)
+
+	k1 *= m
+	k1 ^= k1 >> r
+	k1 *= m
+
+	k2 *= m
+	k2 ^= k2 >> r
+	k2 *= m
+
+	k3 *= m
+	k3 ^= k3 >> r
+	k3 *= m
+
+	k4 *= m
+	k4 ^= k4 >> r
+	k4 *= m
+
+	h := seed
+
+	h *= m
+	h ^= k1
+	h *= m
+	h ^= k2
+	h *= m
+	h ^= k3
+	h *= m
+	h ^= k4
+
+	h ^= h >> 13
+	h *= m
+	h ^= h >> 15
+
+	return h
+}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/cache/cache_test.go

@@ -13,11 +13,26 @@ import (
 	"sync/atomic"
 	"testing"
 	"time"
+	"unsafe"
 )
 
+type int32o int32
+
+func (o *int32o) acquire() {
+	if atomic.AddInt32((*int32)(o), 1) != 1 {
+		panic("BUG: invalid ref")
+	}
+}
+
+func (o *int32o) Release() {
+	if atomic.AddInt32((*int32)(o), -1) != 0 {
+		panic("BUG: invalid ref")
+	}
+}
+
 type releaserFunc struct {
 	fn    func()
-	value interface{}
+	value Value
 }
 
 func (r releaserFunc) Release() {
@@ -26,8 +41,8 @@ func (r releaserFunc) Release() {
 	}
 }
 
-func set(ns Namespace, key uint64, value interface{}, charge int, relf func()) Handle {
-	return ns.Get(key, func() (int, interface{}) {
+func set(c *Cache, ns, key uint64, value Value, charge int, relf func()) *Handle {
+	return c.Get(ns, key, func() (int, Value) {
 		if relf != nil {
 			return charge, releaserFunc{relf, value}
 		} else {
@@ -36,7 +51,246 @@ func set(ns Namespace, key uint64, value interface{}, charge int, relf func()) H
 	})
 }
 
-func TestCache_HitMiss(t *testing.T) {
+func TestCacheMap(t *testing.T) {
+	runtime.GOMAXPROCS(runtime.NumCPU())
+
+	nsx := []struct {
+		nobjects, nhandles, concurrent, repeat int
+	}{
+		{10000, 400, 50, 3},
+		{100000, 1000, 100, 10},
+	}
+
+	var (
+		objects [][]int32o
+		handles [][]unsafe.Pointer
+	)
+
+	for _, x := range nsx {
+		objects = append(objects, make([]int32o, x.nobjects))
+		handles = append(handles, make([]unsafe.Pointer, x.nhandles))
+	}
+
+	c := NewCache(nil)
+
+	wg := new(sync.WaitGroup)
+	var done int32
+
+	for ns, x := range nsx {
+		for i := 0; i < x.concurrent; i++ {
+			wg.Add(1)
+			go func(ns, i, repeat int, objects []int32o, handles []unsafe.Pointer) {
+				defer wg.Done()
+				r := rand.New(rand.NewSource(time.Now().UnixNano()))
+
+				for j := len(objects) * repeat; j >= 0; j-- {
+					key := uint64(r.Intn(len(objects)))
+					h := c.Get(uint64(ns), key, func() (int, Value) {
+						o := &objects[key]
+						o.acquire()
+						return 1, o
+					})
+					if v := h.Value().(*int32o); v != &objects[key] {
+						t.Fatalf("#%d invalid value: want=%p got=%p", ns, &objects[key], v)
+					}
+					if objects[key] != 1 {
+						t.Fatalf("#%d invalid object %d: %d", ns, key, objects[key])
+					}
+					if !atomic.CompareAndSwapPointer(&handles[r.Intn(len(handles))], nil, unsafe.Pointer(h)) {
+						h.Release()
+					}
+				}
+			}(ns, i, x.repeat, objects[ns], handles[ns])
+		}
+
+		go func(handles []unsafe.Pointer) {
+			r := rand.New(rand.NewSource(time.Now().UnixNano()))
+
+			for atomic.LoadInt32(&done) == 0 {
+				i := r.Intn(len(handles))
+				h := (*Handle)(atomic.LoadPointer(&handles[i]))
+				if h != nil && atomic.CompareAndSwapPointer(&handles[i], unsafe.Pointer(h), nil) {
+					h.Release()
+				}
+				time.Sleep(time.Millisecond)
+			}
+		}(handles[ns])
+	}
+
+	go func() {
+		handles := make([]*Handle, 100000)
+		for atomic.LoadInt32(&done) == 0 {
+			for i := range handles {
+				handles[i] = c.Get(999999999, uint64(i), func() (int, Value) {
+					return 1, 1
+				})
+			}
+			for _, h := range handles {
+				h.Release()
+			}
+		}
+	}()
+
+	wg.Wait()
+
+	atomic.StoreInt32(&done, 1)
+
+	for _, handles0 := range handles {
+		for i := range handles0 {
+			h := (*Handle)(atomic.LoadPointer(&handles0[i]))
+			if h != nil && atomic.CompareAndSwapPointer(&handles0[i], unsafe.Pointer(h), nil) {
+				h.Release()
+			}
+		}
+	}
+
+	for ns, objects0 := range objects {
+		for i, o := range objects0 {
+			if o != 0 {
+				t.Fatalf("invalid object #%d.%d: ref=%d", ns, i, o)
+			}
+		}
+	}
+}
+
+func TestCacheMap_NodesAndSize(t *testing.T) {
+	c := NewCache(nil)
+	if c.Nodes() != 0 {
+		t.Errorf("invalid nodes counter: want=%d got=%d", 0, c.Nodes())
+	}
+	if c.Size() != 0 {
+		t.Errorf("invalid size counter: want=%d got=%d", 0, c.Size())
+	}
+	set(c, 0, 1, 1, 1, nil)
+	set(c, 0, 2, 2, 2, nil)
+	set(c, 1, 1, 3, 3, nil)
+	set(c, 2, 1, 4, 1, nil)
+	if c.Nodes() != 4 {
+		t.Errorf("invalid nodes counter: want=%d got=%d", 4, c.Nodes())
+	}
+	if c.Size() != 7 {
+		t.Errorf("invalid size counter: want=%d got=%d", 4, c.Size())
+	}
+}
+
+func TestLRUCache_Capacity(t *testing.T) {
+	c := NewCache(NewLRU(10))
+	if c.Capacity() != 10 {
+		t.Errorf("invalid capacity: want=%d got=%d", 10, c.Capacity())
+	}
+	set(c, 0, 1, 1, 1, nil).Release()
+	set(c, 0, 2, 2, 2, nil).Release()
+	set(c, 1, 1, 3, 3, nil).Release()
+	set(c, 2, 1, 4, 1, nil).Release()
+	set(c, 2, 2, 5, 1, nil).Release()
+	set(c, 2, 3, 6, 1, nil).Release()
+	set(c, 2, 4, 7, 1, nil).Release()
+	set(c, 2, 5, 8, 1, nil).Release()
+	if c.Nodes() != 7 {
+		t.Errorf("invalid nodes counter: want=%d got=%d", 7, c.Nodes())
+	}
+	if c.Size() != 10 {
+		t.Errorf("invalid size counter: want=%d got=%d", 10, c.Size())
+	}
+	c.SetCapacity(9)
+	if c.Capacity() != 9 {
+		t.Errorf("invalid capacity: want=%d got=%d", 9, c.Capacity())
+	}
+	if c.Nodes() != 6 {
+		t.Errorf("invalid nodes counter: want=%d got=%d", 6, c.Nodes())
+	}
+	if c.Size() != 8 {
+		t.Errorf("invalid size counter: want=%d got=%d", 8, c.Size())
+	}
+}
+
+func TestCacheMap_NilValue(t *testing.T) {
+	c := NewCache(NewLRU(10))
+	h := c.Get(0, 0, func() (size int, value Value) {
+		return 1, nil
+	})
+	if h != nil {
+		t.Error("cache handle is non-nil")
+	}
+	if c.Nodes() != 0 {
+		t.Errorf("invalid nodes counter: want=%d got=%d", 0, c.Nodes())
+	}
+	if c.Size() != 0 {
+		t.Errorf("invalid size counter: want=%d got=%d", 0, c.Size())
+	}
+}
+
+func TestLRUCache_GetLatency(t *testing.T) {
+	runtime.GOMAXPROCS(runtime.NumCPU())
+
+	const (
+		concurrentSet = 30
+		concurrentGet = 3
+		duration      = 3 * time.Second
+		delay         = 3 * time.Millisecond
+		maxkey        = 100000
+	)
+
+	var (
+		set, getHit, getAll        int32
+		getMaxLatency, getDuration int64
+	)
+
+	c := NewCache(NewLRU(5000))
+	wg := &sync.WaitGroup{}
+	until := time.Now().Add(duration)
+	for i := 0; i < concurrentSet; i++ {
+		wg.Add(1)
+		go func(i int) {
+			defer wg.Done()
+			r := rand.New(rand.NewSource(time.Now().UnixNano()))
+			for time.Now().Before(until) {
+				c.Get(0, uint64(r.Intn(maxkey)), func() (int, Value) {
+					time.Sleep(delay)
+					atomic.AddInt32(&set, 1)
+					return 1, 1
+				}).Release()
+			}
+		}(i)
+	}
+	for i := 0; i < concurrentGet; i++ {
+		wg.Add(1)
+		go func(i int) {
+			defer wg.Done()
+			r := rand.New(rand.NewSource(time.Now().UnixNano()))
+			for {
+				mark := time.Now()
+				if mark.Before(until) {
+					h := c.Get(0, uint64(r.Intn(maxkey)), nil)
+					latency := int64(time.Now().Sub(mark))
+					m := atomic.LoadInt64(&getMaxLatency)
+					if latency > m {
+						atomic.CompareAndSwapInt64(&getMaxLatency, m, latency)
+					}
+					atomic.AddInt64(&getDuration, latency)
+					if h != nil {
+						atomic.AddInt32(&getHit, 1)
+						h.Release()
+					}
+					atomic.AddInt32(&getAll, 1)
+				} else {
+					break
+				}
+			}
+		}(i)
+	}
+
+	wg.Wait()
+	getAvglatency := time.Duration(getDuration) / time.Duration(getAll)
+	t.Logf("set=%d getHit=%d getAll=%d getMaxLatency=%v getAvgLatency=%v",
+		set, getHit, getAll, time.Duration(getMaxLatency), getAvglatency)
+
+	if getAvglatency > delay/3 {
+		t.Errorf("get avg latency > %v: got=%v", delay/3, getAvglatency)
+	}
+}
+
+func TestLRUCache_HitMiss(t *testing.T) {
 	cases := []struct {
 		key   uint64
 		value string
@@ -54,14 +308,13 @@ func TestCache_HitMiss(t *testing.T) {
 	}
 
 	setfin := 0
-	c := NewLRUCache(1000)
-	ns := c.GetNamespace(0)
+	c := NewCache(NewLRU(1000))
 	for i, x := range cases {
-		set(ns, x.key, x.value, len(x.value), func() {
+		set(c, 0, x.key, x.value, len(x.value), func() {
 			setfin++
 		}).Release()
 		for j, y := range cases {
-			h := ns.Get(y.key, nil)
+			h := c.Get(0, y.key, nil)
 			if j <= i {
 				// should hit
 				if h == nil {
@@ -85,7 +338,7 @@ func TestCache_HitMiss(t *testing.T) {
 
 	for i, x := range cases {
 		finalizerOk := false
-		ns.Delete(x.key, func(exist, pending bool) {
+		c.Delete(0, x.key, func() {
 			finalizerOk = true
 		})
 
@@ -94,7 +347,7 @@ func TestCache_HitMiss(t *testing.T) {
 		}
 
 		for j, y := range cases {
-			h := ns.Get(y.key, nil)
+			h := c.Get(0, y.key, nil)
 			if j > i {
 				// should hit
 				if h == nil {
@@ -122,20 +375,19 @@ func TestCache_HitMiss(t *testing.T) {
 }
 
 func TestLRUCache_Eviction(t *testing.T) {
-	c := NewLRUCache(12)
-	ns := c.GetNamespace(0)
-	o1 := set(ns, 1, 1, 1, nil)
-	set(ns, 2, 2, 1, nil).Release()
-	set(ns, 3, 3, 1, nil).Release()
-	set(ns, 4, 4, 1, nil).Release()
-	set(ns, 5, 5, 1, nil).Release()
-	if h := ns.Get(2, nil); h != nil { // 1,3,4,5,2
+	c := NewCache(NewLRU(12))
+	o1 := set(c, 0, 1, 1, 1, nil)
+	set(c, 0, 2, 2, 1, nil).Release()
+	set(c, 0, 3, 3, 1, nil).Release()
+	set(c, 0, 4, 4, 1, nil).Release()
+	set(c, 0, 5, 5, 1, nil).Release()
+	if h := c.Get(0, 2, nil); h != nil { // 1,3,4,5,2
 		h.Release()
 	}
-	set(ns, 9, 9, 10, nil).Release() // 5,2,9
+	set(c, 0, 9, 9, 10, nil).Release() // 5,2,9
 
 	for _, key := range []uint64{9, 2, 5, 1} {
-		h := ns.Get(key, nil)
+		h := c.Get(0, key, nil)
 		if h == nil {
 			t.Errorf("miss for key '%d'", key)
 		} else {
@@ -147,7 +399,7 @@ func TestLRUCache_Eviction(t *testing.T) {
 	}
 	o1.Release()
 	for _, key := range []uint64{1, 2, 5} {
-		h := ns.Get(key, nil)
+		h := c.Get(0, key, nil)
 		if h == nil {
 			t.Errorf("miss for key '%d'", key)
 		} else {
@@ -158,7 +410,7 @@ func TestLRUCache_Eviction(t *testing.T) {
 		}
 	}
 	for _, key := range []uint64{3, 4, 9} {
-		h := ns.Get(key, nil)
+		h := c.Get(0, key, nil)
 		if h != nil {
 			t.Errorf("hit for key '%d'", key)
 			if x := h.Value().(int); x != int(key) {
@@ -169,487 +421,150 @@ func TestLRUCache_Eviction(t *testing.T) {
 	}
 }
 
-func TestLRUCache_SetGet(t *testing.T) {
-	c := NewLRUCache(13)
-	ns := c.GetNamespace(0)
-	for i := 0; i < 200; i++ {
-		n := uint64(rand.Intn(99999) % 20)
-		set(ns, n, n, 1, nil).Release()
-		if h := ns.Get(n, nil); h != nil {
-			if h.Value() == nil {
-				t.Errorf("key '%d' contains nil value", n)
+func TestLRUCache_Evict(t *testing.T) {
+	c := NewCache(NewLRU(6))
+	set(c, 0, 1, 1, 1, nil).Release()
+	set(c, 0, 2, 2, 1, nil).Release()
+	set(c, 1, 1, 4, 1, nil).Release()
+	set(c, 1, 2, 5, 1, nil).Release()
+	set(c, 2, 1, 6, 1, nil).Release()
+	set(c, 2, 2, 7, 1, nil).Release()
+
+	for ns := 0; ns < 3; ns++ {
+		for key := 1; key < 3; key++ {
+			if h := c.Get(uint64(ns), uint64(key), nil); h != nil {
+				h.Release()
 			} else {
-				if x := h.Value().(uint64); x != n {
-					t.Errorf("invalid value for key '%d' want '%d', got '%d'", n, n, x)
-				}
+				t.Errorf("Cache.Get on #%d.%d return nil", ns, key)
 			}
+		}
+	}
+
+	if ok := c.Evict(0, 1); !ok {
+		t.Error("first Cache.Evict on #0.1 return false")
+	}
+	if ok := c.Evict(0, 1); ok {
+		t.Error("second Cache.Evict on #0.1 return true")
+	}
+	if h := c.Get(0, 1, nil); h != nil {
+		t.Errorf("Cache.Get on #0.1 return non-nil: %v", h.Value())
+	}
+
+	c.EvictNS(1)
+	if h := c.Get(1, 1, nil); h != nil {
+		t.Errorf("Cache.Get on #1.1 return non-nil: %v", h.Value())
+	}
+	if h := c.Get(1, 2, nil); h != nil {
+		t.Errorf("Cache.Get on #1.2 return non-nil: %v", h.Value())
+	}
+
+	c.EvictAll()
+	for ns := 0; ns < 3; ns++ {
+		for key := 1; key < 3; key++ {
+			if h := c.Get(uint64(ns), uint64(key), nil); h != nil {
+				t.Errorf("Cache.Get on #%d.%d return non-nil: %v", ns, key, h.Value())
+			}
+		}
+	}
+}
+
+func TestLRUCache_Delete(t *testing.T) {
+	delFuncCalled := 0
+	delFunc := func() {
+		delFuncCalled++
+	}
+
+	c := NewCache(NewLRU(2))
+	set(c, 0, 1, 1, 1, nil).Release()
+	set(c, 0, 2, 2, 1, nil).Release()
+
+	if ok := c.Delete(0, 1, delFunc); !ok {
+		t.Error("Cache.Delete on #1 return false")
+	}
+	if h := c.Get(0, 1, nil); h != nil {
+		t.Errorf("Cache.Get on #1 return non-nil: %v", h.Value())
+	}
+	if ok := c.Delete(0, 1, delFunc); ok {
+		t.Error("Cache.Delete on #1 return true")
+	}
+
+	h2 := c.Get(0, 2, nil)
+	if h2 == nil {
+		t.Error("Cache.Get on #2 return nil")
+	}
+	if ok := c.Delete(0, 2, delFunc); !ok {
+		t.Error("(1) Cache.Delete on #2 return false")
+	}
+	if ok := c.Delete(0, 2, delFunc); !ok {
+		t.Error("(2) Cache.Delete on #2 return false")
+	}
+
+	set(c, 0, 3, 3, 1, nil).Release()
+	set(c, 0, 4, 4, 1, nil).Release()
+	c.Get(0, 2, nil).Release()
+
+	for key := 2; key <= 4; key++ {
+		if h := c.Get(0, uint64(key), nil); h != nil {
 			h.Release()
 		} else {
-			t.Errorf("key '%d' doesn't exist", n)
-		}
-	}
-}
-
-func TestLRUCache_Purge(t *testing.T) {
-	c := NewLRUCache(3)
-	ns1 := c.GetNamespace(0)
-	o1 := set(ns1, 1, 1, 1, nil)
-	o2 := set(ns1, 2, 2, 1, nil)
-	ns1.Purge(nil)
-	set(ns1, 3, 3, 1, nil).Release()
-	for _, key := range []uint64{1, 2, 3} {
-		h := ns1.Get(key, nil)
-		if h == nil {
-			t.Errorf("miss for key '%d'", key)
-		} else {
-			if x := h.Value().(int); x != int(key) {
-				t.Errorf("invalid value for key '%d' want '%d', got '%d'", key, key, x)
-			}
-			h.Release()
-		}
-	}
-	o1.Release()
-	o2.Release()
-	for _, key := range []uint64{1, 2} {
-		h := ns1.Get(key, nil)
-		if h != nil {
-			t.Errorf("hit for key '%d'", key)
-			if x := h.Value().(int); x != int(key) {
-				t.Errorf("invalid value for key '%d' want '%d', got '%d'", key, key, x)
-			}
-			h.Release()
-		}
-	}
-}
-
-type testingCacheObjectCounter struct {
-	created  uint
-	released uint
-}
-
-func (c *testingCacheObjectCounter) createOne() {
-	c.created++
-}
-
-func (c *testingCacheObjectCounter) releaseOne() {
-	c.released++
-}
-
-type testingCacheObject struct {
-	t   *testing.T
-	cnt *testingCacheObjectCounter
-
-	ns, key uint64
-
-	releaseCalled bool
-}
-
-func (x *testingCacheObject) Release() {
-	if !x.releaseCalled {
-		x.releaseCalled = true
-		x.cnt.releaseOne()
-	} else {
-		x.t.Errorf("duplicate setfin NS#%d KEY#%s", x.ns, x.key)
-	}
-}
-
-func TestLRUCache_ConcurrentSetGet(t *testing.T) {
-	runtime.GOMAXPROCS(runtime.NumCPU())
-
-	seed := time.Now().UnixNano()
-	t.Logf("seed=%d", seed)
-
-	const (
-		N = 2000000
-		M = 4000
-		C = 3
-	)
-
-	var set, get uint32
-
-	wg := &sync.WaitGroup{}
-	c := NewLRUCache(M / 4)
-	for ni := uint64(0); ni < C; ni++ {
-		r0 := rand.New(rand.NewSource(seed + int64(ni)))
-		r1 := rand.New(rand.NewSource(seed + int64(ni) + 1))
-		ns := c.GetNamespace(ni)
-
-		wg.Add(2)
-		go func(ns Namespace, r *rand.Rand) {
-			for i := 0; i < N; i++ {
-				x := uint64(r.Int63n(M))
-				o := ns.Get(x, func() (int, interface{}) {
-					atomic.AddUint32(&set, 1)
-					return 1, x
-				})
-				if v := o.Value().(uint64); v != x {
-					t.Errorf("#%d invalid value, got=%d", x, v)
-				}
-				o.Release()
-			}
-			wg.Done()
-		}(ns, r0)
-		go func(ns Namespace, r *rand.Rand) {
-			for i := 0; i < N; i++ {
-				x := uint64(r.Int63n(M))
-				o := ns.Get(x, nil)
-				if o != nil {
-					atomic.AddUint32(&get, 1)
-					if v := o.Value().(uint64); v != x {
-						t.Errorf("#%d invalid value, got=%d", x, v)
-					}
-					o.Release()
-				}
-			}
-			wg.Done()
-		}(ns, r1)
-	}
-
-	wg.Wait()
-
-	t.Logf("set=%d get=%d", set, get)
-}
-
-func TestLRUCache_Finalizer(t *testing.T) {
-	const (
-		capacity   = 100
-		goroutines = 100
-		iterations = 10000
-		keymax     = 8000
-	)
-
-	cnt := &testingCacheObjectCounter{}
-
-	c := NewLRUCache(capacity)
-
-	type instance struct {
-		seed       int64
-		rnd        *rand.Rand
-		nsid       uint64
-		ns         Namespace
-		effective  int
-		handles    []Handle
-		handlesMap map[uint64]int
-
-		delete          bool
-		purge           bool
-		zap             bool
-		wantDel         int
-		delfinCalled    int
-		delfinCalledAll int
-		delfinCalledEff int
-		purgefinCalled  int
-	}
-
-	instanceGet := func(p *instance, key uint64) {
-		h := p.ns.Get(key, func() (charge int, value interface{}) {
-			to := &testingCacheObject{
-				t: t, cnt: cnt,
-				ns:  p.nsid,
-				key: key,
-			}
-			p.effective++
-			cnt.createOne()
-			return 1, releaserFunc{func() {
-				to.Release()
-				p.effective--
-			}, to}
-		})
-		p.handles = append(p.handles, h)
-		p.handlesMap[key] = p.handlesMap[key] + 1
-	}
-	instanceRelease := func(p *instance, i int) {
-		h := p.handles[i]
-		key := h.Value().(releaserFunc).value.(*testingCacheObject).key
-		if n := p.handlesMap[key]; n == 0 {
-			t.Fatal("key ref == 0")
-		} else if n > 1 {
-			p.handlesMap[key] = n - 1
-		} else {
-			delete(p.handlesMap, key)
-		}
-		h.Release()
-		p.handles = append(p.handles[:i], p.handles[i+1:]...)
-		p.handles[len(p.handles) : len(p.handles)+1][0] = nil
-	}
-
-	seed := time.Now().UnixNano()
-	t.Logf("seed=%d", seed)
-
-	instances := make([]*instance, goroutines)
-	for i := range instances {
-		p := &instance{}
-		p.handlesMap = make(map[uint64]int)
-		p.seed = seed + int64(i)
-		p.rnd = rand.New(rand.NewSource(p.seed))
-		p.nsid = uint64(i)
-		p.ns = c.GetNamespace(p.nsid)
-		p.delete = i%6 == 0
-		p.purge = i%8 == 0
-		p.zap = i%12 == 0 || i%3 == 0
-		instances[i] = p
-	}
-
-	runr := rand.New(rand.NewSource(seed - 1))
-	run := func(rnd *rand.Rand, x []*instance, init func(p *instance) bool, fn func(p *instance, i int) bool) {
-		var (
-			rx []*instance
-			rn []int
-		)
-		if init == nil {
-			rx = append([]*instance{}, x...)
-			rn = make([]int, len(x))
-		} else {
-			for _, p := range x {
-				if init(p) {
-					rx = append(rx, p)
-					rn = append(rn, 0)
-				}
-			}
-		}
-		for len(rx) > 0 {
-			i := rand.Intn(len(rx))
-			if fn(rx[i], rn[i]) {
-				rn[i]++
-			} else {
-				rx = append(rx[:i], rx[i+1:]...)
-				rn = append(rn[:i], rn[i+1:]...)
-			}
+			t.Errorf("Cache.Get on #%d return nil", key)
 		}
 	}
 
-	// Get and release.
-	run(runr, instances, nil, func(p *instance, i int) bool {
-		if i < iterations {
-			if len(p.handles) == 0 || p.rnd.Int()%2 == 0 {
-				instanceGet(p, uint64(p.rnd.Intn(keymax)))
-			} else {
-				instanceRelease(p, p.rnd.Intn(len(p.handles)))
-			}
-			return true
-		} else {
-			return false
+	h2.Release()
+	if h := c.Get(0, 2, nil); h != nil {
+		t.Errorf("Cache.Get on #2 return non-nil: %v", h.Value())
+	}
+
+	if delFuncCalled != 4 {
+		t.Errorf("delFunc isn't called 4 times: got=%d", delFuncCalled)
+	}
+}
+
+func TestLRUCache_Close(t *testing.T) {
+	relFuncCalled := 0
+	relFunc := func() {
+		relFuncCalled++
+	}
+	delFuncCalled := 0
+	delFunc := func() {
+		delFuncCalled++
+	}
+
+	c := NewCache(NewLRU(2))
+	set(c, 0, 1, 1, 1, relFunc).Release()
+	set(c, 0, 2, 2, 1, relFunc).Release()
+
+	h3 := set(c, 0, 3, 3, 1, relFunc)
+	if h3 == nil {
+		t.Error("Cache.Get on #3 return nil")
+	}
+	if ok := c.Delete(0, 3, delFunc); !ok {
+		t.Error("Cache.Delete on #3 return false")
+	}
+
+	c.Close()
+
+	if relFuncCalled != 3 {
+		t.Errorf("relFunc isn't called 3 times: got=%d", relFuncCalled)
+	}
+	if delFuncCalled != 1 {
+		t.Errorf("delFunc isn't called 1 times: got=%d", delFuncCalled)
+	}
+}
+
+func BenchmarkLRUCache(b *testing.B) {
+	c := NewCache(NewLRU(10000))
+
+	b.SetParallelism(10)
+	b.RunParallel(func(pb *testing.PB) {
+		r := rand.New(rand.NewSource(time.Now().UnixNano()))
+
+		for pb.Next() {
+			key := uint64(r.Intn(1000000))
+			c.Get(0, key, func() (int, Value) {
+				return 1, key
+			}).Release()
 		}
 	})
-
-	if used, cap := c.Used(), c.Capacity(); used > cap {
-		t.Errorf("Used > capacity, used=%d cap=%d", used, cap)
-	}
-
-	// Check effective objects.
-	for i, p := range instances {
-		if int(p.effective) < len(p.handlesMap) {
-			t.Errorf("#%d effective objects < acquired handle, eo=%d ah=%d", i, p.effective, len(p.handlesMap))
-		}
-	}
-
-	if want := int(cnt.created - cnt.released); c.Size() != want {
-		t.Errorf("Invalid cache size, want=%d got=%d", want, c.Size())
-	}
-
-	// First delete.
-	run(runr, instances, func(p *instance) bool {
-		p.wantDel = p.effective
-		return p.delete
-	}, func(p *instance, i int) bool {
-		key := uint64(i)
-		if key < keymax {
-			_, wantExist := p.handlesMap[key]
-			gotExist := p.ns.Delete(key, func(exist, pending bool) {
-				p.delfinCalledAll++
-				if exist {
-					p.delfinCalledEff++
-				}
-			})
-			if !gotExist && wantExist {
-				t.Errorf("delete on NS#%d KEY#%d not found", p.nsid, key)
-			}
-			return true
-		} else {
-			return false
-		}
-	})
-
-	// Second delete.
-	run(runr, instances, func(p *instance) bool {
-		p.delfinCalled = 0
-		return p.delete
-	}, func(p *instance, i int) bool {
-		key := uint64(i)
-		if key < keymax {
-			gotExist := p.ns.Delete(key, func(exist, pending bool) {
-				if exist && !pending {
-					t.Errorf("delete fin on NS#%d KEY#%d exist and not pending for deletion", p.nsid, key)
-				}
-				p.delfinCalled++
-			})
-			if gotExist {
-				t.Errorf("delete on NS#%d KEY#%d found", p.nsid, key)
-			}
-			return true
-		} else {
-			if p.delfinCalled != keymax {
-				t.Errorf("(2) #%d not all delete fin called, diff=%d", p.ns, keymax-p.delfinCalled)
-			}
-			return false
-		}
-	})
-
-	// Purge.
-	run(runr, instances, func(p *instance) bool {
-		return p.purge
-	}, func(p *instance, i int) bool {
-		p.ns.Purge(func(ns, key uint64) {
-			p.purgefinCalled++
-		})
-		return false
-	})
-
-	if want := int(cnt.created - cnt.released); c.Size() != want {
-		t.Errorf("Invalid cache size, want=%d got=%d", want, c.Size())
-	}
-
-	// Release.
-	run(runr, instances, func(p *instance) bool {
-		return !p.zap
-	}, func(p *instance, i int) bool {
-		if len(p.handles) > 0 {
-			instanceRelease(p, len(p.handles)-1)
-			return true
-		} else {
-			return false
-		}
-	})
-
-	if want := int(cnt.created - cnt.released); c.Size() != want {
-		t.Errorf("Invalid cache size, want=%d got=%d", want, c.Size())
-	}
-
-	// Zap.
-	run(runr, instances, func(p *instance) bool {
-		return p.zap
-	}, func(p *instance, i int) bool {
-		p.ns.Zap()
-		p.handles = nil
-		p.handlesMap = nil
-		return false
-	})
-
-	if want := int(cnt.created - cnt.released); c.Size() != want {
-		t.Errorf("Invalid cache size, want=%d got=%d", want, c.Size())
-	}
-
-	if notrel, used := int(cnt.created-cnt.released), c.Used(); notrel != used {
-		t.Errorf("Invalid used value, want=%d got=%d", notrel, used)
-	}
-
-	c.Purge(nil)
-
-	for _, p := range instances {
-		if p.delete {
-			if p.delfinCalledAll != keymax {
-				t.Errorf("#%d not all delete fin called, purge=%v zap=%v diff=%d", p.nsid, p.purge, p.zap, keymax-p.delfinCalledAll)
-			}
-			if p.delfinCalledEff != p.wantDel {
-				t.Errorf("#%d not all effective delete fin called, diff=%d", p.nsid, p.wantDel-p.delfinCalledEff)
-			}
-			if p.purge && p.purgefinCalled > 0 {
-				t.Errorf("#%d some purge fin called, delete=%v zap=%v n=%d", p.nsid, p.delete, p.zap, p.purgefinCalled)
-			}
-		} else {
-			if p.purge {
-				if p.purgefinCalled != p.wantDel {
-					t.Errorf("#%d not all purge fin called, delete=%v zap=%v diff=%d", p.nsid, p.delete, p.zap, p.wantDel-p.purgefinCalled)
-				}
-			}
-		}
-	}
-
-	if cnt.created != cnt.released {
-		t.Errorf("Some cache object weren't released, created=%d released=%d", cnt.created, cnt.released)
-	}
-}
-
-func BenchmarkLRUCache_Set(b *testing.B) {
-	c := NewLRUCache(0)
-	ns := c.GetNamespace(0)
-	b.ResetTimer()
-	for i := uint64(0); i < uint64(b.N); i++ {
-		set(ns, i, "", 1, nil)
-	}
-}
-
-func BenchmarkLRUCache_Get(b *testing.B) {
-	c := NewLRUCache(0)
-	ns := c.GetNamespace(0)
-	b.ResetTimer()
-	for i := uint64(0); i < uint64(b.N); i++ {
-		set(ns, i, "", 1, nil)
-	}
-	b.ResetTimer()
-	for i := uint64(0); i < uint64(b.N); i++ {
-		ns.Get(i, nil)
-	}
-}
-
-func BenchmarkLRUCache_Get2(b *testing.B) {
-	c := NewLRUCache(0)
-	ns := c.GetNamespace(0)
-	b.ResetTimer()
-	for i := uint64(0); i < uint64(b.N); i++ {
-		set(ns, i, "", 1, nil)
-	}
-	b.ResetTimer()
-	for i := uint64(0); i < uint64(b.N); i++ {
-		ns.Get(i, func() (charge int, value interface{}) {
-			return 0, nil
-		})
-	}
-}
-
-func BenchmarkLRUCache_Release(b *testing.B) {
-	c := NewLRUCache(0)
-	ns := c.GetNamespace(0)
-	handles := make([]Handle, b.N)
-	for i := uint64(0); i < uint64(b.N); i++ {
-		handles[i] = set(ns, i, "", 1, nil)
-	}
-	b.ResetTimer()
-	for _, h := range handles {
-		h.Release()
-	}
-}
-
-func BenchmarkLRUCache_SetRelease(b *testing.B) {
-	capacity := b.N / 100
-	if capacity <= 0 {
-		capacity = 10
-	}
-	c := NewLRUCache(capacity)
-	ns := c.GetNamespace(0)
-	b.ResetTimer()
-	for i := uint64(0); i < uint64(b.N); i++ {
-		set(ns, i, "", 1, nil).Release()
-	}
-}
-
-func BenchmarkLRUCache_SetReleaseTwice(b *testing.B) {
-	capacity := b.N / 100
-	if capacity <= 0 {
-		capacity = 10
-	}
-	c := NewLRUCache(capacity)
-	ns := c.GetNamespace(0)
-	b.ResetTimer()
-
-	na := b.N / 2
-	nb := b.N - na
-
-	for i := uint64(0); i < uint64(na); i++ {
-		set(ns, i, "", 1, nil).Release()
-	}
-
-	for i := uint64(0); i < uint64(nb); i++ {
-		set(ns, i, "", 1, nil).Release()
-	}
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/cache/lru.go

@@ -0,0 +1,195 @@
+// Copyright (c) 2012, Suryandaru Triandana <syndtr@gmail.com>
+// All rights reserved.
+//
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+package cache
+
+import (
+	"sync"
+	"unsafe"
+)
+
+type lruNode struct {
+	n   *Node
+	h   *Handle
+	ban bool
+
+	next, prev *lruNode
+}
+
+func (n *lruNode) insert(at *lruNode) {
+	x := at.next
+	at.next = n
+	n.prev = at
+	n.next = x
+	x.prev = n
+}
+
+func (n *lruNode) remove() {
+	if n.prev != nil {
+		n.prev.next = n.next
+		n.next.prev = n.prev
+		n.prev = nil
+		n.next = nil
+	} else {
+		panic("BUG: removing removed node")
+	}
+}
+
+type lru struct {
+	mu       sync.Mutex
+	capacity int
+	used     int
+	recent   lruNode
+}
+
+func (r *lru) reset() {
+	r.recent.next = &r.recent
+	r.recent.prev = &r.recent
+	r.used = 0
+}
+
+func (r *lru) Capacity() int {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+	return r.capacity
+}
+
+func (r *lru) SetCapacity(capacity int) {
+	var evicted []*lruNode
+
+	r.mu.Lock()
+	r.capacity = capacity
+	for r.used > r.capacity {
+		rn := r.recent.prev
+		if rn == nil {
+			panic("BUG: invalid LRU used or capacity counter")
+		}
+		rn.remove()
+		rn.n.CacheData = nil
+		r.used -= rn.n.Size()
+		evicted = append(evicted, rn)
+	}
+	r.mu.Unlock()
+
+	for _, rn := range evicted {
+		rn.h.Release()
+	}
+}
+
+func (r *lru) Promote(n *Node) {
+	var evicted []*lruNode
+
+	r.mu.Lock()
+	if n.CacheData == nil {
+		if n.Size() <= r.capacity {
+			rn := &lruNode{n: n, h: n.GetHandle()}
+			rn.insert(&r.recent)
+			n.CacheData = unsafe.Pointer(rn)
+			r.used += n.Size()
+
+			for r.used > r.capacity {
+				rn := r.recent.prev
+				if rn == nil {
+					panic("BUG: invalid LRU used or capacity counter")
+				}
+				rn.remove()
+				rn.n.CacheData = nil
+				r.used -= rn.n.Size()
+				evicted = append(evicted, rn)
+			}
+		}
+	} else {
+		rn := (*lruNode)(n.CacheData)
+		if !rn.ban {
+			rn.remove()
+			rn.insert(&r.recent)
+		}
+	}
+	r.mu.Unlock()
+
+	for _, rn := range evicted {
+		rn.h.Release()
+	}
+}
+
+func (r *lru) Ban(n *Node) {
+	r.mu.Lock()
+	if n.CacheData == nil {
+		n.CacheData = unsafe.Pointer(&lruNode{n: n, ban: true})
+	} else {
+		rn := (*lruNode)(n.CacheData)
+		if !rn.ban {
+			rn.remove()
+			rn.ban = true
+			r.used -= rn.n.Size()
+			r.mu.Unlock()
+
+			rn.h.Release()
+			rn.h = nil
+			return
+		}
+	}
+	r.mu.Unlock()
+}
+
+func (r *lru) Evict(n *Node) {
+	r.mu.Lock()
+	rn := (*lruNode)(n.CacheData)
+	if rn == nil || rn.ban {
+		r.mu.Unlock()
+		return
+	}
+	n.CacheData = nil
+	r.mu.Unlock()
+
+	rn.h.Release()
+}
+
+func (r *lru) EvictNS(ns uint64) {
+	var evicted []*lruNode
+
+	r.mu.Lock()
+	for e := r.recent.prev; e != &r.recent; {
+		rn := e
+		e = e.prev
+		if rn.n.NS() == ns {
+			rn.remove()
+			rn.n.CacheData = nil
+			r.used -= rn.n.Size()
+			evicted = append(evicted, rn)
+		}
+	}
+	r.mu.Unlock()
+
+	for _, rn := range evicted {
+		rn.h.Release()
+	}
+}
+
+func (r *lru) EvictAll() {
+	r.mu.Lock()
+	back := r.recent.prev
+	for rn := back; rn != &r.recent; rn = rn.prev {
+		rn.n.CacheData = nil
+	}
+	r.reset()
+	r.mu.Unlock()
+
+	for rn := back; rn != &r.recent; rn = rn.prev {
+		rn.h.Release()
+	}
+}
+
+func (r *lru) Close() error {
+	return nil
+}
+
+// NewLRU create a new LRU-cache.
+func NewLRU(capacity int) Cacher {
+	r := &lru{capacity: capacity}
+	r.reset()
+	return r
+}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/cache/lru_cache.go

@@ -1,622 +0,0 @@
-// Copyright (c) 2012, Suryandaru Triandana <syndtr@gmail.com>
-// All rights reserved.
-//
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-package cache
-
-import (
-	"sync"
-	"sync/atomic"
-
-	"github.com/syndtr/goleveldb/leveldb/util"
-)
-
-// The LLRB implementation were taken from https://github.com/petar/GoLLRB.
-// Which contains the following header:
-//
-// Copyright 2010 Petar Maymounkov. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-// lruCache represent a LRU cache state.
-type lruCache struct {
-	mu                sync.Mutex
-	recent            lruNode
-	table             map[uint64]*lruNs
-	capacity          int
-	used, size, alive int
-}
-
-// NewLRUCache creates a new initialized LRU cache with the given capacity.
-func NewLRUCache(capacity int) Cache {
-	c := &lruCache{
-		table:    make(map[uint64]*lruNs),
-		capacity: capacity,
-	}
-	c.recent.rNext = &c.recent
-	c.recent.rPrev = &c.recent
-	return c
-}
-
-func (c *lruCache) Capacity() int {
-	c.mu.Lock()
-	defer c.mu.Unlock()
-	return c.capacity
-}
-
-func (c *lruCache) Used() int {
-	c.mu.Lock()
-	defer c.mu.Unlock()
-	return c.used
-}
-
-func (c *lruCache) Size() int {
-	c.mu.Lock()
-	defer c.mu.Unlock()
-	return c.size
-}
-
-func (c *lruCache) NumObjects() int {
-	c.mu.Lock()
-	defer c.mu.Unlock()
-	return c.alive
-}
-
-// SetCapacity set cache capacity.
-func (c *lruCache) SetCapacity(capacity int) {
-	c.mu.Lock()
-	c.capacity = capacity
-	c.evict()
-	c.mu.Unlock()
-}
-
-// GetNamespace return namespace object for given id.
-func (c *lruCache) GetNamespace(id uint64) Namespace {
-	c.mu.Lock()
-	defer c.mu.Unlock()
-
-	if ns, ok := c.table[id]; ok {
-		return ns
-	}
-
-	ns := &lruNs{lru: c, id: id}
-	c.table[id] = ns
-	return ns
-}
-
-func (c *lruCache) ZapNamespace(id uint64) {
-	c.mu.Lock()
-	if ns, exist := c.table[id]; exist {
-		ns.zapNB()
-		delete(c.table, id)
-	}
-	c.mu.Unlock()
-}
-
-func (c *lruCache) PurgeNamespace(id uint64, fin PurgeFin) {
-	c.mu.Lock()
-	if ns, exist := c.table[id]; exist {
-		ns.purgeNB(fin)
-	}
-	c.mu.Unlock()
-}
-
-// Purge purge entire cache.
-func (c *lruCache) Purge(fin PurgeFin) {
-	c.mu.Lock()
-	for _, ns := range c.table {
-		ns.purgeNB(fin)
-	}
-	c.mu.Unlock()
-}
-
-func (c *lruCache) Zap() {
-	c.mu.Lock()
-	for _, ns := range c.table {
-		ns.zapNB()
-	}
-	c.table = make(map[uint64]*lruNs)
-	c.mu.Unlock()
-}
-
-func (c *lruCache) evict() {
-	top := &c.recent
-	for n := c.recent.rPrev; c.used > c.capacity && n != top; {
-		if n.state != nodeEffective {
-			panic("evicting non effective node")
-		}
-		n.state = nodeEvicted
-		n.rRemove()
-		n.derefNB()
-		c.used -= n.charge
-		n = c.recent.rPrev
-	}
-}
-
-type lruNs struct {
-	lru    *lruCache
-	id     uint64
-	rbRoot *lruNode
-	state  nsState
-}
-
-func (ns *lruNs) rbGetOrCreateNode(h *lruNode, key uint64) (hn, n *lruNode) {
-	if h == nil {
-		n = &lruNode{ns: ns, key: key}
-		return n, n
-	}
-
-	if key < h.key {
-		hn, n = ns.rbGetOrCreateNode(h.rbLeft, key)
-		if hn != nil {
-			h.rbLeft = hn
-		} else {
-			return nil, n
-		}
-	} else if key > h.key {
-		hn, n = ns.rbGetOrCreateNode(h.rbRight, key)
-		if hn != nil {
-			h.rbRight = hn
-		} else {
-			return nil, n
-		}
-	} else {
-		return nil, h
-	}
-
-	if rbIsRed(h.rbRight) && !rbIsRed(h.rbLeft) {
-		h = rbRotLeft(h)
-	}
-	if rbIsRed(h.rbLeft) && rbIsRed(h.rbLeft.rbLeft) {
-		h = rbRotRight(h)
-	}
-	if rbIsRed(h.rbLeft) && rbIsRed(h.rbRight) {
-		rbFlip(h)
-	}
-	return h, n
-}
-
-func (ns *lruNs) getOrCreateNode(key uint64) *lruNode {
-	hn, n := ns.rbGetOrCreateNode(ns.rbRoot, key)
-	if hn != nil {
-		ns.rbRoot = hn
-		ns.rbRoot.rbBlack = true
-	}
-	return n
-}
-
-func (ns *lruNs) rbGetNode(key uint64) *lruNode {
-	h := ns.rbRoot
-	for h != nil {
-		switch {
-		case key < h.key:
-			h = h.rbLeft
-		case key > h.key:
-			h = h.rbRight
-		default:
-			return h
-		}
-	}
-	return nil
-}
-
-func (ns *lruNs) getNode(key uint64) *lruNode {
-	return ns.rbGetNode(key)
-}
-
-func (ns *lruNs) rbDeleteNode(h *lruNode, key uint64) *lruNode {
-	if h == nil {
-		return nil
-	}
-
-	if key < h.key {
-		if h.rbLeft == nil { // key not present. Nothing to delete
-			return h
-		}
-		if !rbIsRed(h.rbLeft) && !rbIsRed(h.rbLeft.rbLeft) {
-			h = rbMoveLeft(h)
-		}
-		h.rbLeft = ns.rbDeleteNode(h.rbLeft, key)
-	} else {
-		if rbIsRed(h.rbLeft) {
-			h = rbRotRight(h)
-		}
-		// If @key equals @h.key and no right children at @h
-		if h.key == key && h.rbRight == nil {
-			return nil
-		}
-		if h.rbRight != nil && !rbIsRed(h.rbRight) && !rbIsRed(h.rbRight.rbLeft) {
-			h = rbMoveRight(h)
-		}
-		// If @key equals @h.key, and (from above) 'h.Right != nil'
-		if h.key == key {
-			var x *lruNode
-			h.rbRight, x = rbDeleteMin(h.rbRight)
-			if x == nil {
-				panic("logic")
-			}
-			x.rbLeft, h.rbLeft = h.rbLeft, nil
-			x.rbRight, h.rbRight = h.rbRight, nil
-			x.rbBlack = h.rbBlack
-			h = x
-		} else { // Else, @key is bigger than @h.key
-			h.rbRight = ns.rbDeleteNode(h.rbRight, key)
-		}
-	}
-
-	return rbFixup(h)
-}
-
-func (ns *lruNs) deleteNode(key uint64) {
-	ns.rbRoot = ns.rbDeleteNode(ns.rbRoot, key)
-	if ns.rbRoot != nil {
-		ns.rbRoot.rbBlack = true
-	}
-}
-
-func (ns *lruNs) rbIterateNodes(h *lruNode, pivot uint64, iter func(n *lruNode) bool) bool {
-	if h == nil {
-		return true
-	}
-	if h.key >= pivot {
-		if !ns.rbIterateNodes(h.rbLeft, pivot, iter) {
-			return false
-		}
-		if !iter(h) {
-			return false
-		}
-	}
-	return ns.rbIterateNodes(h.rbRight, pivot, iter)
-}
-
-func (ns *lruNs) iterateNodes(iter func(n *lruNode) bool) {
-	ns.rbIterateNodes(ns.rbRoot, 0, iter)
-}
-
-func (ns *lruNs) Get(key uint64, setf SetFunc) Handle {
-	ns.lru.mu.Lock()
-	defer ns.lru.mu.Unlock()
-
-	if ns.state != nsEffective {
-		return nil
-	}
-
-	var n *lruNode
-	if setf == nil {
-		n = ns.getNode(key)
-		if n == nil {
-			return nil
-		}
-	} else {
-		n = ns.getOrCreateNode(key)
-	}
-	switch n.state {
-	case nodeZero:
-		charge, value := setf()
-		if value == nil {
-			ns.deleteNode(key)
-			return nil
-		}
-		if charge < 0 {
-			charge = 0
-		}
-
-		n.value = value
-		n.charge = charge
-		n.state = nodeEvicted
-
-		ns.lru.size += charge
-		ns.lru.alive++
-
-		fallthrough
-	case nodeEvicted:
-		if n.charge == 0 {
-			break
-		}
-
-		// Insert to recent list.
-		n.state = nodeEffective
-		n.ref++
-		ns.lru.used += n.charge
-		ns.lru.evict()
-
-		fallthrough
-	case nodeEffective:
-		// Bump to front.
-		n.rRemove()
-		n.rInsert(&ns.lru.recent)
-	case nodeDeleted:
-		// Do nothing.
-	default:
-		panic("invalid state")
-	}
-	n.ref++
-
-	return &lruHandle{node: n}
-}
-
-func (ns *lruNs) Delete(key uint64, fin DelFin) bool {
-	ns.lru.mu.Lock()
-	defer ns.lru.mu.Unlock()
-
-	if ns.state != nsEffective {
-		if fin != nil {
-			fin(false, false)
-		}
-		return false
-	}
-
-	n := ns.getNode(key)
-	if n == nil {
-		if fin != nil {
-			fin(false, false)
-		}
-		return false
-
-	}
-
-	switch n.state {
-	case nodeEffective:
-		ns.lru.used -= n.charge
-		n.state = nodeDeleted
-		n.delfin = fin
-		n.rRemove()
-		n.derefNB()
-	case nodeEvicted:
-		n.state = nodeDeleted
-		n.delfin = fin
-	case nodeDeleted:
-		if fin != nil {
-			fin(true, true)
-		}
-		return false
-	default:
-		panic("invalid state")
-	}
-
-	return true
-}
-
-func (ns *lruNs) purgeNB(fin PurgeFin) {
-	if ns.state == nsEffective {
-		var nodes []*lruNode
-		ns.iterateNodes(func(n *lruNode) bool {
-			nodes = append(nodes, n)
-			return true
-		})
-		for _, n := range nodes {
-			switch n.state {
-			case nodeEffective:
-				ns.lru.used -= n.charge
-				n.state = nodeDeleted
-				n.purgefin = fin
-				n.rRemove()
-				n.derefNB()
-			case nodeEvicted:
-				n.state = nodeDeleted
-				n.purgefin = fin
-			case nodeDeleted:
-			default:
-				panic("invalid state")
-			}
-		}
-	}
-}
-
-func (ns *lruNs) Purge(fin PurgeFin) {
-	ns.lru.mu.Lock()
-	ns.purgeNB(fin)
-	ns.lru.mu.Unlock()
-}
-
-func (ns *lruNs) zapNB() {
-	if ns.state == nsEffective {
-		ns.state = nsZapped
-
-		ns.iterateNodes(func(n *lruNode) bool {
-			if n.state == nodeEffective {
-				ns.lru.used -= n.charge
-				n.rRemove()
-			}
-			ns.lru.size -= n.charge
-			n.state = nodeDeleted
-			n.fin()
-
-			return true
-		})
-		ns.rbRoot = nil
-	}
-}
-
-func (ns *lruNs) Zap() {
-	ns.lru.mu.Lock()
-	ns.zapNB()
-	delete(ns.lru.table, ns.id)
-	ns.lru.mu.Unlock()
-}
-
-type lruNode struct {
-	ns *lruNs
-
-	rNext, rPrev    *lruNode
-	rbLeft, rbRight *lruNode
-	rbBlack         bool
-
-	key      uint64
-	value    interface{}
-	charge   int
-	ref      int
-	state    nodeState
-	delfin   DelFin
-	purgefin PurgeFin
-}
-
-func (n *lruNode) rInsert(at *lruNode) {
-	x := at.rNext
-	at.rNext = n
-	n.rPrev = at
-	n.rNext = x
-	x.rPrev = n
-}
-
-func (n *lruNode) rRemove() bool {
-	if n.rPrev == nil {
-		return false
-	}
-
-	n.rPrev.rNext = n.rNext
-	n.rNext.rPrev = n.rPrev
-	n.rPrev = nil
-	n.rNext = nil
-
-	return true
-}
-
-func (n *lruNode) fin() {
-	if r, ok := n.value.(util.Releaser); ok {
-		r.Release()
-	}
-	if n.purgefin != nil {
-		if n.delfin != nil {
-			panic("conflicting delete and purge fin")
-		}
-		n.purgefin(n.ns.id, n.key)
-		n.purgefin = nil
-	} else if n.delfin != nil {
-		n.delfin(true, false)
-		n.delfin = nil
-	}
-}
-
-func (n *lruNode) derefNB() {
-	n.ref--
-	if n.ref == 0 {
-		if n.ns.state == nsEffective {
-			// Remove elemement.
-			n.ns.deleteNode(n.key)
-			n.ns.lru.size -= n.charge
-			n.ns.lru.alive--
-			n.fin()
-		}
-		n.value = nil
-	} else if n.ref < 0 {
-		panic("leveldb/cache: lruCache: negative node reference")
-	}
-}
-
-func (n *lruNode) deref() {
-	n.ns.lru.mu.Lock()
-	n.derefNB()
-	n.ns.lru.mu.Unlock()
-}
-
-type lruHandle struct {
-	node *lruNode
-	once uint32
-}
-
-func (h *lruHandle) Value() interface{} {
-	if atomic.LoadUint32(&h.once) == 0 {
-		return h.node.value
-	}
-	return nil
-}
-
-func (h *lruHandle) Release() {
-	if !atomic.CompareAndSwapUint32(&h.once, 0, 1) {
-		return
-	}
-	h.node.deref()
-	h.node = nil
-}
-
-func rbIsRed(h *lruNode) bool {
-	if h == nil {
-		return false
-	}
-	return !h.rbBlack
-}
-
-func rbRotLeft(h *lruNode) *lruNode {
-	x := h.rbRight
-	if x.rbBlack {
-		panic("rotating a black link")
-	}
-	h.rbRight = x.rbLeft
-	x.rbLeft = h
-	x.rbBlack = h.rbBlack
-	h.rbBlack = false
-	return x
-}
-
-func rbRotRight(h *lruNode) *lruNode {
-	x := h.rbLeft
-	if x.rbBlack {
-		panic("rotating a black link")
-	}
-	h.rbLeft = x.rbRight
-	x.rbRight = h
-	x.rbBlack = h.rbBlack
-	h.rbBlack = false
-	return x
-}
-
-func rbFlip(h *lruNode) {
-	h.rbBlack = !h.rbBlack
-	h.rbLeft.rbBlack = !h.rbLeft.rbBlack
-	h.rbRight.rbBlack = !h.rbRight.rbBlack
-}
-
-func rbMoveLeft(h *lruNode) *lruNode {
-	rbFlip(h)
-	if rbIsRed(h.rbRight.rbLeft) {
-		h.rbRight = rbRotRight(h.rbRight)
-		h = rbRotLeft(h)
-		rbFlip(h)
-	}
-	return h
-}
-
-func rbMoveRight(h *lruNode) *lruNode {
-	rbFlip(h)
-	if rbIsRed(h.rbLeft.rbLeft) {
-		h = rbRotRight(h)
-		rbFlip(h)
-	}
-	return h
-}
-
-func rbFixup(h *lruNode) *lruNode {
-	if rbIsRed(h.rbRight) {
-		h = rbRotLeft(h)
-	}
-
-	if rbIsRed(h.rbLeft) && rbIsRed(h.rbLeft.rbLeft) {
-		h = rbRotRight(h)
-	}
-
-	if rbIsRed(h.rbLeft) && rbIsRed(h.rbRight) {
-		rbFlip(h)
-	}
-
-	return h
-}
-
-func rbDeleteMin(h *lruNode) (hn, n *lruNode) {
-	if h == nil {
-		return nil, nil
-	}
-	if h.rbLeft == nil {
-		return nil, h
-	}
-
-	if !rbIsRed(h.rbLeft) && !rbIsRed(h.rbLeft.rbLeft) {
-		h = rbMoveLeft(h)
-	}
-
-	h.rbLeft, n = rbDeleteMin(h.rbLeft)
-
-	return rbFixup(h), n
-}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/config.go

@@ -1,40 +0,0 @@
-// Copyright (c) 2012, Suryandaru Triandana <syndtr@gmail.com>
-// All rights reserved.
-//
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-package leveldb
-
-const (
-	kNumLevels = 7
-
-	// Level-0 compaction is started when we hit this many files.
-	kL0_CompactionTrigger float64 = 4
-
-	// Soft limit on number of level-0 files.  We slow down writes at this point.
-	kL0_SlowdownWritesTrigger = 8
-
-	// Maximum number of level-0 files.  We stop writes at this point.
-	kL0_StopWritesTrigger = 12
-
-	// Maximum level to which a new compacted memdb is pushed if it
-	// does not create overlap.  We try to push to level 2 to avoid the
-	// relatively expensive level 0=>1 compactions and to avoid some
-	// expensive manifest file operations.  We do not push all the way to
-	// the largest level since that can generate a lot of wasted disk
-	// space if the same key space is being repeatedly overwritten.
-	kMaxMemCompactLevel = 2
-
-	// Maximum size of a table.
-	kMaxTableSize = 2 * 1048576
-
-	// Maximum bytes of overlaps in grandparent (i.e., level+2) before we
-	// stop building a single file in a level->level+1 compaction.
-	kMaxGrandParentOverlapBytes = 10 * kMaxTableSize
-
-	// Maximum number of bytes in all compacted files.  We avoid expanding
-	// the lower level file set of a compaction if it would make the
-	// total compaction cover more than this many bytes.
-	kExpCompactionMaxBytes = 25 * kMaxTableSize
-)

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/corrupt_test.go

@@ -9,13 +9,12 @@ package leveldb
 import (
 	"bytes"
 	"fmt"
+	"github.com/syndtr/goleveldb/leveldb/filter"
+	"github.com/syndtr/goleveldb/leveldb/opt"
+	"github.com/syndtr/goleveldb/leveldb/storage"
 	"io"
 	"math/rand"
 	"testing"
-
-	"github.com/syndtr/goleveldb/leveldb/cache"
-	"github.com/syndtr/goleveldb/leveldb/opt"
-	"github.com/syndtr/goleveldb/leveldb/storage"
 )
 
 const ctValSize = 1000
@@ -32,8 +31,8 @@ func newDbCorruptHarnessWopt(t *testing.T, o *opt.Options) *dbCorruptHarness {
 
 func newDbCorruptHarness(t *testing.T) *dbCorruptHarness {
 	return newDbCorruptHarnessWopt(t, &opt.Options{
-		BlockCache: cache.NewLRUCache(100),
-		Strict:     opt.StrictJournalChecksum,
+		BlockCacheCapacity: 100,
+		Strict:             opt.StrictJournalChecksum,
 	})
 }
 
@@ -96,21 +95,22 @@ func (h *dbCorruptHarness) deleteRand(n, max int, rnd *rand.Rand) {
 	}
 }
 
-func (h *dbCorruptHarness) corrupt(ft storage.FileType, offset, n int) {
+func (h *dbCorruptHarness) corrupt(ft storage.FileType, fi, offset, n int) {
 	p := &h.dbHarness
 	t := p.t
 
-	var file storage.File
 	ff, _ := p.stor.GetFiles(ft)
-	for _, f := range ff {
-		if file == nil || f.Num() > file.Num() {
-			file = f
-		}
+	sff := files(ff)
+	sff.sort()
+	if fi < 0 {
+		fi = len(sff) - 1
 	}
-	if file == nil {
-		t.Fatalf("no such file with type %q", ft)
+	if fi >= len(sff) {
+		t.Fatalf("no such file with type %q with index %d", ft, fi)
 	}
 
+	file := sff[fi]
+
 	r, err := file.Open()
 	if err != nil {
 		t.Fatal("cannot open file: ", err)
@@ -225,8 +225,8 @@ func TestCorruptDB_Journal(t *testing.T) {
 	h.build(100)
 	h.check(100, 100)
 	h.closeDB()
-	h.corrupt(storage.TypeJournal, 19, 1)
-	h.corrupt(storage.TypeJournal, 32*1024+1000, 1)
+	h.corrupt(storage.TypeJournal, -1, 19, 1)
+	h.corrupt(storage.TypeJournal, -1, 32*1024+1000, 1)
 
 	h.openDB()
 	h.check(36, 36)
@@ -242,7 +242,7 @@ func TestCorruptDB_Table(t *testing.T) {
 	h.compactRangeAt(0, "", "")
 	h.compactRangeAt(1, "", "")
 	h.closeDB()
-	h.corrupt(storage.TypeTable, 100, 1)
+	h.corrupt(storage.TypeTable, -1, 100, 1)
 
 	h.openDB()
 	h.check(99, 99)
@@ -256,7 +256,7 @@ func TestCorruptDB_TableIndex(t *testing.T) {
 	h.build(10000)
 	h.compactMem()
 	h.closeDB()
-	h.corrupt(storage.TypeTable, -2000, 500)
+	h.corrupt(storage.TypeTable, -1, -2000, 500)
 
 	h.openDB()
 	h.check(5000, 9999)
@@ -267,9 +267,9 @@ func TestCorruptDB_TableIndex(t *testing.T) {
 func TestCorruptDB_MissingManifest(t *testing.T) {
 	rnd := rand.New(rand.NewSource(0x0badda7a))
 	h := newDbCorruptHarnessWopt(t, &opt.Options{
-		BlockCache:  cache.NewLRUCache(100),
-		Strict:      opt.StrictJournalChecksum,
-		WriteBuffer: 1000 * 60,
+		BlockCacheCapacity: 100,
+		Strict:             opt.StrictJournalChecksum,
+		WriteBuffer:        1000 * 60,
 	})
 
 	h.build(1000)
@@ -355,7 +355,7 @@ func TestCorruptDB_CorruptedManifest(t *testing.T) {
 	h.compactMem()
 	h.compactRange("", "")
 	h.closeDB()
-	h.corrupt(storage.TypeManifest, 0, 1000)
+	h.corrupt(storage.TypeManifest, -1, 0, 1000)
 	h.openAssert(false)
 
 	h.recover()
@@ -370,7 +370,7 @@ func TestCorruptDB_CompactionInputError(t *testing.T) {
 	h.build(10)
 	h.compactMem()
 	h.closeDB()
-	h.corrupt(storage.TypeTable, 100, 1)
+	h.corrupt(storage.TypeTable, -1, 100, 1)
 
 	h.openDB()
 	h.check(9, 9)
@@ -387,7 +387,7 @@ func TestCorruptDB_UnrelatedKeys(t *testing.T) {
 	h.build(10)
 	h.compactMem()
 	h.closeDB()
-	h.corrupt(storage.TypeTable, 100, 1)
+	h.corrupt(storage.TypeTable, -1, 100, 1)
 
 	h.openDB()
 	h.put(string(tkey(1000)), string(tval(1000, ctValSize)))
@@ -470,3 +470,31 @@ func TestCorruptDB_MissingTableFiles(t *testing.T) {
 
 	h.close()
 }
+
+func TestCorruptDB_RecoverTable(t *testing.T) {
+	h := newDbCorruptHarnessWopt(t, &opt.Options{
+		WriteBuffer:         112 * opt.KiB,
+		CompactionTableSize: 90 * opt.KiB,
+		Filter:              filter.NewBloomFilter(10),
+	})
+
+	h.build(1000)
+	h.compactMem()
+	h.compactRangeAt(0, "", "")
+	h.compactRangeAt(1, "", "")
+	seq := h.db.seq
+	h.closeDB()
+	h.corrupt(storage.TypeTable, 0, 1000, 1)
+	h.corrupt(storage.TypeTable, 3, 10000, 1)
+	// Corrupted filter shouldn't affect recovery.
+	h.corrupt(storage.TypeTable, 3, 113888, 10)
+	h.corrupt(storage.TypeTable, -1, 20000, 1)
+
+	h.recover()
+	if h.db.seq != seq {
+		t.Errorf("invalid seq, want=%d got=%d", seq, h.db.seq)
+	}
+	h.check(985, 985)
+
+	h.close()
+}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/db.go

@@ -7,7 +7,7 @@
 package leveldb
 
 import (
-	"errors"
+	"container/list"
 	"fmt"
 	"io"
 	"os"
@@ -17,6 +17,7 @@ import (
 	"sync/atomic"
 	"time"
 
+	"github.com/syndtr/goleveldb/leveldb/errors"
 	"github.com/syndtr/goleveldb/leveldb/iterator"
 	"github.com/syndtr/goleveldb/leveldb/journal"
 	"github.com/syndtr/goleveldb/leveldb/memdb"
@@ -46,7 +47,7 @@ type DB struct {
 
 	// Snapshot.
 	snapsMu   sync.Mutex
-	snapsRoot snapshotElement
+	snapsList *list.List
 
 	// Stats.
 	aliveSnaps, aliveIters int32
@@ -56,18 +57,19 @@ type DB struct {
 	writeMergedC chan bool
 	writeLockC   chan struct{}
 	writeAckC    chan error
+	writeDelay   time.Duration
+	writeDelayN  int
 	journalC     chan *Batch
 	journalAckC  chan error
 
 	// Compaction.
-	tcompCmdC     chan cCmd
-	tcompPauseC   chan chan<- struct{}
-	tcompTriggerC chan struct{}
-	mcompCmdC     chan cCmd
-	mcompTriggerC chan struct{}
-	compErrC      chan error
-	compErrSetC   chan error
-	compStats     [kNumLevels]cStats
+	tcompCmdC   chan cCmd
+	tcompPauseC chan chan<- struct{}
+	mcompCmdC   chan cCmd
+	compErrC    chan error
+	compPerErrC chan error
+	compErrSetC chan error
+	compStats   []cStats
 
 	// Close.
 	closeW sync.WaitGroup
@@ -82,9 +84,11 @@ func openDB(s *session) (*DB, error) {
 	db := &DB{
 		s: s,
 		// Initial sequence
-		seq: s.stSeq,
+		seq: s.stSeqNum,
 		// MemDB
 		memPool: make(chan *memdb.DB, 1),
+		// Snapshot
+		snapsList: list.New(),
 		// Write
 		writeC:       make(chan *Batch),
 		writeMergedC: make(chan bool),
@@ -93,17 +97,16 @@ func openDB(s *session) (*DB, error) {
 		journalC:     make(chan *Batch),
 		journalAckC:  make(chan error),
 		// Compaction
-		tcompCmdC:     make(chan cCmd),
-		tcompPauseC:   make(chan chan<- struct{}),
-		tcompTriggerC: make(chan struct{}, 1),
-		mcompCmdC:     make(chan cCmd),
-		mcompTriggerC: make(chan struct{}, 1),
-		compErrC:      make(chan error),
-		compErrSetC:   make(chan error),
+		tcompCmdC:   make(chan cCmd),
+		tcompPauseC: make(chan chan<- struct{}),
+		mcompCmdC:   make(chan cCmd),
+		compErrC:    make(chan error),
+		compPerErrC: make(chan error),
+		compErrSetC: make(chan error),
+		compStats:   make([]cStats, s.o.GetNumLevel()),
 		// Close
 		closeC: make(chan struct{}),
 	}
-	db.initSnapshot()
 
 	if err := db.recoverJournal(); err != nil {
 		return nil, err
@@ -119,14 +122,14 @@ func openDB(s *session) (*DB, error) {
 		return nil, err
 	}
 
-	// Don't include compaction error goroutine into wait group.
+	// Doesn't need to be included in the wait group.
 	go db.compactionError()
+	go db.mpoolDrain()
 
 	db.closeW.Add(3)
 	go db.tCompaction()
 	go db.mCompaction()
 	go db.jWriter()
-	go db.mpoolDrain()
 
 	s.logf("db@open done T·%v", time.Since(start))
 
@@ -253,6 +256,10 @@ func RecoverFile(path string, o *opt.Options) (db *DB, err error) {
 }
 
 func recoverTable(s *session, o *opt.Options) error {
+	o = dupOptions(o)
+	// Mask StrictReader, lets StrictRecovery doing its job.
+	o.Strict &= ^opt.StrictReader
+
 	// Get all tables and sort it by file number.
 	tableFiles_, err := s.getFiles(storage.TypeTable)
 	if err != nil {
@@ -261,10 +268,16 @@ func recoverTable(s *session, o *opt.Options) error {
 	tableFiles := files(tableFiles_)
 	tableFiles.sort()
 
-	var mSeq uint64
-	var good, corrupted int
-	rec := new(sessionRecord)
-	bpool := util.NewBufferPool(o.GetBlockSize() + 5)
+	var (
+		maxSeq                                                            uint64
+		recoveredKey, goodKey, corruptedKey, corruptedBlock, droppedTable int
+
+		// We will drop corrupted table.
+		strict = o.GetStrict(opt.StrictRecovery)
+
+		rec   = &sessionRecord{numLevel: o.GetNumLevel()}
+		bpool = util.NewBufferPool(o.GetBlockSize() + 5)
+	)
 	buildTable := func(iter iterator.Iterator) (tmp storage.File, size int64, err error) {
 		tmp = s.newTemp()
 		writer, err := tmp.Create()
@@ -311,7 +324,12 @@ func recoverTable(s *session, o *opt.Options) error {
 		if err != nil {
 			return err
 		}
-		defer reader.Close()
+		var closed bool
+		defer func() {
+			if !closed {
+				reader.Close()
+			}
+		}()
 
 		// Get file size.
 		size, err := reader.Seek(0, 2)
@@ -319,25 +337,32 @@ func recoverTable(s *session, o *opt.Options) error {
 			return err
 		}
 
-		var tSeq uint64
-		var tgood, tcorrupted, blockerr int
-		var imin, imax []byte
-		tr := table.NewReader(reader, size, nil, bpool, o)
+		var (
+			tSeq                                     uint64
+			tgoodKey, tcorruptedKey, tcorruptedBlock int
+			imin, imax                               []byte
+		)
+		tr, err := table.NewReader(reader, size, storage.NewFileInfo(file), nil, bpool, o)
+		if err != nil {
+			return err
+		}
 		iter := tr.NewIterator(nil, nil)
 		iter.(iterator.ErrorCallbackSetter).SetErrorCallback(func(err error) {
-			s.logf("table@recovery found error @%d %q", file.Num(), err)
-			blockerr++
+			if errors.IsCorrupted(err) {
+				s.logf("table@recovery block corruption @%d %q", file.Num(), err)
+				tcorruptedBlock++
+			}
 		})
 
 		// Scan the table.
 		for iter.Next() {
 			key := iter.Key()
-			_, seq, _, ok := parseIkey(key)
-			if !ok {
-				tcorrupted++
+			_, seq, _, kerr := parseIkey(key)
+			if kerr != nil {
+				tcorruptedKey++
 				continue
 			}
-			tgood++
+			tgoodKey++
 			if seq > tSeq {
 				tSeq = seq
 			}
@@ -352,8 +377,18 @@ func recoverTable(s *session, o *opt.Options) error {
 		}
 		iter.Release()
 
-		if tgood > 0 {
-			if tcorrupted > 0 || blockerr > 0 {
+		goodKey += tgoodKey
+		corruptedKey += tcorruptedKey
+		corruptedBlock += tcorruptedBlock
+
+		if strict && (tcorruptedKey > 0 || tcorruptedBlock > 0) {
+			droppedTable++
+			s.logf("table@recovery dropped @%d Gk·%d Ck·%d Cb·%d S·%d Q·%d", file.Num(), tgoodKey, tcorruptedKey, tcorruptedBlock, size, tSeq)
+			return nil
+		}
+
+		if tgoodKey > 0 {
+			if tcorruptedKey > 0 || tcorruptedBlock > 0 {
 				// Rebuild the table.
 				s.logf("table@recovery rebuilding @%d", file.Num())
 				iter := tr.NewIterator(nil, nil)
@@ -362,25 +397,25 @@ func recoverTable(s *session, o *opt.Options) error {
 				if err != nil {
 					return err
 				}
+				closed = true
 				reader.Close()
 				if err := file.Replace(tmp); err != nil {
 					return err
 				}
 				size = newSize
 			}
-			if tSeq > mSeq {
-				mSeq = tSeq
+			if tSeq > maxSeq {
+				maxSeq = tSeq
 			}
+			recoveredKey += tgoodKey
 			// Add table to level 0.
 			rec.addTable(0, file.Num(), uint64(size), imin, imax)
-			s.logf("table@recovery recovered @%d N·%d C·%d B·%d S·%d Q·%d", file.Num(), tgood, tcorrupted, blockerr, size, tSeq)
+			s.logf("table@recovery recovered @%d Gk·%d Ck·%d Cb·%d S·%d Q·%d", file.Num(), tgoodKey, tcorruptedKey, tcorruptedBlock, size, tSeq)
 		} else {
-			s.logf("table@recovery unrecoverable @%d C·%d B·%d S·%d", file.Num(), tcorrupted, blockerr, size)
+			droppedTable++
+			s.logf("table@recovery unrecoverable @%d Ck·%d Cb·%d S·%d", file.Num(), tcorruptedKey, tcorruptedBlock, size)
 		}
 
-		good += tgood
-		corrupted += tcorrupted
-
 		return nil
 	}
 
@@ -397,11 +432,11 @@ func recoverTable(s *session, o *opt.Options) error {
 			}
 		}
 
-		s.logf("table@recovery recovered F·%d N·%d C·%d Q·%d", len(tableFiles), good, corrupted, mSeq)
+		s.logf("table@recovery recovered F·%d N·%d Gk·%d Ck·%d Q·%d", len(tableFiles), recoveredKey, goodKey, corruptedKey, maxSeq)
 	}
 
 	// Set sequence number.
-	rec.setSeq(mSeq + 1)
+	rec.setSeqNum(maxSeq)
 
 	// Create new manifest.
 	if err := s.create(); err != nil {
@@ -484,26 +519,30 @@ func (db *DB) recoverJournal() error {
 				if err == io.EOF {
 					break
 				}
-				return err
+				return errors.SetFile(err, file)
 			}
 
 			buf.Reset()
 			if _, err := buf.ReadFrom(r); err != nil {
 				if err == io.ErrUnexpectedEOF {
+					// This is error returned due to corruption, with strict == false.
 					continue
 				} else {
-					return err
+					return errors.SetFile(err, file)
 				}
 			}
-			if err := batch.decode(buf.Bytes()); err != nil {
-				return err
-			}
-			if err := batch.memReplay(mem); err != nil {
-				return err
+			if err := batch.memDecodeAndReplay(db.seq, buf.Bytes(), mem); err != nil {
+				if strict || !errors.IsCorrupted(err) {
+					return errors.SetFile(err, file)
+				} else {
+					db.s.logf("journal error: %v (skipped)", err)
+					// We won't apply sequence number as it might be corrupted.
+					continue
+				}
 			}
 
 			// Save sequence number.
-			db.seq = batch.seq + uint64(batch.len())
+			db.seq = batch.seq + uint64(batch.Len())
 
 			// Flush it if large enough.
 			if mem.Size() >= writeBuffer {
@@ -564,7 +603,7 @@ func (db *DB) recoverJournal() error {
 }
 
 func (db *DB) get(key []byte, seq uint64, ro *opt.ReadOptions) (value []byte, err error) {
-	ikey := newIKey(key, seq, tSeek)
+	ikey := newIkey(key, seq, ktSeek)
 
 	em, fm := db.getMems()
 	for _, m := range [...]*memDB{em, fm} {
@@ -575,9 +614,13 @@ func (db *DB) get(key []byte, seq uint64, ro *opt.ReadOptions) (value []byte, er
 
 		mk, mv, me := m.mdb.Find(ikey)
 		if me == nil {
-			ukey, _, t, ok := parseIkey(mk)
-			if ok && db.s.icmp.uCompare(ukey, key) == 0 {
-				if t == tDel {
+			ukey, _, kt, kerr := parseIkey(mk)
+			if kerr != nil {
+				// Shouldn't have had happen.
+				panic(kerr)
+			}
+			if db.s.icmp.uCompare(ukey, key) == 0 {
+				if kt == ktDel {
 					return nil, ErrNotFound
 				}
 				return append([]byte{}, mv...), nil
@@ -588,17 +631,60 @@ func (db *DB) get(key []byte, seq uint64, ro *opt.ReadOptions) (value []byte, er
 	}
 
 	v := db.s.version()
-	value, cSched, err := v.get(ikey, ro)
+	value, cSched, err := v.get(ikey, ro, false)
 	v.release()
 	if cSched {
 		// Trigger table compaction.
-		db.compTrigger(db.tcompTriggerC)
+		db.compSendTrigger(db.tcompCmdC)
+	}
+	return
+}
+
+func (db *DB) has(key []byte, seq uint64, ro *opt.ReadOptions) (ret bool, err error) {
+	ikey := newIkey(key, seq, ktSeek)
+
+	em, fm := db.getMems()
+	for _, m := range [...]*memDB{em, fm} {
+		if m == nil {
+			continue
+		}
+		defer m.decref()
+
+		mk, _, me := m.mdb.Find(ikey)
+		if me == nil {
+			ukey, _, kt, kerr := parseIkey(mk)
+			if kerr != nil {
+				// Shouldn't have had happen.
+				panic(kerr)
+			}
+			if db.s.icmp.uCompare(ukey, key) == 0 {
+				if kt == ktDel {
+					return false, nil
+				}
+				return true, nil
+			}
+		} else if me != ErrNotFound {
+			return false, me
+		}
+	}
+
+	v := db.s.version()
+	_, cSched, err := v.get(ikey, ro, true)
+	v.release()
+	if cSched {
+		// Trigger table compaction.
+		db.compSendTrigger(db.tcompCmdC)
+	}
+	if err == nil {
+		ret = true
+	} else if err == ErrNotFound {
+		err = nil
 	}
 	return
 }
 
 // Get gets the value for the given key. It returns ErrNotFound if the
-// DB does not contain the key.
+// DB does not contains the key.
 //
 // The returned slice is its own copy, it is safe to modify the contents
 // of the returned slice.
@@ -609,7 +695,23 @@ func (db *DB) Get(key []byte, ro *opt.ReadOptions) (value []byte, err error) {
 		return
 	}
 
-	return db.get(key, db.getSeq(), ro)
+	se := db.acquireSnapshot()
+	defer db.releaseSnapshot(se)
+	return db.get(key, se.seq, ro)
+}
+
+// Has returns true if the DB does contains the given key.
+//
+// It is safe to modify the contents of the argument after Get returns.
+func (db *DB) Has(key []byte, ro *opt.ReadOptions) (ret bool, err error) {
+	err = db.ok()
+	if err != nil {
+		return
+	}
+
+	se := db.acquireSnapshot()
+	defer db.releaseSnapshot(se)
+	return db.has(key, se.seq, ro)
 }
 
 // NewIterator returns an iterator for the latest snapshot of the
@@ -633,9 +735,11 @@ func (db *DB) NewIterator(slice *util.Range, ro *opt.ReadOptions) iterator.Itera
 		return iterator.NewEmptyIterator(err)
 	}
 
-	snap := db.newSnapshot()
-	defer snap.Release()
-	return snap.NewIterator(slice, ro)
+	se := db.acquireSnapshot()
+	defer db.releaseSnapshot(se)
+	// Iterator holds 'version' lock, 'version' is immutable so snapshot
+	// can be released after iterator created.
+	return db.newIterator(se.seq, slice, ro)
 }
 
 // GetSnapshot returns a latest snapshot of the underlying DB. A snapshot
@@ -655,7 +759,7 @@ func (db *DB) GetSnapshot() (*Snapshot, error) {
 //
 // Property names:
 //	leveldb.num-files-at-level{n}
-//		Returns the number of filer at level 'n'.
+//		Returns the number of files at level 'n'.
 //	leveldb.stats
 //		Returns statistics of the underlying DB.
 //	leveldb.sstables
@@ -685,12 +789,13 @@ func (db *DB) GetProperty(name string) (value string, err error) {
 	v := db.s.version()
 	defer v.release()
 
+	numFilesPrefix := "num-files-at-level"
 	switch {
-	case strings.HasPrefix(p, "num-files-at-level"):
+	case strings.HasPrefix(p, numFilesPrefix):
 		var level uint
 		var rest string
-		n, _ := fmt.Scanf("%d%s", &level, &rest)
-		if n != 1 || level >= kNumLevels {
+		n, _ := fmt.Sscanf(p[len(numFilesPrefix):], "%d%s", &level, &rest)
+		if n != 1 || int(level) >= db.s.o.GetNumLevel() {
 			err = errors.New("leveldb: GetProperty: invalid property: " + name)
 		} else {
 			value = fmt.Sprint(v.tLen(int(level)))
@@ -718,8 +823,8 @@ func (db *DB) GetProperty(name string) (value string, err error) {
 	case p == "blockpool":
 		value = fmt.Sprintf("%v", db.s.tops.bpool)
 	case p == "cachedblock":
-		if bc := db.s.o.GetBlockCache(); bc != nil {
-			value = fmt.Sprintf("%d", bc.Size())
+		if db.s.tops.bcache != nil {
+			value = fmt.Sprintf("%d", db.s.tops.bcache.Size())
 		} else {
 			value = "<nil>"
 		}
@@ -752,8 +857,8 @@ func (db *DB) SizeOf(ranges []util.Range) (Sizes, error) {
 
 	sizes := make(Sizes, 0, len(ranges))
 	for _, r := range ranges {
-		imin := newIKey(r.Start, kMaxSeq, tSeek)
-		imax := newIKey(r.Limit, kMaxSeq, tSeek)
+		imin := newIkey(r.Start, kMaxSeq, ktSeek)
+		imax := newIkey(r.Limit, kMaxSeq, ktSeek)
 		start, err := v.offsetOf(imin)
 		if err != nil {
 			return nil, err
@@ -796,18 +901,23 @@ func (db *DB) Close() error {
 	default:
 	}
 
+	// Signal all goroutines.
 	close(db.closeC)
 
-	// Wait for the close WaitGroup.
+	// Wait for all gorotines to exit.
 	db.closeW.Wait()
 
-	// Close journal.
+	// Lock writer and closes journal.
 	db.writeLockC <- struct{}{}
 	if db.journal != nil {
 		db.journal.Close()
 		db.journalWriter.Close()
 	}
 
+	if db.writeDelayN > 0 {
+		db.logf("db@write was delayed N·%d T·%v", db.writeDelayN, db.writeDelay)
+	}
+
 	// Close session.
 	db.s.close()
 	db.logf("db@close done T·%v", time.Since(start))
@@ -827,7 +937,6 @@ func (db *DB) Close() error {
 	db.journalWriter = nil
 	db.journalFile = nil
 	db.frozenJournalFile = nil
-	db.snapsRoot = snapshotElement{}
 	db.closer = nil
 
 	return err

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/db_compaction.go

@@ -7,11 +7,12 @@
 package leveldb
 
 import (
-	"errors"
 	"sync"
 	"time"
 
+	"github.com/syndtr/goleveldb/leveldb/errors"
 	"github.com/syndtr/goleveldb/leveldb/memdb"
+	"github.com/syndtr/goleveldb/leveldb/opt"
 )
 
 var (
@@ -68,7 +69,7 @@ type cMem struct {
 }
 
 func newCMem(s *session) *cMem {
-	return &cMem{s: s, rec: new(sessionRecord)}
+	return &cMem{s: s, rec: &sessionRecord{numLevel: s.o.GetNumLevel()}}
 }
 
 func (c *cMem) flush(mem *memdb.DB, level int) error {
@@ -84,7 +85,9 @@ func (c *cMem) flush(mem *memdb.DB, level int) error {
 
 	// Pick level.
 	if level < 0 {
-		level = s.version_NB().pickLevel(t.imin.ukey(), t.imax.ukey())
+		v := s.version()
+		level = v.pickLevel(t.imin.ukey(), t.imax.ukey())
+		v.release()
 	}
 	c.rec.addTableFile(level, t)
 
@@ -95,24 +98,32 @@ func (c *cMem) flush(mem *memdb.DB, level int) error {
 }
 
 func (c *cMem) reset() {
-	c.rec = new(sessionRecord)
+	c.rec = &sessionRecord{numLevel: c.s.o.GetNumLevel()}
 }
 
 func (c *cMem) commit(journal, seq uint64) error {
 	c.rec.setJournalNum(journal)
-	c.rec.setSeq(seq)
+	c.rec.setSeqNum(seq)
 
 	// Commit changes.
 	return c.s.commit(c.rec)
 }
 
 func (db *DB) compactionError() {
-	var err error
+	var (
+		err     error
+		wlocked bool
+	)
 noerr:
+	// No error.
 	for {
 		select {
 		case err = <-db.compErrSetC:
-			if err != nil {
+			switch {
+			case err == nil:
+			case errors.IsCorrupted(err):
+				goto hasperr
+			default:
 				goto haserr
 			}
 		case _, _ = <-db.closeC:
@@ -120,17 +131,39 @@ noerr:
 		}
 	}
 haserr:
+	// Transient error.
 	for {
 		select {
 		case db.compErrC <- err:
 		case err = <-db.compErrSetC:
-			if err == nil {
+			switch {
+			case err == nil:
 				goto noerr
+			case errors.IsCorrupted(err):
+				goto hasperr
+			default:
 			}
 		case _, _ = <-db.closeC:
 			return
 		}
 	}
+hasperr:
+	// Persistent error.
+	for {
+		select {
+		case db.compErrC <- err:
+		case db.compPerErrC <- err:
+		case db.writeLockC <- struct{}{}:
+			// Hold write lock, so that write won't pass-through.
+			wlocked = true
+		case _, _ = <-db.closeC:
+			if wlocked {
+				// We should release the lock or Close will hang.
+				<-db.writeLockC
+			}
+			return
+		}
+	}
 }
 
 type compactionTransactCounter int
@@ -139,12 +172,17 @@ func (cnt *compactionTransactCounter) incr() {
 	*cnt++
 }
 
-func (db *DB) compactionTransact(name string, exec func(cnt *compactionTransactCounter) error, rollback func() error) {
+type compactionTransactInterface interface {
+	run(cnt *compactionTransactCounter) error
+	revert() error
+}
+
+func (db *DB) compactionTransact(name string, t compactionTransactInterface) {
 	defer func() {
 		if x := recover(); x != nil {
-			if x == errCompactionTransactExiting && rollback != nil {
-				if err := rollback(); err != nil {
-					db.logf("%s rollback error %q", name, err)
+			if x == errCompactionTransactExiting {
+				if err := t.revert(); err != nil {
+					db.logf("%s revert error %q", name, err)
 				}
 			}
 			panic(x)
@@ -156,9 +194,13 @@ func (db *DB) compactionTransact(name string, exec func(cnt *compactionTransactC
 		backoffMax = 8 * time.Second
 		backoffMul = 2 * time.Second
 	)
-	backoff := backoffMin
-	backoffT := time.NewTimer(backoff)
-	lastCnt := compactionTransactCounter(0)
+	var (
+		backoff  = backoffMin
+		backoffT = time.NewTimer(backoff)
+		lastCnt  = compactionTransactCounter(0)
+
+		disableBackoff = db.s.o.GetDisableCompactionBackoff()
+	)
 	for n := 0; ; n++ {
 		// Check wether the DB is closed.
 		if db.isClosed() {
@@ -170,11 +212,19 @@ func (db *DB) compactionTransact(name string, exec func(cnt *compactionTransactC
 
 		// Execute.
 		cnt := compactionTransactCounter(0)
-		err := exec(&cnt)
+		err := t.run(&cnt)
+		if err != nil {
+			db.logf("%s error I·%d %q", name, cnt, err)
+		}
 
 		// Set compaction error status.
 		select {
 		case db.compErrSetC <- err:
+		case perr := <-db.compPerErrC:
+			if err != nil {
+				db.logf("%s exiting (persistent error %q)", name, perr)
+				db.compactionExitTransact()
+			}
 		case _, _ = <-db.closeC:
 			db.logf("%s exiting", name)
 			db.compactionExitTransact()
@@ -182,31 +232,56 @@ func (db *DB) compactionTransact(name string, exec func(cnt *compactionTransactC
 		if err == nil {
 			return
 		}
-		db.logf("%s error I·%d %q", name, cnt, err)
-
-		// Reset backoff duration if counter is advancing.
-		if cnt > lastCnt {
-			backoff = backoffMin
-			lastCnt = cnt
-		}
-
-		// Backoff.
-		backoffT.Reset(backoff)
-		if backoff < backoffMax {
-			backoff *= backoffMul
-			if backoff > backoffMax {
-				backoff = backoffMax
-			}
-		}
-		select {
-		case <-backoffT.C:
-		case _, _ = <-db.closeC:
-			db.logf("%s exiting", name)
+		if errors.IsCorrupted(err) {
+			db.logf("%s exiting (corruption detected)", name)
 			db.compactionExitTransact()
 		}
+
+		if !disableBackoff {
+			// Reset backoff duration if counter is advancing.
+			if cnt > lastCnt {
+				backoff = backoffMin
+				lastCnt = cnt
+			}
+
+			// Backoff.
+			backoffT.Reset(backoff)
+			if backoff < backoffMax {
+				backoff *= backoffMul
+				if backoff > backoffMax {
+					backoff = backoffMax
+				}
+			}
+			select {
+			case <-backoffT.C:
+			case _, _ = <-db.closeC:
+				db.logf("%s exiting", name)
+				db.compactionExitTransact()
+			}
+		}
 	}
 }
 
+type compactionTransactFunc struct {
+	runFunc    func(cnt *compactionTransactCounter) error
+	revertFunc func() error
+}
+
+func (t *compactionTransactFunc) run(cnt *compactionTransactCounter) error {
+	return t.runFunc(cnt)
+}
+
+func (t *compactionTransactFunc) revert() error {
+	if t.revertFunc != nil {
+		return t.revertFunc()
+	}
+	return nil
+}
+
+func (db *DB) compactionTransactFunc(name string, run func(cnt *compactionTransactCounter) error, revert func() error) {
+	db.compactionTransact(name, &compactionTransactFunc{run, revert})
+}
+
 func (db *DB) compactionExitTransact() {
 	panic(errCompactionTransactExiting)
 }
@@ -232,20 +307,23 @@ func (db *DB) memCompaction() {
 	}
 
 	// Pause table compaction.
-	ch := make(chan struct{})
+	resumeC := make(chan struct{})
 	select {
-	case db.tcompPauseC <- (chan<- struct{})(ch):
+	case db.tcompPauseC <- (chan<- struct{})(resumeC):
+	case <-db.compPerErrC:
+		close(resumeC)
+		resumeC = nil
 	case _, _ = <-db.closeC:
 		return
 	}
 
-	db.compactionTransact("mem@flush", func(cnt *compactionTransactCounter) (err error) {
+	db.compactionTransactFunc("mem@flush", func(cnt *compactionTransactCounter) (err error) {
 		stats.startTimer()
 		defer stats.stopTimer()
 		return c.flush(mem.mdb, -1)
 	}, func() error {
 		for _, r := range c.rec.addedTables {
-			db.logf("mem@flush rollback @%d", r.num)
+			db.logf("mem@flush revert @%d", r.num)
 			f := db.s.getTableFile(r.num)
 			if err := f.Remove(); err != nil {
 				return err
@@ -254,7 +332,7 @@ func (db *DB) memCompaction() {
 		return nil
 	})
 
-	db.compactionTransact("mem@commit", func(cnt *compactionTransactCounter) (err error) {
+	db.compactionTransactFunc("mem@commit", func(cnt *compactionTransactCounter) (err error) {
 		stats.startTimer()
 		defer stats.stopTimer()
 		return c.commit(db.journalFile.Num(), db.frozenSeq)
@@ -271,26 +349,223 @@ func (db *DB) memCompaction() {
 	db.dropFrozenMem()
 
 	// Resume table compaction.
-	select {
-	case <-ch:
-	case _, _ = <-db.closeC:
-		return
+	if resumeC != nil {
+		select {
+		case <-resumeC:
+			close(resumeC)
+		case _, _ = <-db.closeC:
+			return
+		}
 	}
 
 	// Trigger table compaction.
-	db.compTrigger(db.mcompTriggerC)
+	db.compSendTrigger(db.tcompCmdC)
+}
+
+type tableCompactionBuilder struct {
+	db           *DB
+	s            *session
+	c            *compaction
+	rec          *sessionRecord
+	stat0, stat1 *cStatsStaging
+
+	snapHasLastUkey bool
+	snapLastUkey    []byte
+	snapLastSeq     uint64
+	snapIter        int
+	snapKerrCnt     int
+	snapDropCnt     int
+
+	kerrCnt int
+	dropCnt int
+
+	minSeq    uint64
+	strict    bool
+	tableSize int
+
+	tw *tWriter
+}
+
+func (b *tableCompactionBuilder) appendKV(key, value []byte) error {
+	// Create new table if not already.
+	if b.tw == nil {
+		// Check for pause event.
+		if b.db != nil {
+			select {
+			case ch := <-b.db.tcompPauseC:
+				b.db.pauseCompaction(ch)
+			case _, _ = <-b.db.closeC:
+				b.db.compactionExitTransact()
+			default:
+			}
+		}
+
+		// Create new table.
+		var err error
+		b.tw, err = b.s.tops.create()
+		if err != nil {
+			return err
+		}
+	}
+
+	// Write key/value into table.
+	return b.tw.append(key, value)
+}
+
+func (b *tableCompactionBuilder) needFlush() bool {
+	return b.tw.tw.BytesLen() >= b.tableSize
+}
+
+func (b *tableCompactionBuilder) flush() error {
+	t, err := b.tw.finish()
+	if err != nil {
+		return err
+	}
+	b.rec.addTableFile(b.c.level+1, t)
+	b.stat1.write += t.size
+	b.s.logf("table@build created L%d@%d N·%d S·%s %q:%q", b.c.level+1, t.file.Num(), b.tw.tw.EntriesLen(), shortenb(int(t.size)), t.imin, t.imax)
+	b.tw = nil
+	return nil
+}
+
+func (b *tableCompactionBuilder) cleanup() {
+	if b.tw != nil {
+		b.tw.drop()
+		b.tw = nil
+	}
+}
+
+func (b *tableCompactionBuilder) run(cnt *compactionTransactCounter) error {
+	snapResumed := b.snapIter > 0
+	hasLastUkey := b.snapHasLastUkey // The key might has zero length, so this is necessary.
+	lastUkey := append([]byte{}, b.snapLastUkey...)
+	lastSeq := b.snapLastSeq
+	b.kerrCnt = b.snapKerrCnt
+	b.dropCnt = b.snapDropCnt
+	// Restore compaction state.
+	b.c.restore()
+
+	defer b.cleanup()
+
+	b.stat1.startTimer()
+	defer b.stat1.stopTimer()
+
+	iter := b.c.newIterator()
+	defer iter.Release()
+	for i := 0; iter.Next(); i++ {
+		// Incr transact counter.
+		cnt.incr()
+
+		// Skip until last state.
+		if i < b.snapIter {
+			continue
+		}
+
+		resumed := false
+		if snapResumed {
+			resumed = true
+			snapResumed = false
+		}
+
+		ikey := iter.Key()
+		ukey, seq, kt, kerr := parseIkey(ikey)
+
+		if kerr == nil {
+			shouldStop := !resumed && b.c.shouldStopBefore(ikey)
+
+			if !hasLastUkey || b.s.icmp.uCompare(lastUkey, ukey) != 0 {
+				// First occurrence of this user key.
+
+				// Only rotate tables if ukey doesn't hop across.
+				if b.tw != nil && (shouldStop || b.needFlush()) {
+					if err := b.flush(); err != nil {
+						return err
+					}
+
+					// Creates snapshot of the state.
+					b.c.save()
+					b.snapHasLastUkey = hasLastUkey
+					b.snapLastUkey = append(b.snapLastUkey[:0], lastUkey...)
+					b.snapLastSeq = lastSeq
+					b.snapIter = i
+					b.snapKerrCnt = b.kerrCnt
+					b.snapDropCnt = b.dropCnt
+				}
+
+				hasLastUkey = true
+				lastUkey = append(lastUkey[:0], ukey...)
+				lastSeq = kMaxSeq
+			}
+
+			switch {
+			case lastSeq <= b.minSeq:
+				// Dropped because newer entry for same user key exist
+				fallthrough // (A)
+			case kt == ktDel && seq <= b.minSeq && b.c.baseLevelForKey(lastUkey):
+				// For this user key:
+				// (1) there is no data in higher levels
+				// (2) data in lower levels will have larger seq numbers
+				// (3) data in layers that are being compacted here and have
+				//     smaller seq numbers will be dropped in the next
+				//     few iterations of this loop (by rule (A) above).
+				// Therefore this deletion marker is obsolete and can be dropped.
+				lastSeq = seq
+				b.dropCnt++
+				continue
+			default:
+				lastSeq = seq
+			}
+		} else {
+			if b.strict {
+				return kerr
+			}
+
+			// Don't drop corrupted keys.
+			hasLastUkey = false
+			lastUkey = lastUkey[:0]
+			lastSeq = kMaxSeq
+			b.kerrCnt++
+		}
+
+		if err := b.appendKV(ikey, iter.Value()); err != nil {
+			return err
+		}
+	}
+
+	if err := iter.Error(); err != nil {
+		return err
+	}
+
+	// Finish last table.
+	if b.tw != nil && !b.tw.empty() {
+		return b.flush()
+	}
+	return nil
+}
+
+func (b *tableCompactionBuilder) revert() error {
+	for _, at := range b.rec.addedTables {
+		b.s.logf("table@build revert @%d", at.num)
+		f := b.s.getTableFile(at.num)
+		if err := f.Remove(); err != nil {
+			return err
+		}
+	}
+	return nil
 }
 
 func (db *DB) tableCompaction(c *compaction, noTrivial bool) {
-	rec := new(sessionRecord)
-	rec.addCompactionPointer(c.level, c.imax)
+	defer c.release()
+
+	rec := &sessionRecord{numLevel: db.s.o.GetNumLevel()}
+	rec.addCompPtr(c.level, c.imax)
 
 	if !noTrivial && c.trivial() {
 		t := c.tables[0][0]
 		db.logf("table@move L%d@%d -> L%d", c.level, t.file.Num(), c.level+1)
-		rec.deleteTable(c.level, t.file.Num())
+		rec.delTable(c.level, t.file.Num())
 		rec.addTableFile(c.level+1, t)
-		db.compactionTransact("table@move", func(cnt *compactionTransactCounter) (err error) {
+		db.compactionTransactFunc("table@move", func(cnt *compactionTransactCounter) (err error) {
 			return db.s.commit(rec)
 		}, nil)
 		return
@@ -301,184 +576,34 @@ func (db *DB) tableCompaction(c *compaction, noTrivial bool) {
 		for _, t := range tables {
 			stats[i].read += t.size
 			// Insert deleted tables into record
-			rec.deleteTable(c.level+i, t.file.Num())
+			rec.delTable(c.level+i, t.file.Num())
 		}
 	}
 	sourceSize := int(stats[0].read + stats[1].read)
 	minSeq := db.minSeq()
 	db.logf("table@compaction L%d·%d -> L%d·%d S·%s Q·%d", c.level, len(c.tables[0]), c.level+1, len(c.tables[1]), shortenb(sourceSize), minSeq)
 
-	var snapUkey []byte
-	var snapHasUkey bool
-	var snapSeq uint64
-	var snapIter int
-	var snapDropCnt int
-	var dropCnt int
-	db.compactionTransact("table@build", func(cnt *compactionTransactCounter) (err error) {
-		ukey := append([]byte{}, snapUkey...)
-		hasUkey := snapHasUkey
-		lseq := snapSeq
-		dropCnt = snapDropCnt
-		snapSched := snapIter == 0
-
-		var tw *tWriter
-		finish := func() error {
-			t, err := tw.finish()
-			if err != nil {
-				return err
-			}
-			rec.addTableFile(c.level+1, t)
-			stats[1].write += t.size
-			db.logf("table@build created L%d@%d N·%d S·%s %q:%q", c.level+1, t.file.Num(), tw.tw.EntriesLen(), shortenb(int(t.size)), t.imin, t.imax)
-			return nil
-		}
-
-		defer func() {
-			stats[1].stopTimer()
-			if tw != nil {
-				tw.drop()
-				tw = nil
-			}
-		}()
-
-		stats[1].startTimer()
-		iter := c.newIterator()
-		defer iter.Release()
-		for i := 0; iter.Next(); i++ {
-			// Incr transact counter.
-			cnt.incr()
-
-			// Skip until last state.
-			if i < snapIter {
-				continue
-			}
-
-			ikey := iKey(iter.Key())
-
-			if c.shouldStopBefore(ikey) && tw != nil {
-				err = finish()
-				if err != nil {
-					return
-				}
-				snapSched = true
-				tw = nil
-			}
-
-			// Scheduled for snapshot, snapshot will used to retry compaction
-			// if error occured.
-			if snapSched {
-				snapUkey = append(snapUkey[:0], ukey...)
-				snapHasUkey = hasUkey
-				snapSeq = lseq
-				snapIter = i
-				snapDropCnt = dropCnt
-				snapSched = false
-			}
-
-			if seq, vt, ok := ikey.parseNum(); !ok {
-				// Don't drop error keys
-				ukey = ukey[:0]
-				hasUkey = false
-				lseq = kMaxSeq
-			} else {
-				if !hasUkey || db.s.icmp.uCompare(ikey.ukey(), ukey) != 0 {
-					// First occurrence of this user key
-					ukey = append(ukey[:0], ikey.ukey()...)
-					hasUkey = true
-					lseq = kMaxSeq
-				}
-
-				drop := false
-				if lseq <= minSeq {
-					// Dropped because newer entry for same user key exist
-					drop = true // (A)
-				} else if vt == tDel && seq <= minSeq && c.baseLevelForKey(ukey) {
-					// For this user key:
-					// (1) there is no data in higher levels
-					// (2) data in lower levels will have larger seq numbers
-					// (3) data in layers that are being compacted here and have
-					//     smaller seq numbers will be dropped in the next
-					//     few iterations of this loop (by rule (A) above).
-					// Therefore this deletion marker is obsolete and can be dropped.
-					drop = true
-				}
-
-				lseq = seq
-				if drop {
-					dropCnt++
-					continue
-				}
-			}
-
-			// Create new table if not already
-			if tw == nil {
-				// Check for pause event.
-				select {
-				case ch := <-db.tcompPauseC:
-					db.pauseCompaction(ch)
-				case _, _ = <-db.closeC:
-					db.compactionExitTransact()
-				default:
-				}
-
-				// Create new table.
-				tw, err = db.s.tops.create()
-				if err != nil {
-					return
-				}
-			}
-
-			// Write key/value into table
-			err = tw.append(ikey, iter.Value())
-			if err != nil {
-				return
-			}
-
-			// Finish table if it is big enough
-			if tw.tw.BytesLen() >= kMaxTableSize {
-				err = finish()
-				if err != nil {
-					return
-				}
-				snapSched = true
-				tw = nil
-			}
-		}
-
-		err = iter.Error()
-		if err != nil {
-			return
-		}
-
-		// Finish last table
-		if tw != nil && !tw.empty() {
-			err = finish()
-			if err != nil {
-				return
-			}
-			tw = nil
-		}
-		return
-	}, func() error {
-		for _, r := range rec.addedTables {
-			db.logf("table@build rollback @%d", r.num)
-			f := db.s.getTableFile(r.num)
-			if err := f.Remove(); err != nil {
-				return err
-			}
-		}
-		return nil
-	})
+	b := &tableCompactionBuilder{
+		db:        db,
+		s:         db.s,
+		c:         c,
+		rec:       rec,
+		stat1:     &stats[1],
+		minSeq:    minSeq,
+		strict:    db.s.o.GetStrict(opt.StrictCompaction),
+		tableSize: db.s.o.GetCompactionTableSize(c.level + 1),
+	}
+	db.compactionTransact("table@build", b)
 
 	// Commit changes
-	db.compactionTransact("table@commit", func(cnt *compactionTransactCounter) (err error) {
+	db.compactionTransactFunc("table@commit", func(cnt *compactionTransactCounter) (err error) {
 		stats[1].startTimer()
 		defer stats[1].stopTimer()
 		return db.s.commit(rec)
 	}, nil)
 
 	resultSize := int(stats[1].write)
-	db.logf("table@compaction committed F%s S%s D·%d T·%v", sint(len(rec.addedTables)-len(rec.deletedTables)), sshortenb(resultSize-sourceSize), dropCnt, stats[1].duration)
+	db.logf("table@compaction committed F%s S%s Ke·%d D·%d T·%v", sint(len(rec.addedTables)-len(rec.deletedTables)), sshortenb(resultSize-sourceSize), b.kerrCnt, b.dropCnt, stats[1].duration)
 
 	// Save compaction stats
 	for i := range stats {
@@ -494,14 +619,14 @@ func (db *DB) tableRangeCompaction(level int, umin, umax []byte) {
 			db.tableCompaction(c, true)
 		}
 	} else {
-		v := db.s.version_NB()
-
+		v := db.s.version()
 		m := 1
 		for i, t := range v.tables[1:] {
 			if t.overlaps(db.s.icmp, umin, umax, false) {
 				m = i + 1
 			}
 		}
+		v.release()
 
 		for level := 0; level < m; level++ {
 			if c := db.s.getCompactionRange(level, umin, umax); c != nil {
@@ -518,7 +643,9 @@ func (db *DB) tableAutoCompaction() {
 }
 
 func (db *DB) tableNeedCompaction() bool {
-	return db.s.version_NB().needCompaction()
+	v := db.s.version()
+	defer v.release()
+	return v.needCompaction()
 }
 
 func (db *DB) pauseCompaction(ch chan<- struct{}) {
@@ -538,7 +665,12 @@ type cIdle struct {
 }
 
 func (r cIdle) ack(err error) {
-	r.ackC <- err
+	if r.ackC != nil {
+		defer func() {
+			recover()
+		}()
+		r.ackC <- err
+	}
 }
 
 type cRange struct {
@@ -548,29 +680,45 @@ type cRange struct {
 }
 
 func (r cRange) ack(err error) {
-	defer func() {
-		recover()
-	}()
 	if r.ackC != nil {
+		defer func() {
+			recover()
+		}()
 		r.ackC <- err
 	}
 }
 
-func (db *DB) compSendIdle(compC chan<- cCmd) error {
+// This will trigger auto compation and/or wait for all compaction to be done.
+func (db *DB) compSendIdle(compC chan<- cCmd) (err error) {
 	ch := make(chan error)
 	defer close(ch)
 	// Send cmd.
 	select {
 	case compC <- cIdle{ch}:
-	case err := <-db.compErrC:
-		return err
+	case err = <-db.compErrC:
+		return
 	case _, _ = <-db.closeC:
 		return ErrClosed
 	}
 	// Wait cmd.
-	return <-ch
+	select {
+	case err = <-ch:
+	case err = <-db.compErrC:
+	case _, _ = <-db.closeC:
+		return ErrClosed
+	}
+	return err
 }
 
+// This will trigger auto compaction but will not wait for it.
+func (db *DB) compSendTrigger(compC chan<- cCmd) {
+	select {
+	case compC <- cIdle{}:
+	default:
+	}
+}
+
+// Send range compaction request.
 func (db *DB) compSendRange(compC chan<- cCmd, level int, min, max []byte) (err error) {
 	ch := make(chan error)
 	defer close(ch)
@@ -584,19 +732,14 @@ func (db *DB) compSendRange(compC chan<- cCmd, level int, min, max []byte) (err
 	}
 	// Wait cmd.
 	select {
-	case err = <-db.compErrC:
 	case err = <-ch:
+	case err = <-db.compErrC:
+	case _, _ = <-db.closeC:
+		return ErrClosed
 	}
 	return err
 }
 
-func (db *DB) compTrigger(compTriggerC chan struct{}) {
-	select {
-	case compTriggerC <- struct{}{}:
-	default:
-	}
-}
-
 func (db *DB) mCompaction() {
 	var x cCmd
 
@@ -615,11 +758,14 @@ func (db *DB) mCompaction() {
 	for {
 		select {
 		case x = <-db.mcompCmdC:
-			db.memCompaction()
-			x.ack(nil)
-			x = nil
-		case <-db.mcompTriggerC:
-			db.memCompaction()
+			switch x.(type) {
+			case cIdle:
+				db.memCompaction()
+				x.ack(nil)
+				x = nil
+			default:
+				panic("leveldb: unknown command")
+			}
 		case _, _ = <-db.closeC:
 			return
 		}
@@ -650,7 +796,6 @@ func (db *DB) tCompaction() {
 		if db.tableNeedCompaction() {
 			select {
 			case x = <-db.tcompCmdC:
-			case <-db.tcompTriggerC:
 			case ch := <-db.tcompPauseC:
 				db.pauseCompaction(ch)
 				continue
@@ -666,7 +811,6 @@ func (db *DB) tCompaction() {
 			ackQ = ackQ[:0]
 			select {
 			case x = <-db.tcompCmdC:
-			case <-db.tcompTriggerC:
 			case ch := <-db.tcompPauseC:
 				db.pauseCompaction(ch)
 				continue
@@ -681,6 +825,8 @@ func (db *DB) tCompaction() {
 			case cRange:
 				db.tableRangeCompaction(cmd.level, cmd.min, cmd.max)
 				x.ack(nil)
+			default:
+				panic("leveldb: unknown command")
 			}
 			x = nil
 		}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/db_iter.go

@@ -48,7 +48,7 @@ func (db *DB) newRawIterator(slice *util.Range, ro *opt.ReadOptions) iterator.It
 		i = append(i, fmi)
 	}
 	i = append(i, ti...)
-	strict := db.s.o.GetStrict(opt.StrictIterator) || ro.GetStrict(opt.StrictIterator)
+	strict := opt.GetStrict(db.s.o.Options, ro, opt.StrictReader)
 	mi := iterator.NewMergedIterator(i, db.s.icmp, strict)
 	mi.SetReleaser(&versionReleaser{v: v})
 	return mi
@@ -59,10 +59,10 @@ func (db *DB) newIterator(seq uint64, slice *util.Range, ro *opt.ReadOptions) *d
 	if slice != nil {
 		islice = &util.Range{}
 		if slice.Start != nil {
-			islice.Start = newIKey(slice.Start, kMaxSeq, tSeek)
+			islice.Start = newIkey(slice.Start, kMaxSeq, ktSeek)
 		}
 		if slice.Limit != nil {
-			islice.Limit = newIKey(slice.Limit, kMaxSeq, tSeek)
+			islice.Limit = newIkey(slice.Limit, kMaxSeq, ktSeek)
 		}
 	}
 	rawIter := db.newRawIterator(islice, ro)
@@ -71,7 +71,7 @@ func (db *DB) newIterator(seq uint64, slice *util.Range, ro *opt.ReadOptions) *d
 		icmp:   db.s.icmp,
 		iter:   rawIter,
 		seq:    seq,
-		strict: db.s.o.GetStrict(opt.StrictIterator) || ro.GetStrict(opt.StrictIterator),
+		strict: opt.GetStrict(db.s.o.Options, ro, opt.StrictReader),
 		key:    make([]byte, 0),
 		value:  make([]byte, 0),
 	}
@@ -162,7 +162,7 @@ func (i *dbIter) Seek(key []byte) bool {
 		return false
 	}
 
-	ikey := newIKey(key, i.seq, tSeek)
+	ikey := newIkey(key, i.seq, ktSeek)
 	if i.iter.Seek(ikey) {
 		i.dir = dirSOI
 		return i.next()
@@ -174,15 +174,14 @@ func (i *dbIter) Seek(key []byte) bool {
 
 func (i *dbIter) next() bool {
 	for {
-		ukey, seq, t, ok := parseIkey(i.iter.Key())
-		if ok {
+		if ukey, seq, kt, kerr := parseIkey(i.iter.Key()); kerr == nil {
 			if seq <= i.seq {
-				switch t {
-				case tDel:
+				switch kt {
+				case ktDel:
 					// Skip deleted key.
 					i.key = append(i.key[:0], ukey...)
 					i.dir = dirForward
-				case tVal:
+				case ktVal:
 					if i.dir == dirSOI || i.icmp.uCompare(ukey, i.key) > 0 {
 						i.key = append(i.key[:0], ukey...)
 						i.value = append(i.value[:0], i.iter.Value()...)
@@ -192,7 +191,7 @@ func (i *dbIter) next() bool {
 				}
 			}
 		} else if i.strict {
-			i.setErr(errInvalidIkey)
+			i.setErr(kerr)
 			break
 		}
 		if !i.iter.Next() {
@@ -225,20 +224,19 @@ func (i *dbIter) prev() bool {
 	del := true
 	if i.iter.Valid() {
 		for {
-			ukey, seq, t, ok := parseIkey(i.iter.Key())
-			if ok {
+			if ukey, seq, kt, kerr := parseIkey(i.iter.Key()); kerr == nil {
 				if seq <= i.seq {
 					if !del && i.icmp.uCompare(ukey, i.key) < 0 {
 						return true
 					}
-					del = (t == tDel)
+					del = (kt == ktDel)
 					if !del {
 						i.key = append(i.key[:0], ukey...)
 						i.value = append(i.value[:0], i.iter.Value()...)
 					}
 				}
 			} else if i.strict {
-				i.setErr(errInvalidIkey)
+				i.setErr(kerr)
 				return false
 			}
 			if !i.iter.Prev() {
@@ -267,13 +265,12 @@ func (i *dbIter) Prev() bool {
 		return i.Last()
 	case dirForward:
 		for i.iter.Prev() {
-			ukey, _, _, ok := parseIkey(i.iter.Key())
-			if ok {
+			if ukey, _, _, kerr := parseIkey(i.iter.Key()); kerr == nil {
 				if i.icmp.uCompare(ukey, i.key) < 0 {
 					goto cont
 				}
 			} else if i.strict {
-				i.setErr(errInvalidIkey)
+				i.setErr(kerr)
 				return false
 			}
 		}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/db_snapshot.go

@@ -7,6 +7,8 @@
 package leveldb
 
 import (
+	"container/list"
+	"fmt"
 	"runtime"
 	"sync"
 	"sync/atomic"
@@ -19,51 +21,41 @@ import (
 type snapshotElement struct {
 	seq uint64
 	ref int
-	// Next and previous pointers in the doubly-linked list of elements.
-	next, prev *snapshotElement
-}
-
-// Initialize the snapshot.
-func (db *DB) initSnapshot() {
-	db.snapsRoot.next = &db.snapsRoot
-	db.snapsRoot.prev = &db.snapsRoot
+	e   *list.Element
 }
 
 // Acquires a snapshot, based on latest sequence.
 func (db *DB) acquireSnapshot() *snapshotElement {
 	db.snapsMu.Lock()
+	defer db.snapsMu.Unlock()
+
 	seq := db.getSeq()
-	elem := db.snapsRoot.prev
-	if elem == &db.snapsRoot || elem.seq != seq {
-		at := db.snapsRoot.prev
-		next := at.next
-		elem = &snapshotElement{
-			seq:  seq,
-			prev: at,
-			next: next,
+
+	if e := db.snapsList.Back(); e != nil {
+		se := e.Value.(*snapshotElement)
+		if se.seq == seq {
+			se.ref++
+			return se
+		} else if seq < se.seq {
+			panic("leveldb: sequence number is not increasing")
 		}
-		at.next = elem
-		next.prev = elem
 	}
-	elem.ref++
-	db.snapsMu.Unlock()
-	return elem
+	se := &snapshotElement{seq: seq, ref: 1}
+	se.e = db.snapsList.PushBack(se)
+	return se
 }
 
 // Releases given snapshot element.
-func (db *DB) releaseSnapshot(elem *snapshotElement) {
-	if !db.isClosed() {
-		db.snapsMu.Lock()
-		elem.ref--
-		if elem.ref == 0 {
-			elem.prev.next = elem.next
-			elem.next.prev = elem.prev
-			elem.next = nil
-			elem.prev = nil
-		} else if elem.ref < 0 {
-			panic("leveldb: Snapshot: negative element reference")
-		}
-		db.snapsMu.Unlock()
+func (db *DB) releaseSnapshot(se *snapshotElement) {
+	db.snapsMu.Lock()
+	defer db.snapsMu.Unlock()
+
+	se.ref--
+	if se.ref == 0 {
+		db.snapsList.Remove(se.e)
+		se.e = nil
+	} else if se.ref < 0 {
+		panic("leveldb: Snapshot: negative element reference")
 	}
 }
 
@@ -71,10 +63,11 @@ func (db *DB) releaseSnapshot(elem *snapshotElement) {
 func (db *DB) minSeq() uint64 {
 	db.snapsMu.Lock()
 	defer db.snapsMu.Unlock()
-	elem := db.snapsRoot.next
-	if elem != &db.snapsRoot {
-		return elem.seq
+
+	if e := db.snapsList.Front(); e != nil {
+		return e.Value.(*snapshotElement).seq
 	}
+
 	return db.getSeq()
 }
 
@@ -97,8 +90,12 @@ func (db *DB) newSnapshot() *Snapshot {
 	return snap
 }
 
+func (snap *Snapshot) String() string {
+	return fmt.Sprintf("leveldb.Snapshot{%d}", snap.elem.seq)
+}
+
 // Get gets the value for the given key. It returns ErrNotFound if
-// the DB does not contain the key.
+// the DB does not contains the key.
 //
 // The caller should not modify the contents of the returned slice, but
 // it is safe to modify the contents of the argument after Get returns.
@@ -116,6 +113,23 @@ func (snap *Snapshot) Get(key []byte, ro *opt.ReadOptions) (value []byte, err er
 	return snap.db.get(key, snap.elem.seq, ro)
 }
 
+// Has returns true if the DB does contains the given key.
+//
+// It is safe to modify the contents of the argument after Get returns.
+func (snap *Snapshot) Has(key []byte, ro *opt.ReadOptions) (ret bool, err error) {
+	err = snap.db.ok()
+	if err != nil {
+		return
+	}
+	snap.mu.RLock()
+	defer snap.mu.RUnlock()
+	if snap.released {
+		err = ErrSnapshotReleased
+		return
+	}
+	return snap.db.has(key, snap.elem.seq, ro)
+}
+
 // NewIterator returns an iterator for the snapshot of the uderlying DB.
 // The returned iterator is not goroutine-safe, but it is safe to use
 // multiple iterators concurrently, with each in a dedicated goroutine.

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/db_util.go

@@ -7,8 +7,7 @@
 package leveldb
 
 import (
-	"errors"
-
+	"github.com/syndtr/goleveldb/leveldb/errors"
 	"github.com/syndtr/goleveldb/leveldb/iterator"
 	"github.com/syndtr/goleveldb/leveldb/opt"
 	"github.com/syndtr/goleveldb/leveldb/storage"
@@ -38,7 +37,9 @@ func (db *DB) logf(format string, v ...interface{}) { db.s.logf(format, v...) }
 
 // Check and clean files.
 func (db *DB) checkAndCleanFiles() error {
-	v := db.s.version_NB()
+	v := db.s.version()
+	defer v.release()
+
 	tablesMap := make(map[uint64]bool)
 	for _, tables := range v.tables {
 		for _, t := range tables {
@@ -78,12 +79,14 @@ func (db *DB) checkAndCleanFiles() error {
 	}
 
 	if nTables != len(tablesMap) {
+		var missing []*storage.FileInfo
 		for num, present := range tablesMap {
 			if !present {
+				missing = append(missing, &storage.FileInfo{Type: storage.TypeTable, Num: num})
 				db.logf("db@janitor table missing @%d", num)
 			}
 		}
-		return ErrCorrupted{Type: MissingFiles, Err: errors.New("leveldb: table files missing")}
+		return errors.NewErrCorrupted(nil, &errors.ErrMissingFiles{Files: missing})
 	}
 
 	db.logf("db@janitor F·%d G·%d", len(files), len(rem))

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/db_write.go

@@ -59,7 +59,7 @@ func (db *DB) rotateMem(n int) (mem *memDB, err error) {
 	}
 
 	// Schedule memdb compaction.
-	db.compTrigger(db.mcompTriggerC)
+	db.compSendTrigger(db.mcompCmdC)
 	return
 }
 
@@ -77,12 +77,12 @@ func (db *DB) flush(n int) (mem *memDB, nn int, err error) {
 		}()
 		nn = mem.mdb.Free()
 		switch {
-		case v.tLen(0) >= kL0_SlowdownWritesTrigger && !delayed:
+		case v.tLen(0) >= db.s.o.GetWriteL0SlowdownTrigger() && !delayed:
 			delayed = true
 			time.Sleep(time.Millisecond)
 		case nn >= n:
 			return false
-		case v.tLen(0) >= kL0_StopWritesTrigger:
+		case v.tLen(0) >= db.s.o.GetWriteL0PauseTrigger():
 			delayed = true
 			err = db.compSendIdle(db.tcompCmdC)
 			if err != nil {
@@ -109,7 +109,12 @@ func (db *DB) flush(n int) (mem *memDB, nn int, err error) {
 	for flush() {
 	}
 	if delayed {
-		db.logf("db@write delayed T·%v", time.Since(start))
+		db.writeDelay += time.Since(start)
+		db.writeDelayN++
+	} else if db.writeDelayN > 0 {
+		db.logf("db@write was delayed N·%d T·%v", db.writeDelayN, db.writeDelay)
+		db.writeDelay = 0
+		db.writeDelayN = 0
 	}
 	return
 }
@@ -120,28 +125,33 @@ func (db *DB) flush(n int) (mem *memDB, nn int, err error) {
 // It is safe to modify the contents of the arguments after Write returns.
 func (db *DB) Write(b *Batch, wo *opt.WriteOptions) (err error) {
 	err = db.ok()
-	if err != nil || b == nil || b.len() == 0 {
+	if err != nil || b == nil || b.Len() == 0 {
 		return
 	}
 
 	b.init(wo.GetSync())
 
 	// The write happen synchronously.
-retry:
 	select {
 	case db.writeC <- b:
 		if <-db.writeMergedC {
 			return <-db.writeAckC
 		}
-		goto retry
 	case db.writeLockC <- struct{}{}:
+	case err = <-db.compPerErrC:
+		return
 	case _, _ = <-db.closeC:
 		return ErrClosed
 	}
 
 	merged := 0
+	danglingMerge := false
 	defer func() {
-		<-db.writeLockC
+		if danglingMerge {
+			db.writeMergedC <- false
+		} else {
+			<-db.writeLockC
+		}
 		for i := 0; i < merged; i++ {
 			db.writeAckC <- err
 		}
@@ -170,7 +180,7 @@ drain:
 				db.writeMergedC <- true
 				merged++
 			} else {
-				db.writeMergedC <- false
+				danglingMerge = true
 				break drain
 			}
 		default:
@@ -185,35 +195,43 @@ drain:
 	if b.size() >= (128 << 10) {
 		// Push the write batch to the journal writer
 		select {
+		case db.journalC <- b:
+			// Write into memdb
+			if berr := b.memReplay(mem.mdb); berr != nil {
+				panic(berr)
+			}
+		case err = <-db.compPerErrC:
+			return
 		case _, _ = <-db.closeC:
 			err = ErrClosed
 			return
-		case db.journalC <- b:
-			// Write into memdb
-			b.memReplay(mem.mdb)
 		}
 		// Wait for journal writer
 		select {
-		case _, _ = <-db.closeC:
-			err = ErrClosed
-			return
 		case err = <-db.journalAckC:
 			if err != nil {
 				// Revert memdb if error detected
-				b.revertMemReplay(mem.mdb)
+				if berr := b.revertMemReplay(mem.mdb); berr != nil {
+					panic(berr)
+				}
 				return
 			}
+		case _, _ = <-db.closeC:
+			err = ErrClosed
+			return
 		}
 	} else {
 		err = db.writeJournal(b)
 		if err != nil {
 			return
 		}
-		b.memReplay(mem.mdb)
+		if berr := b.memReplay(mem.mdb); berr != nil {
+			panic(berr)
+		}
 	}
 
 	// Set last seq number.
-	db.addSeq(uint64(b.len()))
+	db.addSeq(uint64(b.Len()))
 
 	if b.size() >= memFree {
 		db.rotateMem(0)
@@ -262,8 +280,11 @@ func (db *DB) CompactRange(r util.Range) error {
 		return err
 	}
 
+	// Lock writer.
 	select {
 	case db.writeLockC <- struct{}{}:
+	case err := <-db.compPerErrC:
+		return err
 	case _, _ = <-db.closeC:
 		return ErrClosed
 	}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/errors.go

@@ -7,32 +7,12 @@
 package leveldb
 
 import (
-	"errors"
-
-	"github.com/syndtr/goleveldb/leveldb/util"
+	"github.com/syndtr/goleveldb/leveldb/errors"
 )
 
 var (
-	ErrNotFound         = util.ErrNotFound
+	ErrNotFound         = errors.ErrNotFound
 	ErrSnapshotReleased = errors.New("leveldb: snapshot released")
 	ErrIterReleased     = errors.New("leveldb: iterator released")
 	ErrClosed           = errors.New("leveldb: closed")
 )
-
-type CorruptionType int
-
-const (
-	CorruptedManifest CorruptionType = iota
-	MissingFiles
-)
-
-// ErrCorrupted is the type that wraps errors that indicate corruption in
-// the database.
-type ErrCorrupted struct {
-	Type CorruptionType
-	Err  error
-}
-
-func (e ErrCorrupted) Error() string {
-	return e.Err.Error()
-}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/errors/errors.go

@@ -0,0 +1,76 @@
+// Copyright (c) 2014, Suryandaru Triandana <syndtr@gmail.com>
+// All rights reserved.
+//
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+// Package errors provides common error types used throughout leveldb.
+package errors
+
+import (
+	"errors"
+	"fmt"
+
+	"github.com/syndtr/goleveldb/leveldb/storage"
+	"github.com/syndtr/goleveldb/leveldb/util"
+)
+
+var (
+	ErrNotFound    = New("leveldb: not found")
+	ErrReleased    = util.ErrReleased
+	ErrHasReleaser = util.ErrHasReleaser
+)
+
+// New returns an error that formats as the given text.
+func New(text string) error {
+	return errors.New(text)
+}
+
+// ErrCorrupted is the type that wraps errors that indicate corruption in
+// the database.
+type ErrCorrupted struct {
+	File *storage.FileInfo
+	Err  error
+}
+
+func (e *ErrCorrupted) Error() string {
+	if e.File != nil {
+		return fmt.Sprintf("%v [file=%v]", e.Err, e.File)
+	} else {
+		return e.Err.Error()
+	}
+}
+
+// NewErrCorrupted creates new ErrCorrupted error.
+func NewErrCorrupted(f storage.File, err error) error {
+	return &ErrCorrupted{storage.NewFileInfo(f), err}
+}
+
+// IsCorrupted returns a boolean indicating whether the error is indicating
+// a corruption.
+func IsCorrupted(err error) bool {
+	switch err.(type) {
+	case *ErrCorrupted:
+		return true
+	}
+	return false
+}
+
+// ErrMissingFiles is the type that indicating a corruption due to missing
+// files.
+type ErrMissingFiles struct {
+	Files []*storage.FileInfo
+}
+
+func (e *ErrMissingFiles) Error() string { return "file missing" }
+
+// SetFile sets 'file info' of the given error with the given file.
+// Currently only ErrCorrupted is supported, otherwise will do nothing.
+func SetFile(err error, f storage.File) error {
+	switch x := err.(type) {
+	case *ErrCorrupted:
+		x.File = storage.NewFileInfo(f)
+		return x
+	}
+	return err
+}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/external_test.go

@@ -17,13 +17,14 @@ import (
 var _ = testutil.Defer(func() {
 	Describe("Leveldb external", func() {
 		o := &opt.Options{
-			BlockCache:           opt.NoCache,
-			BlockRestartInterval: 5,
-			BlockSize:            50,
-			Compression:          opt.NoCompression,
-			CachedOpenFiles:      -1,
-			Strict:               opt.StrictAll,
-			WriteBuffer:          1000,
+			DisableBlockCache:      true,
+			BlockRestartInterval:   5,
+			BlockSize:              80,
+			Compression:            opt.NoCompression,
+			OpenFilesCacheCapacity: -1,
+			Strict:                 opt.StrictAll,
+			WriteBuffer:            1000,
+			CompactionTableSize:    2000,
 		}
 
 		Describe("write test", func() {

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/iterator/indexed_iter.go

@@ -7,6 +7,7 @@
 package iterator
 
 import (
+	"github.com/syndtr/goleveldb/leveldb/errors"
 	"github.com/syndtr/goleveldb/leveldb/util"
 )
 
@@ -22,9 +23,8 @@ type IteratorIndexer interface {
 
 type indexedIterator struct {
 	util.BasicReleaser
-	index     IteratorIndexer
-	strict    bool
-	strictGet bool
+	index  IteratorIndexer
+	strict bool
 
 	data   Iterator
 	err    error
@@ -37,11 +37,6 @@ func (i *indexedIterator) setData() {
 		i.data.Release()
 	}
 	i.data = i.index.Get()
-	if i.strictGet {
-		if err := i.data.Error(); err != nil {
-			i.err = err
-		}
-	}
 }
 
 func (i *indexedIterator) clearData() {
@@ -61,13 +56,11 @@ func (i *indexedIterator) indexErr() {
 }
 
 func (i *indexedIterator) dataErr() bool {
-	if i.errf != nil {
-		if err := i.data.Error(); err != nil {
+	if err := i.data.Error(); err != nil {
+		if i.errf != nil {
 			i.errf(err)
 		}
-	}
-	if i.strict {
-		if err := i.data.Error(); err != nil {
+		if i.strict || !errors.IsCorrupted(err) {
 			i.err = err
 			return true
 		}
@@ -236,16 +229,14 @@ func (i *indexedIterator) SetErrorCallback(f func(err error)) {
 	i.errf = f
 }
 
-// NewIndexedIterator returns an indexed iterator. An index is iterator
-// that returns another iterator, a data iterator. A data iterator is the
+// NewIndexedIterator returns an 'indexed iterator'. An index is iterator
+// that returns another iterator, a 'data iterator'. A 'data iterator' is the
 // iterator that contains actual key/value pairs.
 //
-// If strict is true then error yield by data iterator will halt the indexed
-// iterator, on contrary if strict is false then the indexed iterator will
-// ignore those error and move on to the next index. If strictGet is true and
-// index.Get() yield an 'error iterator' then the indexed iterator will be halted.
-// An 'error iterator' is iterator which its Error() method always return non-nil
-// even before any 'seeks method' is called.
-func NewIndexedIterator(index IteratorIndexer, strict, strictGet bool) Iterator {
-	return &indexedIterator{index: index, strict: strict, strictGet: strictGet}
+// If strict is true the any 'corruption errors' (i.e errors.IsCorrupted(err) == true)
+// won't be ignored and will halt 'indexed iterator', otherwise the iterator will
+// continue to the next 'data iterator'. Corruption on 'index iterator' will not be
+// ignored and will halt the iterator.
+func NewIndexedIterator(index IteratorIndexer, strict bool) Iterator {
+	return &indexedIterator{index: index, strict: strict}
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/iterator/indexed_iter_test.go

@@ -65,7 +65,7 @@ var _ = testutil.Defer(func() {
 					// Test the iterator.
 					t := testutil.IteratorTesting{
 						KeyValue: kv.Clone(),
-						Iter:     NewIndexedIterator(NewArrayIndexer(index), true, true),
+						Iter:     NewIndexedIterator(NewArrayIndexer(index), true),
 					}
 					testutil.DoIteratorTesting(&t)
 					done <- true

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/iterator/iter_suite_test.go

@@ -3,15 +3,9 @@ package iterator_test
 import (
 	"testing"
 
-	. "github.com/onsi/ginkgo"
-	. "github.com/onsi/gomega"
-
 	"github.com/syndtr/goleveldb/leveldb/testutil"
 )
 
 func TestIterator(t *testing.T) {
-	testutil.RunDefer()
-
-	RegisterFailHandler(Fail)
-	RunSpecs(t, "Iterator Suite")
+	testutil.RunSuite(t, "Iterator Suite")
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/iterator/merged_iter.go

@@ -8,6 +8,7 @@ package iterator
 
 import (
 	"github.com/syndtr/goleveldb/leveldb/comparer"
+	"github.com/syndtr/goleveldb/leveldb/errors"
 	"github.com/syndtr/goleveldb/leveldb/util"
 )
 
@@ -42,13 +43,11 @@ func assertKey(key []byte) []byte {
 }
 
 func (i *mergedIterator) iterErr(iter Iterator) bool {
-	if i.errf != nil {
-		if err := iter.Error(); err != nil {
+	if err := iter.Error(); err != nil {
+		if i.errf != nil {
 			i.errf(err)
 		}
-	}
-	if i.strict {
-		if err := iter.Error(); err != nil {
+		if i.strict || !errors.IsCorrupted(err) {
 			i.err = err
 			return true
 		}
@@ -292,9 +291,9 @@ func (i *mergedIterator) SetErrorCallback(f func(err error)) {
 // keys: if iters[i] contains a key k then iters[j] will not contain that key k.
 // None of the iters may be nil.
 //
-// If strict is true then error yield by any iterators will halt the merged
-// iterator, on contrary if strict is false then the merged iterator will
-// ignore those error and move on to the next iterator.
+// If strict is true the any 'corruption errors' (i.e errors.IsCorrupted(err) == true)
+// won't be ignored and will halt 'merged iterator', otherwise the iterator will
+// continue to the next 'input iterator'.
 func NewMergedIterator(iters []Iterator, cmp comparer.Comparer, strict bool) Iterator {
 	return &mergedIterator{
 		iters:  iters,

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/journal/journal.go

@@ -79,10 +79,10 @@ package journal
 
 import (
 	"encoding/binary"
-	"errors"
 	"fmt"
 	"io"
 
+	"github.com/syndtr/goleveldb/leveldb/errors"
 	"github.com/syndtr/goleveldb/leveldb/util"
 )
 
@@ -109,7 +109,7 @@ type ErrCorrupted struct {
 	Reason string
 }
 
-func (e ErrCorrupted) Error() string {
+func (e *ErrCorrupted) Error() string {
 	return fmt.Sprintf("leveldb/journal: block/chunk corrupted: %s (%d bytes)", e.Reason, e.Size)
 }
 
@@ -162,10 +162,10 @@ var errSkip = errors.New("leveldb/journal: skipped")
 
 func (r *Reader) corrupt(n int, reason string, skip bool) error {
 	if r.dropper != nil {
-		r.dropper.Drop(ErrCorrupted{n, reason})
+		r.dropper.Drop(&ErrCorrupted{n, reason})
 	}
 	if r.strict && !skip {
-		r.err = ErrCorrupted{n, reason}
+		r.err = errors.NewErrCorrupted(nil, &ErrCorrupted{n, reason})
 		return r.err
 	}
 	return errSkip

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/key.go

@@ -9,15 +9,30 @@ package leveldb
 import (
 	"encoding/binary"
 	"fmt"
+
+	"github.com/syndtr/goleveldb/leveldb/errors"
 )
 
-type vType int
+type ErrIkeyCorrupted struct {
+	Ikey   []byte
+	Reason string
+}
 
-func (t vType) String() string {
-	switch t {
-	case tDel:
+func (e *ErrIkeyCorrupted) Error() string {
+	return fmt.Sprintf("leveldb: iKey %q corrupted: %s", e.Ikey, e.Reason)
+}
+
+func newErrIkeyCorrupted(ikey []byte, reason string) error {
+	return errors.NewErrCorrupted(nil, &ErrIkeyCorrupted{append([]byte{}, ikey...), reason})
+}
+
+type kType int
+
+func (kt kType) String() string {
+	switch kt {
+	case ktDel:
 		return "d"
-	case tVal:
+	case ktVal:
 		return "v"
 	}
 	return "x"
@@ -26,16 +41,16 @@ func (t vType) String() string {
 // Value types encoded as the last component of internal keys.
 // Don't modify; this value are saved to disk.
 const (
-	tDel vType = iota
-	tVal
+	ktDel kType = iota
+	ktVal
 )
 
-// tSeek defines the vType that should be passed when constructing an
+// ktSeek defines the kType that should be passed when constructing an
 // internal key for seeking to a particular sequence number (since we
 // sort sequence numbers in decreasing order and the value type is
 // embedded as the low 8 bits in the sequence number in internal keys,
 // we need to use the highest-numbered ValueType, not the lowest).
-const tSeek = tVal
+const ktSeek = ktVal
 
 const (
 	// Maximum value possible for sequence number; the 8-bits are
@@ -43,7 +58,7 @@ const (
 	// 64-bit integer.
 	kMaxSeq uint64 = (uint64(1) << 56) - 1
 	// Maximum value possible for packed sequence number and type.
-	kMaxNum uint64 = (kMaxSeq << 8) | uint64(tSeek)
+	kMaxNum uint64 = (kMaxSeq << 8) | uint64(ktSeek)
 )
 
 // Maximum number encoded in bytes.
@@ -55,85 +70,73 @@ func init() {
 
 type iKey []byte
 
-func newIKey(ukey []byte, seq uint64, t vType) iKey {
-	if seq > kMaxSeq || t > tVal {
-		panic("invalid seq number or value type")
+func newIkey(ukey []byte, seq uint64, kt kType) iKey {
+	if seq > kMaxSeq {
+		panic("leveldb: invalid sequence number")
+	} else if kt > ktVal {
+		panic("leveldb: invalid type")
 	}
 
-	b := make(iKey, len(ukey)+8)
-	copy(b, ukey)
-	binary.LittleEndian.PutUint64(b[len(ukey):], (seq<<8)|uint64(t))
-	return b
+	ik := make(iKey, len(ukey)+8)
+	copy(ik, ukey)
+	binary.LittleEndian.PutUint64(ik[len(ukey):], (seq<<8)|uint64(kt))
+	return ik
 }
 
-func parseIkey(p []byte) (ukey []byte, seq uint64, t vType, ok bool) {
-	if len(p) < 8 {
-		return
+func parseIkey(ik []byte) (ukey []byte, seq uint64, kt kType, err error) {
+	if len(ik) < 8 {
+		return nil, 0, 0, newErrIkeyCorrupted(ik, "invalid length")
 	}
-	num := binary.LittleEndian.Uint64(p[len(p)-8:])
-	seq, t = uint64(num>>8), vType(num&0xff)
-	if t > tVal {
-		return
+	num := binary.LittleEndian.Uint64(ik[len(ik)-8:])
+	seq, kt = uint64(num>>8), kType(num&0xff)
+	if kt > ktVal {
+		return nil, 0, 0, newErrIkeyCorrupted(ik, "invalid type")
 	}
-	ukey = p[:len(p)-8]
-	ok = true
+	ukey = ik[:len(ik)-8]
 	return
 }
 
-func validIkey(p []byte) bool {
-	_, _, _, ok := parseIkey(p)
-	return ok
+func validIkey(ik []byte) bool {
+	_, _, _, err := parseIkey(ik)
+	return err == nil
 }
 
-func (p iKey) assert() {
-	if p == nil {
-		panic("nil iKey")
+func (ik iKey) assert() {
+	if ik == nil {
+		panic("leveldb: nil iKey")
 	}
-	if len(p) < 8 {
-		panic(fmt.Sprintf("invalid iKey %q, len=%d", []byte(p), len(p)))
+	if len(ik) < 8 {
+		panic(fmt.Sprintf("leveldb: iKey %q, len=%d: invalid length", []byte(ik), len(ik)))
 	}
 }
 
-func (p iKey) ok() bool {
-	if len(p) < 8 {
-		return false
-	}
-	_, _, ok := p.parseNum()
-	return ok
+func (ik iKey) ukey() []byte {
+	ik.assert()
+	return ik[:len(ik)-8]
 }
 
-func (p iKey) ukey() []byte {
-	p.assert()
-	return p[:len(p)-8]
+func (ik iKey) num() uint64 {
+	ik.assert()
+	return binary.LittleEndian.Uint64(ik[len(ik)-8:])
 }
 
-func (p iKey) num() uint64 {
-	p.assert()
-	return binary.LittleEndian.Uint64(p[len(p)-8:])
-}
-
-func (p iKey) parseNum() (seq uint64, t vType, ok bool) {
-	if p == nil {
-		panic("nil iKey")
+func (ik iKey) parseNum() (seq uint64, kt kType) {
+	num := ik.num()
+	seq, kt = uint64(num>>8), kType(num&0xff)
+	if kt > ktVal {
+		panic(fmt.Sprintf("leveldb: iKey %q, len=%d: invalid type %#x", []byte(ik), len(ik), kt))
 	}
-	if len(p) < 8 {
-		return
-	}
-	num := p.num()
-	seq, t = uint64(num>>8), vType(num&0xff)
-	if t > tVal {
-		return 0, 0, false
-	}
-	ok = true
 	return
 }
 
-func (p iKey) String() string {
-	if len(p) == 0 {
+func (ik iKey) String() string {
+	if ik == nil {
 		return "<nil>"
 	}
-	if seq, t, ok := p.parseNum(); ok {
-		return fmt.Sprintf("%s,%s%d", shorten(string(p.ukey())), t, seq)
+
+	if ukey, seq, kt, err := parseIkey(ik); err == nil {
+		return fmt.Sprintf("%s,%s%d", shorten(string(ukey)), kt, seq)
+	} else {
+		return "<invalid>"
 	}
-	return "<invalid>"
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/key_test.go

@@ -15,8 +15,8 @@ import (
 
 var defaultIComparer = &iComparer{comparer.DefaultComparer}
 
-func ikey(key string, seq uint64, t vType) iKey {
-	return newIKey([]byte(key), uint64(seq), t)
+func ikey(key string, seq uint64, kt kType) iKey {
+	return newIkey([]byte(key), uint64(seq), kt)
 }
 
 func shortSep(a, b []byte) []byte {
@@ -37,27 +37,37 @@ func shortSuccessor(b []byte) []byte {
 	return dst
 }
 
-func testSingleKey(t *testing.T, key string, seq uint64, vt vType) {
-	ik := ikey(key, seq, vt)
+func testSingleKey(t *testing.T, key string, seq uint64, kt kType) {
+	ik := ikey(key, seq, kt)
 
 	if !bytes.Equal(ik.ukey(), []byte(key)) {
 		t.Errorf("user key does not equal, got %v, want %v", string(ik.ukey()), key)
 	}
 
-	if rseq, rt, ok := ik.parseNum(); ok {
+	rseq, rt := ik.parseNum()
+	if rseq != seq {
+		t.Errorf("seq number does not equal, got %v, want %v", rseq, seq)
+	}
+	if rt != kt {
+		t.Errorf("type does not equal, got %v, want %v", rt, kt)
+	}
+
+	if rukey, rseq, rt, kerr := parseIkey(ik); kerr == nil {
+		if !bytes.Equal(rukey, []byte(key)) {
+			t.Errorf("user key does not equal, got %v, want %v", string(ik.ukey()), key)
+		}
 		if rseq != seq {
 			t.Errorf("seq number does not equal, got %v, want %v", rseq, seq)
 		}
-
-		if rt != vt {
-			t.Errorf("type does not equal, got %v, want %v", rt, vt)
+		if rt != kt {
+			t.Errorf("type does not equal, got %v, want %v", rt, kt)
 		}
 	} else {
-		t.Error("cannot parse seq and type")
+		t.Errorf("key error: %v", kerr)
 	}
 }
 
-func TestIKey_EncodeDecode(t *testing.T) {
+func TestIkey_EncodeDecode(t *testing.T) {
 	keys := []string{"", "k", "hello", "longggggggggggggggggggggg"}
 	seqs := []uint64{
 		1, 2, 3,
@@ -67,8 +77,8 @@ func TestIKey_EncodeDecode(t *testing.T) {
 	}
 	for _, key := range keys {
 		for _, seq := range seqs {
-			testSingleKey(t, key, seq, tVal)
-			testSingleKey(t, "hello", 1, tDel)
+			testSingleKey(t, key, seq, ktVal)
+			testSingleKey(t, "hello", 1, ktDel)
 		}
 	}
 }
@@ -79,45 +89,45 @@ func assertBytes(t *testing.T, want, got []byte) {
 	}
 }
 
-func TestIKeyShortSeparator(t *testing.T) {
+func TestIkeyShortSeparator(t *testing.T) {
 	// When user keys are same
-	assertBytes(t, ikey("foo", 100, tVal),
-		shortSep(ikey("foo", 100, tVal),
-			ikey("foo", 99, tVal)))
-	assertBytes(t, ikey("foo", 100, tVal),
-		shortSep(ikey("foo", 100, tVal),
-			ikey("foo", 101, tVal)))
-	assertBytes(t, ikey("foo", 100, tVal),
-		shortSep(ikey("foo", 100, tVal),
-			ikey("foo", 100, tVal)))
-	assertBytes(t, ikey("foo", 100, tVal),
-		shortSep(ikey("foo", 100, tVal),
-			ikey("foo", 100, tDel)))
+	assertBytes(t, ikey("foo", 100, ktVal),
+		shortSep(ikey("foo", 100, ktVal),
+			ikey("foo", 99, ktVal)))
+	assertBytes(t, ikey("foo", 100, ktVal),
+		shortSep(ikey("foo", 100, ktVal),
+			ikey("foo", 101, ktVal)))
+	assertBytes(t, ikey("foo", 100, ktVal),
+		shortSep(ikey("foo", 100, ktVal),
+			ikey("foo", 100, ktVal)))
+	assertBytes(t, ikey("foo", 100, ktVal),
+		shortSep(ikey("foo", 100, ktVal),
+			ikey("foo", 100, ktDel)))
 
 	// When user keys are misordered
-	assertBytes(t, ikey("foo", 100, tVal),
-		shortSep(ikey("foo", 100, tVal),
-			ikey("bar", 99, tVal)))
+	assertBytes(t, ikey("foo", 100, ktVal),
+		shortSep(ikey("foo", 100, ktVal),
+			ikey("bar", 99, ktVal)))
 
 	// When user keys are different, but correctly ordered
-	assertBytes(t, ikey("g", uint64(kMaxSeq), tSeek),
-		shortSep(ikey("foo", 100, tVal),
-			ikey("hello", 200, tVal)))
+	assertBytes(t, ikey("g", uint64(kMaxSeq), ktSeek),
+		shortSep(ikey("foo", 100, ktVal),
+			ikey("hello", 200, ktVal)))
 
 	// When start user key is prefix of limit user key
-	assertBytes(t, ikey("foo", 100, tVal),
-		shortSep(ikey("foo", 100, tVal),
-			ikey("foobar", 200, tVal)))
+	assertBytes(t, ikey("foo", 100, ktVal),
+		shortSep(ikey("foo", 100, ktVal),
+			ikey("foobar", 200, ktVal)))
 
 	// When limit user key is prefix of start user key
-	assertBytes(t, ikey("foobar", 100, tVal),
-		shortSep(ikey("foobar", 100, tVal),
-			ikey("foo", 200, tVal)))
+	assertBytes(t, ikey("foobar", 100, ktVal),
+		shortSep(ikey("foobar", 100, ktVal),
+			ikey("foo", 200, ktVal)))
 }
 
-func TestIKeyShortestSuccessor(t *testing.T) {
-	assertBytes(t, ikey("g", uint64(kMaxSeq), tSeek),
-		shortSuccessor(ikey("foo", 100, tVal)))
-	assertBytes(t, ikey("\xff\xff", 100, tVal),
-		shortSuccessor(ikey("\xff\xff", 100, tVal)))
+func TestIkeyShortestSuccessor(t *testing.T) {
+	assertBytes(t, ikey("g", uint64(kMaxSeq), ktSeek),
+		shortSuccessor(ikey("foo", 100, ktVal)))
+	assertBytes(t, ikey("\xff\xff", 100, ktVal),
+		shortSuccessor(ikey("\xff\xff", 100, ktVal)))
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/leveldb_suite_test.go

@@ -3,18 +3,9 @@ package leveldb
 import (
 	"testing"
 
-	. "github.com/onsi/ginkgo"
-	. "github.com/onsi/gomega"
-
 	"github.com/syndtr/goleveldb/leveldb/testutil"
 )
 
-func TestLeveldb(t *testing.T) {
-	testutil.RunDefer()
-
-	RegisterFailHandler(Fail)
-	RunSpecs(t, "Leveldb Suite")
-
-	RegisterTestingT(t)
-	testutil.RunDefer("teardown")
+func TestLevelDB(t *testing.T) {
+	testutil.RunSuite(t, "LevelDB Suite")
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/memdb/memdb.go

@@ -8,17 +8,17 @@
 package memdb
 
 import (
-	"errors"
 	"math/rand"
 	"sync"
 
 	"github.com/syndtr/goleveldb/leveldb/comparer"
+	"github.com/syndtr/goleveldb/leveldb/errors"
 	"github.com/syndtr/goleveldb/leveldb/iterator"
 	"github.com/syndtr/goleveldb/leveldb/util"
 )
 
 var (
-	ErrNotFound     = util.ErrNotFound
+	ErrNotFound     = errors.ErrNotFound
 	ErrIterReleased = errors.New("leveldb/memdb: iterator released")
 )
 

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/memdb/memdb_suite_test.go

@@ -3,15 +3,9 @@ package memdb
 import (
 	"testing"
 
-	. "github.com/onsi/ginkgo"
-	. "github.com/onsi/gomega"
-
 	"github.com/syndtr/goleveldb/leveldb/testutil"
 )
 
-func TestMemdb(t *testing.T) {
-	testutil.RunDefer()
-
-	RegisterFailHandler(Fail)
-	RunSpecs(t, "Memdb Suite")
+func TestMemDB(t *testing.T) {
+	testutil.RunSuite(t, "MemDB Suite")
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/opt/options.go

@@ -11,6 +11,7 @@ import (
 	"github.com/syndtr/goleveldb/leveldb/cache"
 	"github.com/syndtr/goleveldb/leveldb/comparer"
 	"github.com/syndtr/goleveldb/leveldb/filter"
+	"math"
 )
 
 const (
@@ -19,31 +20,56 @@ const (
 	GiB = MiB * 1024
 )
 
-const (
-	DefaultBlockCacheSize       = 8 * MiB
-	DefaultBlockRestartInterval = 16
-	DefaultBlockSize            = 4 * KiB
-	DefaultCompressionType      = SnappyCompression
-	DefaultCachedOpenFiles      = 500
-	DefaultWriteBuffer          = 4 * MiB
+var (
+	DefaultBlockCacher                   = LRUCacher
+	DefaultBlockCacheCapacity            = 8 * MiB
+	DefaultBlockRestartInterval          = 16
+	DefaultBlockSize                     = 4 * KiB
+	DefaultCompactionExpandLimitFactor   = 25
+	DefaultCompactionGPOverlapsFactor    = 10
+	DefaultCompactionL0Trigger           = 4
+	DefaultCompactionSourceLimitFactor   = 1
+	DefaultCompactionTableSize           = 2 * MiB
+	DefaultCompactionTableSizeMultiplier = 1.0
+	DefaultCompactionTotalSize           = 10 * MiB
+	DefaultCompactionTotalSizeMultiplier = 10.0
+	DefaultCompressionType               = SnappyCompression
+	DefaultOpenFilesCacher               = LRUCacher
+	DefaultOpenFilesCacheCapacity        = 500
+	DefaultMaxMemCompationLevel          = 2
+	DefaultNumLevel                      = 7
+	DefaultWriteBuffer                   = 4 * MiB
+	DefaultWriteL0PauseTrigger           = 12
+	DefaultWriteL0SlowdownTrigger        = 8
 )
 
-type noCache struct{}
+// Cacher is a caching algorithm.
+type Cacher interface {
+	New(capacity int) cache.Cacher
+}
 
-func (noCache) SetCapacity(capacity int)                     {}
-func (noCache) Capacity() int                                { return 0 }
-func (noCache) Used() int                                    { return 0 }
-func (noCache) Size() int                                    { return 0 }
-func (noCache) NumObjects() int                              { return 0 }
-func (noCache) GetNamespace(id uint64) cache.Namespace       { return nil }
-func (noCache) PurgeNamespace(id uint64, fin cache.PurgeFin) {}
-func (noCache) ZapNamespace(id uint64)                       {}
-func (noCache) Purge(fin cache.PurgeFin)                     {}
-func (noCache) Zap()                                         {}
+type CacherFunc struct {
+	NewFunc func(capacity int) cache.Cacher
+}
 
-var NoCache cache.Cache = noCache{}
+func (f *CacherFunc) New(capacity int) cache.Cacher {
+	if f.NewFunc != nil {
+		return f.NewFunc(capacity)
+	}
+	return nil
+}
 
-// Compression is the per-block compression algorithm to use.
+func noCacher(int) cache.Cacher { return nil }
+
+var (
+	// LRUCacher is the LRU-cache algorithm.
+	LRUCacher = &CacherFunc{cache.NewLRU}
+
+	// NoCacher is the value to disable caching algorithm.
+	NoCacher = &CacherFunc{}
+)
+
+// Compression is the 'sorted table' block compression algorithm to use.
 type Compression uint
 
 func (c Compression) String() string {
@@ -65,34 +91,47 @@ const (
 	nCompression
 )
 
-// Strict is the DB strict level.
+// Strict is the DB 'strict level'.
 type Strict uint
 
 const (
 	// If present then a corrupted or invalid chunk or block in manifest
-	// journal will cause an error istead of being dropped.
+	// journal will cause an error instead of being dropped.
+	// This will prevent database with corrupted manifest to be opened.
 	StrictManifest Strict = 1 << iota
 
-	// If present then a corrupted or invalid chunk or block in journal
-	// will cause an error istead of being dropped.
-	StrictJournal
-
 	// If present then journal chunk checksum will be verified.
 	StrictJournalChecksum
 
-	// If present then an invalid key/value pair will cause an error
-	// instead of being skipped.
-	StrictIterator
+	// If present then a corrupted or invalid chunk or block in journal
+	// will cause an error instead of being dropped.
+	// This will prevent database with corrupted journal to be opened.
+	StrictJournal
 
 	// If present then 'sorted table' block checksum will be verified.
+	// This has effect on both 'read operation' and compaction.
 	StrictBlockChecksum
 
+	// If present then a corrupted 'sorted table' will fails compaction.
+	// The database will enter read-only mode.
+	StrictCompaction
+
+	// If present then a corrupted 'sorted table' will halts 'read operation'.
+	StrictReader
+
+	// If present then leveldb.Recover will drop corrupted 'sorted table'.
+	StrictRecovery
+
+	// This only applicable for ReadOptions, if present then this ReadOptions
+	// 'strict level' will override global ones.
+	StrictOverride
+
 	// StrictAll enables all strict flags.
-	StrictAll = StrictManifest | StrictJournal | StrictJournalChecksum | StrictIterator | StrictBlockChecksum
+	StrictAll = StrictManifest | StrictJournalChecksum | StrictJournal | StrictBlockChecksum | StrictCompaction | StrictReader | StrictRecovery
 
 	// DefaultStrict is the default strict flags. Specify any strict flags
 	// will override default strict flags as whole (i.e. not OR'ed).
-	DefaultStrict = StrictJournalChecksum | StrictBlockChecksum
+	DefaultStrict = StrictJournalChecksum | StrictBlockChecksum | StrictCompaction | StrictReader
 
 	// NoStrict disables all strict flags. Override default strict flags.
 	NoStrict = ^StrictAll
@@ -107,11 +146,17 @@ type Options struct {
 	// The default value is nil
 	AltFilters []filter.Filter
 
-	// BlockCache provides per-block caching for LevelDB. Specify NoCache to
-	// disable block caching.
+	// BlockCacher provides cache algorithm for LevelDB 'sorted table' block caching.
+	// Specify NoCacher to disable caching algorithm.
 	//
-	// By default LevelDB will create LRU-cache with capacity of 8MiB.
-	BlockCache cache.Cache
+	// The default value is LRUCacher.
+	BlockCacher Cacher
+
+	// BlockCacheCapacity defines the capacity of the 'sorted table' block caching.
+	// Use -1 for zero, this has same effect with specifying NoCacher to BlockCacher.
+	//
+	// The default value is 8MiB.
+	BlockCacheCapacity int
 
 	// BlockRestartInterval is the number of keys between restart points for
 	// delta encoding of keys.
@@ -125,12 +170,72 @@ type Options struct {
 	// The default value is 4KiB.
 	BlockSize int
 
-	// CachedOpenFiles defines number of open files to kept around when not
-	// in-use, the counting includes still in-use files.
-	// Set this to negative value to disable caching.
+	// CompactionExpandLimitFactor limits compaction size after expanded.
+	// This will be multiplied by table size limit at compaction target level.
 	//
-	// The default value is 500.
-	CachedOpenFiles int
+	// The default value is 25.
+	CompactionExpandLimitFactor int
+
+	// CompactionGPOverlapsFactor limits overlaps in grandparent (Level + 2) that a
+	// single 'sorted table' generates.
+	// This will be multiplied by table size limit at grandparent level.
+	//
+	// The default value is 10.
+	CompactionGPOverlapsFactor int
+
+	// CompactionL0Trigger defines number of 'sorted table' at level-0 that will
+	// trigger compaction.
+	//
+	// The default value is 4.
+	CompactionL0Trigger int
+
+	// CompactionSourceLimitFactor limits compaction source size. This doesn't apply to
+	// level-0.
+	// This will be multiplied by table size limit at compaction target level.
+	//
+	// The default value is 1.
+	CompactionSourceLimitFactor int
+
+	// CompactionTableSize limits size of 'sorted table' that compaction generates.
+	// The limits for each level will be calculated as:
+	//   CompactionTableSize * (CompactionTableSizeMultiplier ^ Level)
+	// The multiplier for each level can also fine-tuned using CompactionTableSizeMultiplierPerLevel.
+	//
+	// The default value is 2MiB.
+	CompactionTableSize int
+
+	// CompactionTableSizeMultiplier defines multiplier for CompactionTableSize.
+	//
+	// The default value is 1.
+	CompactionTableSizeMultiplier float64
+
+	// CompactionTableSizeMultiplierPerLevel defines per-level multiplier for
+	// CompactionTableSize.
+	// Use zero to skip a level.
+	//
+	// The default value is nil.
+	CompactionTableSizeMultiplierPerLevel []float64
+
+	// CompactionTotalSize limits total size of 'sorted table' for each level.
+	// The limits for each level will be calculated as:
+	//   CompactionTotalSize * (CompactionTotalSizeMultiplier ^ Level)
+	// The multiplier for each level can also fine-tuned using
+	// CompactionTotalSizeMultiplierPerLevel.
+	//
+	// The default value is 10MiB.
+	CompactionTotalSize int
+
+	// CompactionTotalSizeMultiplier defines multiplier for CompactionTotalSize.
+	//
+	// The default value is 10.
+	CompactionTotalSizeMultiplier float64
+
+	// CompactionTotalSizeMultiplierPerLevel defines per-level multiplier for
+	// CompactionTotalSize.
+	// Use zero to skip a level.
+	//
+	// The default value is nil.
+	CompactionTotalSizeMultiplierPerLevel []float64
 
 	// Comparer defines a total ordering over the space of []byte keys: a 'less
 	// than' relationship. The same comparison algorithm must be used for reads
@@ -139,11 +244,22 @@ type Options struct {
 	// The default value uses the same ordering as bytes.Compare.
 	Comparer comparer.Comparer
 
-	// Compression defines the per-block compression to use.
+	// Compression defines the 'sorted table' block compression to use.
 	//
 	// The default value (DefaultCompression) uses snappy compression.
 	Compression Compression
 
+	// DisableBlockCache allows disable use of cache.Cache functionality on
+	// 'sorted table' block.
+	//
+	// The default value is false.
+	DisableBlockCache bool
+
+	// DisableCompactionBackoff allows disable compaction retry backoff.
+	//
+	// The default value is false.
+	DisableCompactionBackoff bool
+
 	// ErrorIfExist defines whether an error should returned if the DB already
 	// exist.
 	//
@@ -172,6 +288,31 @@ type Options struct {
 	// The default value is nil.
 	Filter filter.Filter
 
+	// MaxMemCompationLevel defines maximum level a newly compacted 'memdb'
+	// will be pushed into if doesn't creates overlap. This should less than
+	// NumLevel. Use -1 for level-0.
+	//
+	// The default is 2.
+	MaxMemCompationLevel int
+
+	// NumLevel defines number of database level. The level shouldn't changed
+	// between opens, or the database will panic.
+	//
+	// The default is 7.
+	NumLevel int
+
+	// OpenFilesCacher provides cache algorithm for open files caching.
+	// Specify NoCacher to disable caching algorithm.
+	//
+	// The default value is LRUCacher.
+	OpenFilesCacher Cacher
+
+	// OpenFilesCacheCapacity defines the capacity of the open files caching.
+	// Use -1 for zero, this has same effect with specifying NoCacher to OpenFilesCacher.
+	//
+	// The default value is 500.
+	OpenFilesCacheCapacity int
+
 	// Strict defines the DB strict level.
 	Strict Strict
 
@@ -183,6 +324,18 @@ type Options struct {
 	//
 	// The default value is 4MiB.
 	WriteBuffer int
+
+	// WriteL0StopTrigger defines number of 'sorted table' at level-0 that will
+	// pause write.
+	//
+	// The default value is 12.
+	WriteL0PauseTrigger int
+
+	// WriteL0SlowdownTrigger defines number of 'sorted table' at level-0 that
+	// will trigger write slowdown.
+	//
+	// The default value is 8.
+	WriteL0SlowdownTrigger int
 }
 
 func (o *Options) GetAltFilters() []filter.Filter {
@@ -192,11 +345,22 @@ func (o *Options) GetAltFilters() []filter.Filter {
 	return o.AltFilters
 }
 
-func (o *Options) GetBlockCache() cache.Cache {
-	if o == nil {
+func (o *Options) GetBlockCacher() Cacher {
+	if o == nil || o.BlockCacher == nil {
+		return DefaultBlockCacher
+	} else if o.BlockCacher == NoCacher {
 		return nil
 	}
-	return o.BlockCache
+	return o.BlockCacher
+}
+
+func (o *Options) GetBlockCacheCapacity() int {
+	if o == nil || o.BlockCacheCapacity <= 0 {
+		return DefaultBlockCacheCapacity
+	} else if o.BlockCacheCapacity == -1 {
+		return 0
+	}
+	return o.BlockCacheCapacity
 }
 
 func (o *Options) GetBlockRestartInterval() int {
@@ -213,13 +377,77 @@ func (o *Options) GetBlockSize() int {
 	return o.BlockSize
 }
 
-func (o *Options) GetCachedOpenFiles() int {
-	if o == nil || o.CachedOpenFiles == 0 {
-		return DefaultCachedOpenFiles
-	} else if o.CachedOpenFiles < 0 {
-		return 0
+func (o *Options) GetCompactionExpandLimit(level int) int {
+	factor := DefaultCompactionExpandLimitFactor
+	if o != nil && o.CompactionExpandLimitFactor > 0 {
+		factor = o.CompactionExpandLimitFactor
 	}
-	return o.CachedOpenFiles
+	return o.GetCompactionTableSize(level+1) * factor
+}
+
+func (o *Options) GetCompactionGPOverlaps(level int) int {
+	factor := DefaultCompactionGPOverlapsFactor
+	if o != nil && o.CompactionGPOverlapsFactor > 0 {
+		factor = o.CompactionGPOverlapsFactor
+	}
+	return o.GetCompactionTableSize(level+2) * factor
+}
+
+func (o *Options) GetCompactionL0Trigger() int {
+	if o == nil || o.CompactionL0Trigger == 0 {
+		return DefaultCompactionL0Trigger
+	}
+	return o.CompactionL0Trigger
+}
+
+func (o *Options) GetCompactionSourceLimit(level int) int {
+	factor := DefaultCompactionSourceLimitFactor
+	if o != nil && o.CompactionSourceLimitFactor > 0 {
+		factor = o.CompactionSourceLimitFactor
+	}
+	return o.GetCompactionTableSize(level+1) * factor
+}
+
+func (o *Options) GetCompactionTableSize(level int) int {
+	var (
+		base = DefaultCompactionTableSize
+		mult float64
+	)
+	if o != nil {
+		if o.CompactionTableSize > 0 {
+			base = o.CompactionTableSize
+		}
+		if len(o.CompactionTableSizeMultiplierPerLevel) > level && o.CompactionTableSizeMultiplierPerLevel[level] > 0 {
+			mult = o.CompactionTableSizeMultiplierPerLevel[level]
+		} else if o.CompactionTableSizeMultiplier > 0 {
+			mult = math.Pow(o.CompactionTableSizeMultiplier, float64(level))
+		}
+	}
+	if mult == 0 {
+		mult = math.Pow(DefaultCompactionTableSizeMultiplier, float64(level))
+	}
+	return int(float64(base) * mult)
+}
+
+func (o *Options) GetCompactionTotalSize(level int) int64 {
+	var (
+		base = DefaultCompactionTotalSize
+		mult float64
+	)
+	if o != nil {
+		if o.CompactionTotalSize > 0 {
+			base = o.CompactionTotalSize
+		}
+		if len(o.CompactionTotalSizeMultiplierPerLevel) > level && o.CompactionTotalSizeMultiplierPerLevel[level] > 0 {
+			mult = o.CompactionTotalSizeMultiplierPerLevel[level]
+		} else if o.CompactionTotalSizeMultiplier > 0 {
+			mult = math.Pow(o.CompactionTotalSizeMultiplier, float64(level))
+		}
+	}
+	if mult == 0 {
+		mult = math.Pow(DefaultCompactionTotalSizeMultiplier, float64(level))
+	}
+	return int64(float64(base) * mult)
 }
 
 func (o *Options) GetComparer() comparer.Comparer {
@@ -236,6 +464,13 @@ func (o *Options) GetCompression() Compression {
 	return o.Compression
 }
 
+func (o *Options) GetDisableCompactionBackoff() bool {
+	if o == nil {
+		return false
+	}
+	return o.DisableCompactionBackoff
+}
+
 func (o *Options) GetErrorIfExist() bool {
 	if o == nil {
 		return false
@@ -257,6 +492,47 @@ func (o *Options) GetFilter() filter.Filter {
 	return o.Filter
 }
 
+func (o *Options) GetMaxMemCompationLevel() int {
+	level := DefaultMaxMemCompationLevel
+	if o != nil {
+		if o.MaxMemCompationLevel > 0 {
+			level = o.MaxMemCompationLevel
+		} else if o.MaxMemCompationLevel == -1 {
+			level = 0
+		}
+	}
+	if level >= o.GetNumLevel() {
+		return o.GetNumLevel() - 1
+	}
+	return level
+}
+
+func (o *Options) GetNumLevel() int {
+	if o == nil || o.NumLevel <= 0 {
+		return DefaultNumLevel
+	}
+	return o.NumLevel
+}
+
+func (o *Options) GetOpenFilesCacher() Cacher {
+	if o == nil || o.OpenFilesCacher == nil {
+		return DefaultOpenFilesCacher
+	}
+	if o.OpenFilesCacher == NoCacher {
+		return nil
+	}
+	return o.OpenFilesCacher
+}
+
+func (o *Options) GetOpenFilesCacheCapacity() int {
+	if o == nil || o.OpenFilesCacheCapacity <= 0 {
+		return DefaultOpenFilesCacheCapacity
+	} else if o.OpenFilesCacheCapacity == -1 {
+		return 0
+	}
+	return o.OpenFilesCacheCapacity
+}
+
 func (o *Options) GetStrict(strict Strict) bool {
 	if o == nil || o.Strict == 0 {
 		return DefaultStrict&strict != 0
@@ -271,6 +547,20 @@ func (o *Options) GetWriteBuffer() int {
 	return o.WriteBuffer
 }
 
+func (o *Options) GetWriteL0PauseTrigger() int {
+	if o == nil || o.WriteL0PauseTrigger == 0 {
+		return DefaultWriteL0PauseTrigger
+	}
+	return o.WriteL0PauseTrigger
+}
+
+func (o *Options) GetWriteL0SlowdownTrigger() int {
+	if o == nil || o.WriteL0SlowdownTrigger == 0 {
+		return DefaultWriteL0SlowdownTrigger
+	}
+	return o.WriteL0SlowdownTrigger
+}
+
 // ReadOptions holds the optional parameters for 'read operation'. The
 // 'read operation' includes Get, Find and NewIterator.
 type ReadOptions struct {
@@ -281,8 +571,8 @@ type ReadOptions struct {
 	// The default value is false.
 	DontFillCache bool
 
-	// Strict overrides global DB strict level. Only StrictIterator and
-	// StrictBlockChecksum that does have effects here.
+	// Strict will be OR'ed with global DB 'strict level' unless StrictOverride
+	// is present. Currently only StrictReader that has effect here.
 	Strict Strict
 }
 
@@ -324,3 +614,11 @@ func (wo *WriteOptions) GetSync() bool {
 	}
 	return wo.Sync
 }
+
+func GetStrict(o *Options, ro *ReadOptions, strict Strict) bool {
+	if ro.GetStrict(StrictOverride) {
+		return ro.GetStrict(strict)
+	} else {
+		return o.GetStrict(strict) || ro.GetStrict(strict)
+	}
+}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/options.go

@@ -7,35 +7,86 @@
 package leveldb
 
 import (
-	"github.com/syndtr/goleveldb/leveldb/cache"
 	"github.com/syndtr/goleveldb/leveldb/filter"
 	"github.com/syndtr/goleveldb/leveldb/opt"
 )
 
-func (s *session) setOptions(o *opt.Options) {
-	s.o = &opt.Options{}
+func dupOptions(o *opt.Options) *opt.Options {
+	newo := &opt.Options{}
 	if o != nil {
-		*s.o = *o
+		*newo = *o
 	}
+	if newo.Strict == 0 {
+		newo.Strict = opt.DefaultStrict
+	}
+	return newo
+}
+
+func (s *session) setOptions(o *opt.Options) {
+	no := dupOptions(o)
 	// Alternative filters.
 	if filters := o.GetAltFilters(); len(filters) > 0 {
-		s.o.AltFilters = make([]filter.Filter, len(filters))
+		no.AltFilters = make([]filter.Filter, len(filters))
 		for i, filter := range filters {
-			s.o.AltFilters[i] = &iFilter{filter}
+			no.AltFilters[i] = &iFilter{filter}
 		}
 	}
-	// Block cache.
-	switch o.GetBlockCache() {
-	case nil:
-		s.o.BlockCache = cache.NewLRUCache(opt.DefaultBlockCacheSize)
-	case opt.NoCache:
-		s.o.BlockCache = nil
-	}
 	// Comparer.
 	s.icmp = &iComparer{o.GetComparer()}
-	s.o.Comparer = s.icmp
+	no.Comparer = s.icmp
 	// Filter.
 	if filter := o.GetFilter(); filter != nil {
-		s.o.Filter = &iFilter{filter}
+		no.Filter = &iFilter{filter}
+	}
+
+	s.o = &cachedOptions{Options: no}
+	s.o.cache()
+}
+
+type cachedOptions struct {
+	*opt.Options
+
+	compactionExpandLimit []int
+	compactionGPOverlaps  []int
+	compactionSourceLimit []int
+	compactionTableSize   []int
+	compactionTotalSize   []int64
+}
+
+func (co *cachedOptions) cache() {
+	numLevel := co.Options.GetNumLevel()
+
+	co.compactionExpandLimit = make([]int, numLevel)
+	co.compactionGPOverlaps = make([]int, numLevel)
+	co.compactionSourceLimit = make([]int, numLevel)
+	co.compactionTableSize = make([]int, numLevel)
+	co.compactionTotalSize = make([]int64, numLevel)
+
+	for level := 0; level < numLevel; level++ {
+		co.compactionExpandLimit[level] = co.Options.GetCompactionExpandLimit(level)
+		co.compactionGPOverlaps[level] = co.Options.GetCompactionGPOverlaps(level)
+		co.compactionSourceLimit[level] = co.Options.GetCompactionSourceLimit(level)
+		co.compactionTableSize[level] = co.Options.GetCompactionTableSize(level)
+		co.compactionTotalSize[level] = co.Options.GetCompactionTotalSize(level)
 	}
 }
+
+func (co *cachedOptions) GetCompactionExpandLimit(level int) int {
+	return co.compactionExpandLimit[level]
+}
+
+func (co *cachedOptions) GetCompactionGPOverlaps(level int) int {
+	return co.compactionGPOverlaps[level]
+}
+
+func (co *cachedOptions) GetCompactionSourceLimit(level int) int {
+	return co.compactionSourceLimit[level]
+}
+
+func (co *cachedOptions) GetCompactionTableSize(level int) int {
+	return co.compactionTableSize[level]
+}
+
+func (co *cachedOptions) GetCompactionTotalSize(level int) int64 {
+	return co.compactionTotalSize[level]
+}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/session.go

@@ -7,12 +7,13 @@
 package leveldb
 
 import (
-	"errors"
+	"fmt"
 	"io"
 	"os"
 	"sync"
 	"sync/atomic"
 
+	"github.com/syndtr/goleveldb/leveldb/errors"
 	"github.com/syndtr/goleveldb/leveldb/iterator"
 	"github.com/syndtr/goleveldb/leveldb/journal"
 	"github.com/syndtr/goleveldb/leveldb/opt"
@@ -20,18 +21,31 @@ import (
 	"github.com/syndtr/goleveldb/leveldb/util"
 )
 
+type ErrManifestCorrupted struct {
+	Field  string
+	Reason string
+}
+
+func (e *ErrManifestCorrupted) Error() string {
+	return fmt.Sprintf("leveldb: manifest corrupted (field '%s'): %s", e.Field, e.Reason)
+}
+
+func newErrManifestCorrupted(f storage.File, field, reason string) error {
+	return errors.NewErrCorrupted(f, &ErrManifestCorrupted{field, reason})
+}
+
 // session represent a persistent database session.
 type session struct {
 	// Need 64-bit alignment.
-	stFileNum        uint64 // current unused file number
+	stNextFileNum    uint64 // current unused file number
 	stJournalNum     uint64 // current journal file number; need external synchronization
 	stPrevJournalNum uint64 // prev journal file number; no longer used; for compatibility with older version of leveldb
-	stSeq            uint64 // last mem compacted seq; need external synchronization
+	stSeqNum         uint64 // last mem compacted seq; need external synchronization
 	stTempFileNum    uint64
 
 	stor     storage.Storage
 	storLock util.Releaser
-	o        *opt.Options
+	o        *cachedOptions
 	icmp     *iComparer
 	tops     *tOps
 
@@ -39,9 +53,9 @@ type session struct {
 	manifestWriter storage.Writer
 	manifestFile   storage.File
 
-	stCptrs   [kNumLevels]iKey // compact pointers; need external synchronization
-	stVersion *version         // current version
-	vmu       sync.Mutex
+	stCompPtrs []iKey   // compaction pointers; need external synchronization
+	stVersion  *version // current version
+	vmu        sync.Mutex
 }
 
 // Creates new initialized session instance.
@@ -54,22 +68,20 @@ func newSession(stor storage.Storage, o *opt.Options) (s *session, err error) {
 		return
 	}
 	s = &session{
-		stor:     stor,
-		storLock: storLock,
+		stor:       stor,
+		storLock:   storLock,
+		stCompPtrs: make([]iKey, o.GetNumLevel()),
 	}
 	s.setOptions(o)
-	s.tops = newTableOps(s, s.o.GetCachedOpenFiles())
-	s.setVersion(&version{s: s})
-	s.log("log@legend F·NumFile S·FileSize N·Entry C·BadEntry B·BadBlock D·DeletedEntry L·Level Q·SeqNum T·TimeElapsed")
+	s.tops = newTableOps(s)
+	s.setVersion(newVersion(s))
+	s.log("log@legend F·NumFile S·FileSize N·Entry C·BadEntry B·BadBlock Ke·KeyError D·DroppedEntry L·Level Q·SeqNum T·TimeElapsed")
 	return
 }
 
 // Close session.
 func (s *session) close() {
 	s.tops.close()
-	if bc := s.o.GetBlockCache(); bc != nil {
-		bc.Purge(nil)
-	}
 	if s.manifest != nil {
 		s.manifest.Close()
 	}
@@ -100,26 +112,26 @@ func (s *session) recover() (err error) {
 			// Don't return os.ErrNotExist if the underlying storage contains
 			// other files that belong to LevelDB. So the DB won't get trashed.
 			if files, _ := s.stor.GetFiles(storage.TypeAll); len(files) > 0 {
-				err = ErrCorrupted{Type: CorruptedManifest, Err: errors.New("leveldb: manifest file missing")}
+				err = &errors.ErrCorrupted{File: &storage.FileInfo{Type: storage.TypeManifest}, Err: &errors.ErrMissingFiles{}}
 			}
 		}
 	}()
 
-	file, err := s.stor.GetManifest()
+	m, err := s.stor.GetManifest()
 	if err != nil {
 		return
 	}
 
-	reader, err := file.Open()
+	reader, err := m.Open()
 	if err != nil {
 		return
 	}
 	defer reader.Close()
 	strict := s.o.GetStrict(opt.StrictManifest)
-	jr := journal.NewReader(reader, dropper{s, file}, strict, true)
+	jr := journal.NewReader(reader, dropper{s, m}, strict, true)
 
-	staging := s.version_NB().newStaging()
-	rec := &sessionRecord{}
+	staging := s.stVersion.newStaging()
+	rec := &sessionRecord{numLevel: s.o.GetNumLevel()}
 	for {
 		var r io.Reader
 		r, err = jr.Next()
@@ -128,51 +140,57 @@ func (s *session) recover() (err error) {
 				err = nil
 				break
 			}
-			return
+			return errors.SetFile(err, m)
 		}
 
 		err = rec.decode(r)
 		if err == nil {
 			// save compact pointers
-			for _, r := range rec.compactionPointers {
-				s.stCptrs[r.level] = iKey(r.ikey)
+			for _, r := range rec.compPtrs {
+				s.stCompPtrs[r.level] = iKey(r.ikey)
 			}
 			// commit record to version staging
 			staging.commit(rec)
-		} else if strict {
-			return ErrCorrupted{Type: CorruptedManifest, Err: err}
 		} else {
-			s.logf("manifest error: %v (skipped)", err)
+			err = errors.SetFile(err, m)
+			if strict || !errors.IsCorrupted(err) {
+				return
+			} else {
+				s.logf("manifest error: %v (skipped)", errors.SetFile(err, m))
+			}
 		}
-		rec.resetCompactionPointers()
+		rec.resetCompPtrs()
 		rec.resetAddedTables()
 		rec.resetDeletedTables()
 	}
 
 	switch {
 	case !rec.has(recComparer):
-		return ErrCorrupted{Type: CorruptedManifest, Err: errors.New("leveldb: manifest missing comparer name")}
+		return newErrManifestCorrupted(m, "comparer", "missing")
 	case rec.comparer != s.icmp.uName():
-		return ErrCorrupted{Type: CorruptedManifest, Err: errors.New("leveldb: comparer mismatch, " + "want '" + s.icmp.uName() + "', " + "got '" + rec.comparer + "'")}
-	case !rec.has(recNextNum):
-		return ErrCorrupted{Type: CorruptedManifest, Err: errors.New("leveldb: manifest missing next file number")}
+		return newErrManifestCorrupted(m, "comparer", fmt.Sprintf("mismatch: want '%s', got '%s'", s.icmp.uName(), rec.comparer))
+	case !rec.has(recNextFileNum):
+		return newErrManifestCorrupted(m, "next-file-num", "missing")
 	case !rec.has(recJournalNum):
-		return ErrCorrupted{Type: CorruptedManifest, Err: errors.New("leveldb: manifest missing journal file number")}
-	case !rec.has(recSeq):
-		return ErrCorrupted{Type: CorruptedManifest, Err: errors.New("leveldb: manifest missing seq number")}
+		return newErrManifestCorrupted(m, "journal-file-num", "missing")
+	case !rec.has(recSeqNum):
+		return newErrManifestCorrupted(m, "seq-num", "missing")
 	}
 
-	s.manifestFile = file
+	s.manifestFile = m
 	s.setVersion(staging.finish())
-	s.setFileNum(rec.nextNum)
+	s.setNextFileNum(rec.nextFileNum)
 	s.recordCommited(rec)
 	return nil
 }
 
 // Commit session; need external synchronization.
 func (s *session) commit(r *sessionRecord) (err error) {
+	v := s.version()
+	defer v.release()
+
 	// spawn new version based on current version
-	nv := s.version_NB().spawn(r)
+	nv := v.spawn(r)
 
 	if s.manifest == nil {
 		// manifest journal writer not yet created, create one
@@ -191,13 +209,13 @@ func (s *session) commit(r *sessionRecord) (err error) {
 
 // Pick a compaction based on current state; need external synchronization.
 func (s *session) pickCompaction() *compaction {
-	v := s.version_NB()
+	v := s.version()
 
 	var level int
 	var t0 tFiles
 	if v.cScore >= 1 {
 		level = v.cLevel
-		cptr := s.stCptrs[level]
+		cptr := s.stCompPtrs[level]
 		tables := v.tables[level]
 		for _, t := range tables {
 			if cptr == nil || s.icmp.Compare(t.imax, cptr) > 0 {
@@ -214,27 +232,21 @@ func (s *session) pickCompaction() *compaction {
 			level = ts.level
 			t0 = append(t0, ts.table)
 		} else {
+			v.release()
 			return nil
 		}
 	}
 
-	c := &compaction{s: s, v: v, level: level}
-	if level == 0 {
-		imin, imax := t0.getRange(s.icmp)
-		t0 = v.tables[0].getOverlaps(t0[:0], s.icmp, imin.ukey(), imax.ukey(), true)
-	}
-
-	c.tables[0] = t0
-	c.expand()
-	return c
+	return newCompaction(s, v, level, t0)
 }
 
 // Create compaction from given level and range; need external synchronization.
 func (s *session) getCompactionRange(level int, umin, umax []byte) *compaction {
-	v := s.version_NB()
+	v := s.version()
 
 	t0 := v.tables[level].getOverlaps(nil, s.icmp, umin, umax, level == 0)
 	if len(t0) == 0 {
+		v.release()
 		return nil
 	}
 
@@ -243,7 +255,7 @@ func (s *session) getCompactionRange(level int, umin, umax []byte) *compaction {
 	// and we must not pick one file and drop another older file if the
 	// two files overlap.
 	if level > 0 {
-		limit := uint64(kMaxTableSize)
+		limit := uint64(v.s.o.GetCompactionSourceLimit(level))
 		total := uint64(0)
 		for i, t := range t0 {
 			total += t.size
@@ -255,9 +267,20 @@ func (s *session) getCompactionRange(level int, umin, umax []byte) *compaction {
 		}
 	}
 
-	c := &compaction{s: s, v: v, level: level}
-	c.tables[0] = t0
+	return newCompaction(s, v, level, t0)
+}
+
+func newCompaction(s *session, v *version, level int, t0 tFiles) *compaction {
+	c := &compaction{
+		s:             s,
+		v:             v,
+		level:         level,
+		tables:        [2]tFiles{t0, nil},
+		maxGPOverlaps: uint64(s.o.GetCompactionGPOverlaps(level)),
+		tPtrs:         make([]int, s.o.GetNumLevel()),
+	}
 	c.expand()
+	c.save()
 	return c
 }
 
@@ -266,25 +289,57 @@ type compaction struct {
 	s *session
 	v *version
 
-	level  int
-	tables [2]tFiles
+	level         int
+	tables        [2]tFiles
+	maxGPOverlaps uint64
 
-	gp              tFiles
-	gpidx           int
-	seenKey         bool
-	overlappedBytes uint64
-	imin, imax      iKey
+	gp                tFiles
+	gpi               int
+	seenKey           bool
+	gpOverlappedBytes uint64
+	imin, imax        iKey
+	tPtrs             []int
+	released          bool
 
-	tPtrs [kNumLevels]int
+	snapGPI               int
+	snapSeenKey           bool
+	snapGPOverlappedBytes uint64
+	snapTPtrs             []int
+}
+
+func (c *compaction) save() {
+	c.snapGPI = c.gpi
+	c.snapSeenKey = c.seenKey
+	c.snapGPOverlappedBytes = c.gpOverlappedBytes
+	c.snapTPtrs = append(c.snapTPtrs[:0], c.tPtrs...)
+}
+
+func (c *compaction) restore() {
+	c.gpi = c.snapGPI
+	c.seenKey = c.snapSeenKey
+	c.gpOverlappedBytes = c.snapGPOverlappedBytes
+	c.tPtrs = append(c.tPtrs[:0], c.snapTPtrs...)
+}
+
+func (c *compaction) release() {
+	if !c.released {
+		c.released = true
+		c.v.release()
+	}
 }
 
 // Expand compacted tables; need external synchronization.
 func (c *compaction) expand() {
-	level := c.level
-	vt0, vt1 := c.v.tables[level], c.v.tables[level+1]
+	limit := uint64(c.s.o.GetCompactionExpandLimit(c.level))
+	vt0, vt1 := c.v.tables[c.level], c.v.tables[c.level+1]
 
 	t0, t1 := c.tables[0], c.tables[1]
 	imin, imax := t0.getRange(c.s.icmp)
+	// We expand t0 here just incase ukey hop across tables.
+	t0 = vt0.getOverlaps(t0, c.s.icmp, imin.ukey(), imax.ukey(), c.level == 0)
+	if len(t0) != len(c.tables[0]) {
+		imin, imax = t0.getRange(c.s.icmp)
+	}
 	t1 = vt1.getOverlaps(t1, c.s.icmp, imin.ukey(), imax.ukey(), false)
 	// Get entire range covered by compaction.
 	amin, amax := append(t0, t1...).getRange(c.s.icmp)
@@ -292,13 +347,13 @@ func (c *compaction) expand() {
 	// See if we can grow the number of inputs in "level" without
 	// changing the number of "level+1" files we pick up.
 	if len(t1) > 0 {
-		exp0 := vt0.getOverlaps(nil, c.s.icmp, amin.ukey(), amax.ukey(), level == 0)
-		if len(exp0) > len(t0) && t1.size()+exp0.size() < kExpCompactionMaxBytes {
+		exp0 := vt0.getOverlaps(nil, c.s.icmp, amin.ukey(), amax.ukey(), c.level == 0)
+		if len(exp0) > len(t0) && t1.size()+exp0.size() < limit {
 			xmin, xmax := exp0.getRange(c.s.icmp)
 			exp1 := vt1.getOverlaps(nil, c.s.icmp, xmin.ukey(), xmax.ukey(), false)
 			if len(exp1) == len(t1) {
 				c.s.logf("table@compaction expanding L%d+L%d (F·%d S·%s)+(F·%d S·%s) -> (F·%d S·%s)+(F·%d S·%s)",
-					level, level+1, len(t0), shortenb(int(t0.size())), len(t1), shortenb(int(t1.size())),
+					c.level, c.level+1, len(t0), shortenb(int(t0.size())), len(t1), shortenb(int(t1.size())),
 					len(exp0), shortenb(int(exp0.size())), len(exp1), shortenb(int(exp1.size())))
 				imin, imax = xmin, xmax
 				t0, t1 = exp0, exp1
@@ -309,8 +364,8 @@ func (c *compaction) expand() {
 
 	// Compute the set of grandparent files that overlap this compaction
 	// (parent == level+1; grandparent == level+2)
-	if level+2 < kNumLevels {
-		c.gp = c.v.tables[level+2].getOverlaps(c.gp, c.s.icmp, amin.ukey(), amax.ukey(), false)
+	if c.level+2 < c.s.o.GetNumLevel() {
+		c.gp = c.v.tables[c.level+2].getOverlaps(c.gp, c.s.icmp, amin.ukey(), amax.ukey(), false)
 	}
 
 	c.tables[0], c.tables[1] = t0, t1
@@ -319,7 +374,7 @@ func (c *compaction) expand() {
 
 // Check whether compaction is trivial.
 func (c *compaction) trivial() bool {
-	return len(c.tables[0]) == 1 && len(c.tables[1]) == 0 && c.gp.size() <= kMaxGrandParentOverlapBytes
+	return len(c.tables[0]) == 1 && len(c.tables[1]) == 0 && c.gp.size() <= c.maxGPOverlaps
 }
 
 func (c *compaction) baseLevelForKey(ukey []byte) bool {
@@ -341,20 +396,20 @@ func (c *compaction) baseLevelForKey(ukey []byte) bool {
 }
 
 func (c *compaction) shouldStopBefore(ikey iKey) bool {
-	for ; c.gpidx < len(c.gp); c.gpidx++ {
-		gp := c.gp[c.gpidx]
+	for ; c.gpi < len(c.gp); c.gpi++ {
+		gp := c.gp[c.gpi]
 		if c.s.icmp.Compare(ikey, gp.imax) <= 0 {
 			break
 		}
 		if c.seenKey {
-			c.overlappedBytes += gp.size
+			c.gpOverlappedBytes += gp.size
 		}
 	}
 	c.seenKey = true
 
-	if c.overlappedBytes > kMaxGrandParentOverlapBytes {
+	if c.gpOverlappedBytes > c.maxGPOverlaps {
 		// Too much overlap for current output; start new output.
-		c.overlappedBytes = 0
+		c.gpOverlappedBytes = 0
 		return true
 	}
 	return false
@@ -373,8 +428,12 @@ func (c *compaction) newIterator() iterator.Iterator {
 	// Options.
 	ro := &opt.ReadOptions{
 		DontFillCache: true,
+		Strict:        opt.StrictOverride,
+	}
+	strict := c.s.o.GetStrict(opt.StrictCompaction)
+	if strict {
+		ro.Strict |= opt.StrictReader
 	}
-	strict := c.s.o.GetStrict(opt.StrictIterator)
 
 	for i, tables := range c.tables {
 		if len(tables) == 0 {
@@ -387,10 +446,10 @@ func (c *compaction) newIterator() iterator.Iterator {
 				its = append(its, c.s.tops.newIterator(t, nil, ro))
 			}
 		} else {
-			it := iterator.NewIndexedIterator(tables.newIndexIterator(c.s.tops, c.s.icmp, nil, ro), strict, true)
+			it := iterator.NewIndexedIterator(tables.newIndexIterator(c.s.tops, c.s.icmp, nil, ro), strict)
 			its = append(its, it)
 		}
 	}
 
-	return iterator.NewMergedIterator(its, c.s.icmp, true)
+	return iterator.NewMergedIterator(its, c.s.icmp, strict)
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/session_record.go

@@ -9,11 +9,11 @@ package leveldb
 import (
 	"bufio"
 	"encoding/binary"
-	"errors"
 	"io"
-)
+	"strings"
 
-var errCorruptManifest = errors.New("leveldb: corrupt manifest")
+	"github.com/syndtr/goleveldb/leveldb/errors"
+)
 
 type byteReader interface {
 	io.Reader
@@ -22,13 +22,13 @@ type byteReader interface {
 
 // These numbers are written to disk and should not be changed.
 const (
-	recComparer          = 1
-	recJournalNum        = 2
-	recNextNum           = 3
-	recSeq               = 4
-	recCompactionPointer = 5
-	recDeletedTable      = 6
-	recNewTable          = 7
+	recComparer    = 1
+	recJournalNum  = 2
+	recNextFileNum = 3
+	recSeqNum      = 4
+	recCompPtr     = 5
+	recDelTable    = 6
+	recAddTable    = 7
 	// 8 was used for large value refs
 	recPrevJournalNum = 9
 )
@@ -38,7 +38,7 @@ type cpRecord struct {
 	ikey  iKey
 }
 
-type ntRecord struct {
+type atRecord struct {
 	level int
 	num   uint64
 	size  uint64
@@ -46,27 +46,26 @@ type ntRecord struct {
 	imax  iKey
 }
 
-func (r ntRecord) makeFile(s *session) *tFile {
-	return newTableFile(s.getTableFile(r.num), r.size, r.imin, r.imax)
-}
-
 type dtRecord struct {
 	level int
 	num   uint64
 }
 
 type sessionRecord struct {
-	hasRec             int
-	comparer           string
-	journalNum         uint64
-	prevJournalNum     uint64
-	nextNum            uint64
-	seq                uint64
-	compactionPointers []cpRecord
-	addedTables        []ntRecord
-	deletedTables      []dtRecord
-	scratch            [binary.MaxVarintLen64]byte
-	err                error
+	numLevel int
+
+	hasRec         int
+	comparer       string
+	journalNum     uint64
+	prevJournalNum uint64
+	nextFileNum    uint64
+	seqNum         uint64
+	compPtrs       []cpRecord
+	addedTables    []atRecord
+	deletedTables  []dtRecord
+
+	scratch [binary.MaxVarintLen64]byte
+	err     error
 }
 
 func (p *sessionRecord) has(rec int) bool {
@@ -88,29 +87,29 @@ func (p *sessionRecord) setPrevJournalNum(num uint64) {
 	p.prevJournalNum = num
 }
 
-func (p *sessionRecord) setNextNum(num uint64) {
-	p.hasRec |= 1 << recNextNum
-	p.nextNum = num
+func (p *sessionRecord) setNextFileNum(num uint64) {
+	p.hasRec |= 1 << recNextFileNum
+	p.nextFileNum = num
 }
 
-func (p *sessionRecord) setSeq(seq uint64) {
-	p.hasRec |= 1 << recSeq
-	p.seq = seq
+func (p *sessionRecord) setSeqNum(num uint64) {
+	p.hasRec |= 1 << recSeqNum
+	p.seqNum = num
 }
 
-func (p *sessionRecord) addCompactionPointer(level int, ikey iKey) {
-	p.hasRec |= 1 << recCompactionPointer
-	p.compactionPointers = append(p.compactionPointers, cpRecord{level, ikey})
+func (p *sessionRecord) addCompPtr(level int, ikey iKey) {
+	p.hasRec |= 1 << recCompPtr
+	p.compPtrs = append(p.compPtrs, cpRecord{level, ikey})
 }
 
-func (p *sessionRecord) resetCompactionPointers() {
-	p.hasRec &= ^(1 << recCompactionPointer)
-	p.compactionPointers = p.compactionPointers[:0]
+func (p *sessionRecord) resetCompPtrs() {
+	p.hasRec &= ^(1 << recCompPtr)
+	p.compPtrs = p.compPtrs[:0]
 }
 
 func (p *sessionRecord) addTable(level int, num, size uint64, imin, imax iKey) {
-	p.hasRec |= 1 << recNewTable
-	p.addedTables = append(p.addedTables, ntRecord{level, num, size, imin, imax})
+	p.hasRec |= 1 << recAddTable
+	p.addedTables = append(p.addedTables, atRecord{level, num, size, imin, imax})
 }
 
 func (p *sessionRecord) addTableFile(level int, t *tFile) {
@@ -118,17 +117,17 @@ func (p *sessionRecord) addTableFile(level int, t *tFile) {
 }
 
 func (p *sessionRecord) resetAddedTables() {
-	p.hasRec &= ^(1 << recNewTable)
+	p.hasRec &= ^(1 << recAddTable)
 	p.addedTables = p.addedTables[:0]
 }
 
-func (p *sessionRecord) deleteTable(level int, num uint64) {
-	p.hasRec |= 1 << recDeletedTable
+func (p *sessionRecord) delTable(level int, num uint64) {
+	p.hasRec |= 1 << recDelTable
 	p.deletedTables = append(p.deletedTables, dtRecord{level, num})
 }
 
 func (p *sessionRecord) resetDeletedTables() {
-	p.hasRec &= ^(1 << recDeletedTable)
+	p.hasRec &= ^(1 << recDelTable)
 	p.deletedTables = p.deletedTables[:0]
 }
 
@@ -161,26 +160,26 @@ func (p *sessionRecord) encode(w io.Writer) error {
 		p.putUvarint(w, recJournalNum)
 		p.putUvarint(w, p.journalNum)
 	}
-	if p.has(recNextNum) {
-		p.putUvarint(w, recNextNum)
-		p.putUvarint(w, p.nextNum)
+	if p.has(recNextFileNum) {
+		p.putUvarint(w, recNextFileNum)
+		p.putUvarint(w, p.nextFileNum)
 	}
-	if p.has(recSeq) {
-		p.putUvarint(w, recSeq)
-		p.putUvarint(w, p.seq)
+	if p.has(recSeqNum) {
+		p.putUvarint(w, recSeqNum)
+		p.putUvarint(w, p.seqNum)
 	}
-	for _, r := range p.compactionPointers {
-		p.putUvarint(w, recCompactionPointer)
+	for _, r := range p.compPtrs {
+		p.putUvarint(w, recCompPtr)
 		p.putUvarint(w, uint64(r.level))
 		p.putBytes(w, r.ikey)
 	}
 	for _, r := range p.deletedTables {
-		p.putUvarint(w, recDeletedTable)
+		p.putUvarint(w, recDelTable)
 		p.putUvarint(w, uint64(r.level))
 		p.putUvarint(w, r.num)
 	}
 	for _, r := range p.addedTables {
-		p.putUvarint(w, recNewTable)
+		p.putUvarint(w, recAddTable)
 		p.putUvarint(w, uint64(r.level))
 		p.putUvarint(w, r.num)
 		p.putUvarint(w, r.size)
@@ -190,14 +189,16 @@ func (p *sessionRecord) encode(w io.Writer) error {
 	return p.err
 }
 
-func (p *sessionRecord) readUvarint(r io.ByteReader) uint64 {
+func (p *sessionRecord) readUvarintMayEOF(field string, r io.ByteReader, mayEOF bool) uint64 {
 	if p.err != nil {
 		return 0
 	}
 	x, err := binary.ReadUvarint(r)
 	if err != nil {
-		if err == io.EOF {
-			p.err = errCorruptManifest
+		if err == io.ErrUnexpectedEOF || (mayEOF == false && err == io.EOF) {
+			p.err = errors.NewErrCorrupted(nil, &ErrManifestCorrupted{field, "short read"})
+		} else if strings.HasPrefix(err.Error(), "binary:") {
+			p.err = errors.NewErrCorrupted(nil, &ErrManifestCorrupted{field, err.Error()})
 		} else {
 			p.err = err
 		}
@@ -206,35 +207,39 @@ func (p *sessionRecord) readUvarint(r io.ByteReader) uint64 {
 	return x
 }
 
-func (p *sessionRecord) readBytes(r byteReader) []byte {
+func (p *sessionRecord) readUvarint(field string, r io.ByteReader) uint64 {
+	return p.readUvarintMayEOF(field, r, false)
+}
+
+func (p *sessionRecord) readBytes(field string, r byteReader) []byte {
 	if p.err != nil {
 		return nil
 	}
-	n := p.readUvarint(r)
+	n := p.readUvarint(field, r)
 	if p.err != nil {
 		return nil
 	}
 	x := make([]byte, n)
 	_, p.err = io.ReadFull(r, x)
 	if p.err != nil {
-		if p.err == io.EOF {
-			p.err = errCorruptManifest
+		if p.err == io.ErrUnexpectedEOF {
+			p.err = errors.NewErrCorrupted(nil, &ErrManifestCorrupted{field, "short read"})
 		}
 		return nil
 	}
 	return x
 }
 
-func (p *sessionRecord) readLevel(r io.ByteReader) int {
+func (p *sessionRecord) readLevel(field string, r io.ByteReader) int {
 	if p.err != nil {
 		return 0
 	}
-	x := p.readUvarint(r)
+	x := p.readUvarint(field, r)
 	if p.err != nil {
 		return 0
 	}
-	if x >= kNumLevels {
-		p.err = errCorruptManifest
+	if x >= uint64(p.numLevel) {
+		p.err = errors.NewErrCorrupted(nil, &ErrManifestCorrupted{field, "invalid level number"})
 		return 0
 	}
 	return int(x)
@@ -247,59 +252,59 @@ func (p *sessionRecord) decode(r io.Reader) error {
 	}
 	p.err = nil
 	for p.err == nil {
-		rec, err := binary.ReadUvarint(br)
-		if err != nil {
-			if err == io.EOF {
-				err = nil
+		rec := p.readUvarintMayEOF("field-header", br, true)
+		if p.err != nil {
+			if p.err == io.EOF {
+				return nil
 			}
-			return err
+			return p.err
 		}
 		switch rec {
 		case recComparer:
-			x := p.readBytes(br)
+			x := p.readBytes("comparer", br)
 			if p.err == nil {
 				p.setComparer(string(x))
 			}
 		case recJournalNum:
-			x := p.readUvarint(br)
+			x := p.readUvarint("journal-num", br)
 			if p.err == nil {
 				p.setJournalNum(x)
 			}
 		case recPrevJournalNum:
-			x := p.readUvarint(br)
+			x := p.readUvarint("prev-journal-num", br)
 			if p.err == nil {
 				p.setPrevJournalNum(x)
 			}
-		case recNextNum:
-			x := p.readUvarint(br)
+		case recNextFileNum:
+			x := p.readUvarint("next-file-num", br)
 			if p.err == nil {
-				p.setNextNum(x)
+				p.setNextFileNum(x)
 			}
-		case recSeq:
-			x := p.readUvarint(br)
+		case recSeqNum:
+			x := p.readUvarint("seq-num", br)
 			if p.err == nil {
-				p.setSeq(x)
+				p.setSeqNum(x)
 			}
-		case recCompactionPointer:
-			level := p.readLevel(br)
-			ikey := p.readBytes(br)
+		case recCompPtr:
+			level := p.readLevel("comp-ptr.level", br)
+			ikey := p.readBytes("comp-ptr.ikey", br)
 			if p.err == nil {
-				p.addCompactionPointer(level, iKey(ikey))
+				p.addCompPtr(level, iKey(ikey))
 			}
-		case recNewTable:
-			level := p.readLevel(br)
-			num := p.readUvarint(br)
-			size := p.readUvarint(br)
-			imin := p.readBytes(br)
-			imax := p.readBytes(br)
+		case recAddTable:
+			level := p.readLevel("add-table.level", br)
+			num := p.readUvarint("add-table.num", br)
+			size := p.readUvarint("add-table.size", br)
+			imin := p.readBytes("add-table.imin", br)
+			imax := p.readBytes("add-table.imax", br)
 			if p.err == nil {
 				p.addTable(level, num, size, imin, imax)
 			}
-		case recDeletedTable:
-			level := p.readLevel(br)
-			num := p.readUvarint(br)
+		case recDelTable:
+			level := p.readLevel("del-table.level", br)
+			num := p.readUvarint("del-table.num", br)
 			if p.err == nil {
-				p.deleteTable(level, num)
+				p.delTable(level, num)
 			}
 		}
 	}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/session_record_test.go

@@ -9,6 +9,8 @@ package leveldb
 import (
 	"bytes"
 	"testing"
+
+	"github.com/syndtr/goleveldb/leveldb/opt"
 )
 
 func decodeEncode(v *sessionRecord) (res bool, err error) {
@@ -17,7 +19,7 @@ func decodeEncode(v *sessionRecord) (res bool, err error) {
 	if err != nil {
 		return
 	}
-	v2 := new(sessionRecord)
+	v2 := &sessionRecord{numLevel: opt.DefaultNumLevel}
 	err = v.decode(b)
 	if err != nil {
 		return
@@ -32,7 +34,7 @@ func decodeEncode(v *sessionRecord) (res bool, err error) {
 
 func TestSessionRecord_EncodeDecode(t *testing.T) {
 	big := uint64(1) << 50
-	v := new(sessionRecord)
+	v := &sessionRecord{numLevel: opt.DefaultNumLevel}
 	i := uint64(0)
 	test := func() {
 		res, err := decodeEncode(v)
@@ -47,16 +49,16 @@ func TestSessionRecord_EncodeDecode(t *testing.T) {
 	for ; i < 4; i++ {
 		test()
 		v.addTable(3, big+300+i, big+400+i,
-			newIKey([]byte("foo"), big+500+1, tVal),
-			newIKey([]byte("zoo"), big+600+1, tDel))
-		v.deleteTable(4, big+700+i)
-		v.addCompactionPointer(int(i), newIKey([]byte("x"), big+900+1, tVal))
+			newIkey([]byte("foo"), big+500+1, ktVal),
+			newIkey([]byte("zoo"), big+600+1, ktDel))
+		v.delTable(4, big+700+i)
+		v.addCompPtr(int(i), newIkey([]byte("x"), big+900+1, ktVal))
 	}
 
 	v.setComparer("foo")
 	v.setJournalNum(big + 100)
 	v.setPrevJournalNum(big + 99)
-	v.setNextNum(big + 200)
-	v.setSeq(big + 1000)
+	v.setNextFileNum(big + 200)
+	v.setSeqNum(big + 1000)
 	test()
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/session_util.go

@@ -22,7 +22,7 @@ type dropper struct {
 }
 
 func (d dropper) Drop(err error) {
-	if e, ok := err.(journal.ErrCorrupted); ok {
+	if e, ok := err.(*journal.ErrCorrupted); ok {
 		d.s.logf("journal@drop %s-%d S·%s %q", d.file.Type(), d.file.Num(), shortenb(e.Size), e.Reason)
 	} else {
 		d.s.logf("journal@drop %s-%d %q", d.file.Type(), d.file.Num(), err)
@@ -51,9 +51,14 @@ func (s *session) newTemp() storage.File {
 	return s.stor.GetFile(num, storage.TypeTemp)
 }
 
+func (s *session) tableFileFromRecord(r atRecord) *tFile {
+	return newTableFile(s.getTableFile(r.num), r.size, r.imin, r.imax)
+}
+
 // Session state.
 
-// Get current version.
+// Get current version. This will incr version ref, must call
+// version.release (exactly once) after use.
 func (s *session) version() *version {
 	s.vmu.Lock()
 	defer s.vmu.Unlock()
@@ -61,61 +66,56 @@ func (s *session) version() *version {
 	return s.stVersion
 }
 
-// Get current version; no barrier.
-func (s *session) version_NB() *version {
-	return s.stVersion
-}
-
 // Set current version to v.
 func (s *session) setVersion(v *version) {
 	s.vmu.Lock()
-	v.ref = 1
+	v.ref = 1 // Holds by session.
 	if old := s.stVersion; old != nil {
-		v.ref++
+		v.ref++ // Holds by old version.
 		old.next = v
-		old.release_NB()
+		old.releaseNB()
 	}
 	s.stVersion = v
 	s.vmu.Unlock()
 }
 
 // Get current unused file number.
-func (s *session) fileNum() uint64 {
-	return atomic.LoadUint64(&s.stFileNum)
+func (s *session) nextFileNum() uint64 {
+	return atomic.LoadUint64(&s.stNextFileNum)
 }
 
-// Get current unused file number to num.
-func (s *session) setFileNum(num uint64) {
-	atomic.StoreUint64(&s.stFileNum, num)
+// Set current unused file number to num.
+func (s *session) setNextFileNum(num uint64) {
+	atomic.StoreUint64(&s.stNextFileNum, num)
 }
 
 // Mark file number as used.
 func (s *session) markFileNum(num uint64) {
-	num += 1
+	nextFileNum := num + 1
 	for {
-		old, x := s.stFileNum, num
+		old, x := s.stNextFileNum, nextFileNum
 		if old > x {
 			x = old
 		}
-		if atomic.CompareAndSwapUint64(&s.stFileNum, old, x) {
+		if atomic.CompareAndSwapUint64(&s.stNextFileNum, old, x) {
 			break
 		}
 	}
 }
 
 // Allocate a file number.
-func (s *session) allocFileNum() (num uint64) {
-	return atomic.AddUint64(&s.stFileNum, 1) - 1
+func (s *session) allocFileNum() uint64 {
+	return atomic.AddUint64(&s.stNextFileNum, 1) - 1
 }
 
 // Reuse given file number.
 func (s *session) reuseFileNum(num uint64) {
 	for {
-		old, x := s.stFileNum, num
+		old, x := s.stNextFileNum, num
 		if old != x+1 {
 			x = old
 		}
-		if atomic.CompareAndSwapUint64(&s.stFileNum, old, x) {
+		if atomic.CompareAndSwapUint64(&s.stNextFileNum, old, x) {
 			break
 		}
 	}
@@ -126,20 +126,20 @@ func (s *session) reuseFileNum(num uint64) {
 // Fill given session record obj with current states; need external
 // synchronization.
 func (s *session) fillRecord(r *sessionRecord, snapshot bool) {
-	r.setNextNum(s.fileNum())
+	r.setNextFileNum(s.nextFileNum())
 
 	if snapshot {
 		if !r.has(recJournalNum) {
 			r.setJournalNum(s.stJournalNum)
 		}
 
-		if !r.has(recSeq) {
-			r.setSeq(s.stSeq)
+		if !r.has(recSeqNum) {
+			r.setSeqNum(s.stSeqNum)
 		}
 
-		for level, ik := range s.stCptrs {
+		for level, ik := range s.stCompPtrs {
 			if ik != nil {
-				r.addCompactionPointer(level, ik)
+				r.addCompPtr(level, ik)
 			}
 		}
 
@@ -158,12 +158,12 @@ func (s *session) recordCommited(r *sessionRecord) {
 		s.stPrevJournalNum = r.prevJournalNum
 	}
 
-	if r.has(recSeq) {
-		s.stSeq = r.seq
+	if r.has(recSeqNum) {
+		s.stSeqNum = r.seqNum
 	}
 
-	for _, p := range r.compactionPointers {
-		s.stCptrs[p.level] = iKey(p.ikey)
+	for _, p := range r.compPtrs {
+		s.stCompPtrs[p.level] = iKey(p.ikey)
 	}
 }
 
@@ -178,10 +178,11 @@ func (s *session) newManifest(rec *sessionRecord, v *version) (err error) {
 	jw := journal.NewWriter(writer)
 
 	if v == nil {
-		v = s.version_NB()
+		v = s.version()
+		defer v.release()
 	}
 	if rec == nil {
-		rec = new(sessionRecord)
+		rec = &sessionRecord{numLevel: s.o.GetNumLevel()}
 	}
 	s.fillRecord(rec, true)
 	v.fillRecord(rec)

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/storage/file_storage.go

@@ -221,7 +221,7 @@ func (fs *fileStorage) GetManifest() (f File, err error) {
 					fs.log(fmt.Sprintf("skipping %s: invalid file name", fn))
 					continue
 				}
-				if _, e1 := strconv.ParseUint(fn[7:], 10, 0); e1 != nil {
+				if _, e1 := strconv.ParseUint(fn[8:], 10, 0); e1 != nil {
 					fs.log(fmt.Sprintf("skipping %s: invalid file num: %v", fn, e1))
 					continue
 				}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/storage/storage.go

@@ -125,3 +125,33 @@ type Storage interface {
 	// Other methods should not be called after the storage has been closed.
 	Close() error
 }
+
+// FileInfo wraps basic file info.
+type FileInfo struct {
+	Type FileType
+	Num  uint64
+}
+
+func (fi FileInfo) String() string {
+	switch fi.Type {
+	case TypeManifest:
+		return fmt.Sprintf("MANIFEST-%06d", fi.Num)
+	case TypeJournal:
+		return fmt.Sprintf("%06d.log", fi.Num)
+	case TypeTable:
+		return fmt.Sprintf("%06d.ldb", fi.Num)
+	case TypeTemp:
+		return fmt.Sprintf("%06d.tmp", fi.Num)
+	default:
+		return fmt.Sprintf("%#x-%d", fi.Type, fi.Num)
+	}
+}
+
+// NewFileInfo creates new FileInfo from the given File. It will returns nil
+// if File is nil.
+func NewFileInfo(f File) *FileInfo {
+	if f == nil {
+		return nil
+	}
+	return &FileInfo{f.Type(), f.Num()}
+}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/storage_test.go

@@ -11,6 +11,7 @@ import (
 	"fmt"
 	"io"
 	"io/ioutil"
+	"math/rand"
 	"os"
 	"path/filepath"
 	"sync"
@@ -28,11 +29,25 @@ var (
 )
 
 var (
-	tsFSEnv  = os.Getenv("GOLEVELDB_USEFS")
-	tsKeepFS = tsFSEnv == "2"
-	tsFS     = tsKeepFS || tsFSEnv == "" || tsFSEnv == "1"
-	tsMU     = &sync.Mutex{}
-	tsNum    = 0
+	tsFSEnv   = os.Getenv("GOLEVELDB_USEFS")
+	tsTempdir = os.Getenv("GOLEVELDB_TEMPDIR")
+	tsKeepFS  = tsFSEnv == "2"
+	tsFS      = tsKeepFS || tsFSEnv == "" || tsFSEnv == "1"
+	tsMU      = &sync.Mutex{}
+	tsNum     = 0
+)
+
+type tsOp uint
+
+const (
+	tsOpOpen tsOp = iota
+	tsOpCreate
+	tsOpRead
+	tsOpReadAt
+	tsOpWrite
+	tsOpSync
+
+	tsOpNum
 )
 
 type tsLock struct {
@@ -53,6 +68,9 @@ type tsReader struct {
 func (tr tsReader) Read(b []byte) (n int, err error) {
 	ts := tr.tf.ts
 	ts.countRead(tr.tf.Type())
+	if tr.tf.shouldErrLocked(tsOpRead) {
+		return 0, errors.New("leveldb.testStorage: emulated read error")
+	}
 	n, err = tr.Reader.Read(b)
 	if err != nil && err != io.EOF {
 		ts.t.Errorf("E: read error, num=%d type=%v n=%d: %v", tr.tf.Num(), tr.tf.Type(), n, err)
@@ -63,6 +81,9 @@ func (tr tsReader) Read(b []byte) (n int, err error) {
 func (tr tsReader) ReadAt(b []byte, off int64) (n int, err error) {
 	ts := tr.tf.ts
 	ts.countRead(tr.tf.Type())
+	if tr.tf.shouldErrLocked(tsOpReadAt) {
+		return 0, errors.New("leveldb.testStorage: emulated readAt error")
+	}
 	n, err = tr.Reader.ReadAt(b, off)
 	if err != nil && err != io.EOF {
 		ts.t.Errorf("E: readAt error, num=%d type=%v off=%d n=%d: %v", tr.tf.Num(), tr.tf.Type(), off, n, err)
@@ -82,15 +103,12 @@ type tsWriter struct {
 }
 
 func (tw tsWriter) Write(b []byte) (n int, err error) {
-	ts := tw.tf.ts
-	ts.mu.Lock()
-	defer ts.mu.Unlock()
-	if ts.emuWriteErr&tw.tf.Type() != 0 {
+	if tw.tf.shouldErrLocked(tsOpWrite) {
 		return 0, errors.New("leveldb.testStorage: emulated write error")
 	}
 	n, err = tw.Writer.Write(b)
 	if err != nil {
-		ts.t.Errorf("E: write error, num=%d type=%v n=%d: %v", tw.tf.Num(), tw.tf.Type(), n, err)
+		tw.tf.ts.t.Errorf("E: write error, num=%d type=%v n=%d: %v", tw.tf.Num(), tw.tf.Type(), n, err)
 	}
 	return
 }
@@ -98,23 +116,23 @@ func (tw tsWriter) Write(b []byte) (n int, err error) {
 func (tw tsWriter) Sync() (err error) {
 	ts := tw.tf.ts
 	ts.mu.Lock()
-	defer ts.mu.Unlock()
 	for ts.emuDelaySync&tw.tf.Type() != 0 {
 		ts.cond.Wait()
 	}
-	if ts.emuSyncErr&tw.tf.Type() != 0 {
+	ts.mu.Unlock()
+	if tw.tf.shouldErrLocked(tsOpSync) {
 		return errors.New("leveldb.testStorage: emulated sync error")
 	}
 	err = tw.Writer.Sync()
 	if err != nil {
-		ts.t.Errorf("E: sync error, num=%d type=%v: %v", tw.tf.Num(), tw.tf.Type(), err)
+		tw.tf.ts.t.Errorf("E: sync error, num=%d type=%v: %v", tw.tf.Num(), tw.tf.Type(), err)
 	}
 	return
 }
 
 func (tw tsWriter) Close() (err error) {
 	err = tw.Writer.Close()
-	tw.tf.close("reader", err)
+	tw.tf.close("writer", err)
 	return
 }
 
@@ -127,6 +145,16 @@ func (tf tsFile) x() uint64 {
 	return tf.Num()<<typeShift | uint64(tf.Type())
 }
 
+func (tf tsFile) shouldErr(op tsOp) bool {
+	return tf.ts.shouldErr(tf, op)
+}
+
+func (tf tsFile) shouldErrLocked(op tsOp) bool {
+	tf.ts.mu.Lock()
+	defer tf.ts.mu.Unlock()
+	return tf.shouldErr(op)
+}
+
 func (tf tsFile) checkOpen(m string) error {
 	ts := tf.ts
 	if writer, ok := ts.opens[tf.x()]; ok {
@@ -163,7 +191,7 @@ func (tf tsFile) Open() (r storage.Reader, err error) {
 	if err != nil {
 		return
 	}
-	if ts.emuOpenErr&tf.Type() != 0 {
+	if tf.shouldErr(tsOpOpen) {
 		err = errors.New("leveldb.testStorage: emulated open error")
 		return
 	}
@@ -190,7 +218,7 @@ func (tf tsFile) Create() (w storage.Writer, err error) {
 	if err != nil {
 		return
 	}
-	if ts.emuCreateErr&tf.Type() != 0 {
+	if tf.shouldErr(tsOpCreate) {
 		err = errors.New("leveldb.testStorage: emulated create error")
 		return
 	}
@@ -205,6 +233,23 @@ func (tf tsFile) Create() (w storage.Writer, err error) {
 	return
 }
 
+func (tf tsFile) Replace(newfile storage.File) (err error) {
+	ts := tf.ts
+	ts.mu.Lock()
+	defer ts.mu.Unlock()
+	err = tf.checkOpen("replace")
+	if err != nil {
+		return
+	}
+	err = tf.File.Replace(newfile.(tsFile).File)
+	if err != nil {
+		ts.t.Errorf("E: cannot replace file, num=%d type=%v: %v", tf.Num(), tf.Type(), err)
+	} else {
+		ts.t.Logf("I: file replace, num=%d type=%v", tf.Num(), tf.Type())
+	}
+	return
+}
+
 func (tf tsFile) Remove() (err error) {
 	ts := tf.ts
 	ts.mu.Lock()
@@ -231,25 +276,61 @@ type testStorage struct {
 	cond sync.Cond
 	// Open files, true=writer, false=reader
 	opens         map[uint64]bool
-	emuOpenErr    storage.FileType
-	emuCreateErr  storage.FileType
 	emuDelaySync  storage.FileType
-	emuWriteErr   storage.FileType
-	emuSyncErr    storage.FileType
 	ignoreOpenErr storage.FileType
 	readCnt       uint64
 	readCntEn     storage.FileType
+
+	emuErr         [tsOpNum]storage.FileType
+	emuErrOnce     [tsOpNum]storage.FileType
+	emuRandErr     [tsOpNum]storage.FileType
+	emuRandErrProb int
+	emuErrOnceMap  map[uint64]uint
+	emuRandRand    *rand.Rand
 }
 
-func (ts *testStorage) SetOpenErr(t storage.FileType) {
+func (ts *testStorage) shouldErr(tf tsFile, op tsOp) bool {
+	if ts.emuErr[op]&tf.Type() != 0 {
+		return true
+	} else if ts.emuRandErr[op]&tf.Type() != 0 || ts.emuErrOnce[op]&tf.Type() != 0 {
+		sop := uint(1) << op
+		eop := ts.emuErrOnceMap[tf.x()]
+		if eop&sop == 0 && (ts.emuRandRand.Int()%ts.emuRandErrProb == 0 || ts.emuErrOnce[op]&tf.Type() != 0) {
+			ts.emuErrOnceMap[tf.x()] = eop | sop
+			ts.t.Logf("I: emulated error: file=%d type=%v op=%v", tf.Num(), tf.Type(), op)
+			return true
+		}
+	}
+	return false
+}
+
+func (ts *testStorage) SetEmuErr(t storage.FileType, ops ...tsOp) {
 	ts.mu.Lock()
-	ts.emuOpenErr = t
+	for _, op := range ops {
+		ts.emuErr[op] = t
+	}
 	ts.mu.Unlock()
 }
 
-func (ts *testStorage) SetCreateErr(t storage.FileType) {
+func (ts *testStorage) SetEmuErrOnce(t storage.FileType, ops ...tsOp) {
 	ts.mu.Lock()
-	ts.emuCreateErr = t
+	for _, op := range ops {
+		ts.emuErrOnce[op] = t
+	}
+	ts.mu.Unlock()
+}
+
+func (ts *testStorage) SetEmuRandErr(t storage.FileType, ops ...tsOp) {
+	ts.mu.Lock()
+	for _, op := range ops {
+		ts.emuRandErr[op] = t
+	}
+	ts.mu.Unlock()
+}
+
+func (ts *testStorage) SetEmuRandErrProb(prob int) {
+	ts.mu.Lock()
+	ts.emuRandErrProb = prob
 	ts.mu.Unlock()
 }
 
@@ -267,18 +348,6 @@ func (ts *testStorage) ReleaseSync(t storage.FileType) {
 	ts.mu.Unlock()
 }
 
-func (ts *testStorage) SetWriteErr(t storage.FileType) {
-	ts.mu.Lock()
-	ts.emuWriteErr = t
-	ts.mu.Unlock()
-}
-
-func (ts *testStorage) SetSyncErr(t storage.FileType) {
-	ts.mu.Lock()
-	ts.emuSyncErr = t
-	ts.mu.Unlock()
-}
-
 func (ts *testStorage) ReadCounter() uint64 {
 	ts.mu.Lock()
 	defer ts.mu.Unlock()
@@ -413,7 +482,11 @@ func newTestStorage(t *testing.T) *testStorage {
 			num := tsNum
 			tsNum++
 			tsMU.Unlock()
-			path := filepath.Join(os.TempDir(), fmt.Sprintf("goleveldb-test%d0%d0%d", os.Getuid(), os.Getpid(), num))
+			tempdir := tsTempdir
+			if tempdir == "" {
+				tempdir = os.TempDir()
+			}
+			path := filepath.Join(tempdir, fmt.Sprintf("goleveldb-test%d0%d0%d", os.Getuid(), os.Getpid(), num))
 			if _, err := os.Stat(path); err != nil {
 				stor, err = storage.OpenFile(path)
 				if err != nil {
@@ -436,6 +509,10 @@ func newTestStorage(t *testing.T) *testStorage {
 						}
 						f.Close()
 					}
+					if t.Failed() {
+						t.Logf("testing failed, test DB preserved at %s", path)
+						return nil
+					}
 					if tsKeepFS {
 						return nil
 					}
@@ -449,10 +526,13 @@ func newTestStorage(t *testing.T) *testStorage {
 		stor = storage.NewMemStorage()
 	}
 	ts := &testStorage{
-		t:       t,
-		Storage: stor,
-		closeFn: closeFn,
-		opens:   make(map[uint64]bool),
+		t:              t,
+		Storage:        stor,
+		closeFn:        closeFn,
+		opens:          make(map[uint64]bool),
+		emuErrOnceMap:  make(map[uint64]uint),
+		emuRandErrProb: 0x999,
+		emuRandRand:    rand.New(rand.NewSource(0xfacedead)),
 	}
 	ts.cond.L = &ts.mu
 	return ts

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/table.go

@@ -9,7 +9,6 @@ package leveldb
 import (
 	"fmt"
 	"sort"
-	"sync"
 	"sync/atomic"
 
 	"github.com/syndtr/goleveldb/leveldb/cache"
@@ -84,6 +83,18 @@ type tFiles []*tFile
 func (tf tFiles) Len() int      { return len(tf) }
 func (tf tFiles) Swap(i, j int) { tf[i], tf[j] = tf[j], tf[i] }
 
+func (tf tFiles) nums() string {
+	x := "[ "
+	for i, f := range tf {
+		if i != 0 {
+			x += ", "
+		}
+		x += fmt.Sprint(f.file.Num())
+	}
+	x += " ]"
+	return x
+}
+
 // Returns true if i smallest key is less than j.
 // This used for sort by key in ascending order.
 func (tf tFiles) lessByKey(icmp *iComparer, i, j int) bool {
@@ -151,7 +162,7 @@ func (tf tFiles) overlaps(icmp *iComparer, umin, umax []byte, unsorted bool) boo
 	i := 0
 	if len(umin) > 0 {
 		// Find the earliest possible internal key for min.
-		i = tf.searchMax(icmp, newIKey(umin, kMaxSeq, tSeek))
+		i = tf.searchMax(icmp, newIkey(umin, kMaxSeq, ktSeek))
 	}
 	if i >= len(tf) {
 		// Beginning of range is after all files, so no overlap.
@@ -161,24 +172,25 @@ func (tf tFiles) overlaps(icmp *iComparer, umin, umax []byte, unsorted bool) boo
 }
 
 // Returns tables whose its key range overlaps with given key range.
-// If overlapped is true then the search will be expanded to tables that
-// overlaps with each other.
+// Range will be expanded if ukey found hop across tables.
+// If overlapped is true then the search will be restarted if umax
+// expanded.
+// The dst content will be overwritten.
 func (tf tFiles) getOverlaps(dst tFiles, icmp *iComparer, umin, umax []byte, overlapped bool) tFiles {
-	x := len(dst)
+	dst = dst[:0]
 	for i := 0; i < len(tf); {
 		t := tf[i]
 		if t.overlaps(icmp, umin, umax) {
-			if overlapped {
-				// For overlapped files, check if the newly added file has
-				// expanded the range. If so, restart search.
-				if umin != nil && icmp.uCompare(t.imin.ukey(), umin) < 0 {
-					umin = t.imin.ukey()
-					dst = dst[:x]
-					i = 0
-					continue
-				} else if umax != nil && icmp.uCompare(t.imax.ukey(), umax) > 0 {
-					umax = t.imax.ukey()
-					dst = dst[:x]
+			if umin != nil && icmp.uCompare(t.imin.ukey(), umin) < 0 {
+				umin = t.imin.ukey()
+				dst = dst[:0]
+				i = 0
+				continue
+			} else if umax != nil && icmp.uCompare(t.imax.ukey(), umax) > 0 {
+				umax = t.imax.ukey()
+				// Restart search if it is overlapped.
+				if overlapped {
+					dst = dst[:0]
 					i = 0
 					continue
 				}
@@ -274,12 +286,10 @@ func (x *tFilesSortByNum) Less(i, j int) bool {
 
 // Table operations.
 type tOps struct {
-	s       *session
-	cache   cache.Cache
-	cacheNS cache.Namespace
-	bpool   *util.BufferPool
-	mu      sync.Mutex
-	closed  bool
+	s      *session
+	cache  *cache.Cache
+	bcache *cache.Cache
+	bpool  *util.BufferPool
 }
 
 // Creates an empty table and returns table writer.
@@ -293,7 +303,7 @@ func (t *tOps) create() (*tWriter, error) {
 		t:    t,
 		file: file,
 		w:    fw,
-		tw:   table.NewWriter(fw, t.s.o),
+		tw:   table.NewWriter(fw, t.s.o.Options),
 	}, nil
 }
 
@@ -326,61 +336,34 @@ func (t *tOps) createFrom(src iterator.Iterator) (f *tFile, n int, err error) {
 	return
 }
 
-type trWrapper struct {
-	*table.Reader
-	t   *tOps
-	ref int
-}
-
-func (w *trWrapper) Release() {
-	if w.ref != 0 && !w.t.closed {
-		panic(fmt.Sprintf("BUG: invalid ref %d, refer to issue #72", w.ref))
-	}
-	w.Reader.Release()
-}
-
-type trCacheHandleWrapper struct {
-	cache.Handle
-	t        *tOps
-	released bool
-}
-
-func (w *trCacheHandleWrapper) Release() {
-	w.t.mu.Lock()
-	defer w.t.mu.Unlock()
-
-	if !w.released {
-		w.released = true
-		w.Value().(*trWrapper).ref--
-	}
-	w.Handle.Release()
-}
-
 // Opens table. It returns a cache handle, which should
 // be released after use.
-func (t *tOps) open(f *tFile) (ch cache.Handle, err error) {
-	t.mu.Lock()
-	defer t.mu.Unlock()
-
+func (t *tOps) open(f *tFile) (ch *cache.Handle, err error) {
 	num := f.file.Num()
-	ch = t.cacheNS.Get(num, func() (charge int, value interface{}) {
+	ch = t.cache.Get(0, num, func() (size int, value cache.Value) {
 		var r storage.Reader
 		r, err = f.file.Open()
 		if err != nil {
 			return 0, nil
 		}
 
-		var bcacheNS cache.Namespace
-		if bc := t.s.o.GetBlockCache(); bc != nil {
-			bcacheNS = bc.GetNamespace(num)
+		var bcache *cache.CacheGetter
+		if t.bcache != nil {
+			bcache = &cache.CacheGetter{Cache: t.bcache, NS: num}
 		}
-		return 1, &trWrapper{table.NewReader(r, int64(f.size), bcacheNS, t.bpool, t.s.o), t, 0}
+
+		var tr *table.Reader
+		tr, err = table.NewReader(r, int64(f.size), storage.NewFileInfo(f.file), bcache, t.bpool, t.s.o.Options)
+		if err != nil {
+			r.Close()
+			return 0, nil
+		}
+		return 1, tr
+
 	})
 	if ch == nil && err == nil {
 		err = ErrClosed
 	}
-	ch.Value().(*trWrapper).ref++
-	ch = &trCacheHandleWrapper{ch, t, false}
 	return
 }
 
@@ -392,7 +375,17 @@ func (t *tOps) find(f *tFile, key []byte, ro *opt.ReadOptions) (rkey, rvalue []b
 		return nil, nil, err
 	}
 	defer ch.Release()
-	return ch.Value().(*trWrapper).Find(key, ro)
+	return ch.Value().(*table.Reader).Find(key, true, ro)
+}
+
+// Finds key that is greater than or equal to the given key.
+func (t *tOps) findKey(f *tFile, key []byte, ro *opt.ReadOptions) (rkey []byte, err error) {
+	ch, err := t.open(f)
+	if err != nil {
+		return nil, err
+	}
+	defer ch.Release()
+	return ch.Value().(*table.Reader).FindKey(key, true, ro)
 }
 
 // Returns approximate offset of the given key.
@@ -402,7 +395,7 @@ func (t *tOps) offsetOf(f *tFile, key []byte) (offset uint64, err error) {
 		return
 	}
 	defer ch.Release()
-	offset_, err := ch.Value().(*trWrapper).OffsetOf(key)
+	offset_, err := ch.Value().(*table.Reader).OffsetOf(key)
 	return uint64(offset_), err
 }
 
@@ -412,7 +405,7 @@ func (t *tOps) newIterator(f *tFile, slice *util.Range, ro *opt.ReadOptions) ite
 	if err != nil {
 		return iterator.NewEmptyIterator(err)
 	}
-	iter := ch.Value().(*trWrapper).NewIterator(slice, ro)
+	iter := ch.Value().(*table.Reader).NewIterator(slice, ro)
 	iter.SetReleaser(ch)
 	return iter
 }
@@ -420,20 +413,15 @@ func (t *tOps) newIterator(f *tFile, slice *util.Range, ro *opt.ReadOptions) ite
 // Removes table from persistent storage. It waits until
 // no one use the the table.
 func (t *tOps) remove(f *tFile) {
-	t.mu.Lock()
-	defer t.mu.Unlock()
-
 	num := f.file.Num()
-	t.cacheNS.Delete(num, func(exist, pending bool) {
-		if !pending {
-			if err := f.file.Remove(); err != nil {
-				t.s.logf("table@remove removing @%d %q", num, err)
-			} else {
-				t.s.logf("table@remove removed @%d", num)
-			}
-			if bc := t.s.o.GetBlockCache(); bc != nil {
-				bc.ZapNamespace(num)
-			}
+	t.cache.Delete(0, num, func() {
+		if err := f.file.Remove(); err != nil {
+			t.s.logf("table@remove removing @%d %q", num, err)
+		} else {
+			t.s.logf("table@remove removed @%d", num)
+		}
+		if t.bcache != nil {
+			t.bcache.EvictNS(num)
 		}
 	})
 }
@@ -441,22 +429,34 @@ func (t *tOps) remove(f *tFile) {
 // Closes the table ops instance. It will close all tables,
 // regadless still used or not.
 func (t *tOps) close() {
-	t.mu.Lock()
-	defer t.mu.Unlock()
-
-	t.closed = true
-	t.cache.Zap()
 	t.bpool.Close()
+	t.cache.Close()
+	if t.bcache != nil {
+		t.bcache.Close()
+	}
 }
 
 // Creates new initialized table ops instance.
-func newTableOps(s *session, cacheCap int) *tOps {
-	c := cache.NewLRUCache(cacheCap)
+func newTableOps(s *session) *tOps {
+	var (
+		cacher cache.Cacher
+		bcache *cache.Cache
+	)
+	if s.o.GetOpenFilesCacheCapacity() > 0 {
+		cacher = cache.NewLRU(s.o.GetOpenFilesCacheCapacity())
+	}
+	if !s.o.DisableBlockCache {
+		var bcacher cache.Cacher
+		if s.o.GetBlockCacheCapacity() > 0 {
+			bcacher = cache.NewLRU(s.o.GetBlockCacheCapacity())
+		}
+		bcache = cache.NewCache(bcacher)
+	}
 	return &tOps{
-		s:       s,
-		cache:   c,
-		cacheNS: c.GetNamespace(0),
-		bpool:   util.NewBufferPool(s.o.GetBlockSize() + 5),
+		s:      s,
+		cache:  cache.NewCache(cacher),
+		bcache: bcache,
+		bpool:  util.NewBufferPool(s.o.GetBlockSize() + 5),
 	}
 }
 
@@ -486,28 +486,34 @@ func (w *tWriter) empty() bool {
 	return w.first == nil
 }
 
+// Closes the storage.Writer.
+func (w *tWriter) close() {
+	if w.w != nil {
+		w.w.Close()
+		w.w = nil
+	}
+}
+
 // Finalizes the table and returns table file.
 func (w *tWriter) finish() (f *tFile, err error) {
+	defer w.close()
 	err = w.tw.Close()
 	if err != nil {
 		return
 	}
 	err = w.w.Sync()
 	if err != nil {
-		w.w.Close()
 		return
 	}
-	w.w.Close()
 	f = newTableFile(w.file, uint64(w.tw.BytesLen()), iKey(w.first), iKey(w.last))
 	return
 }
 
 // Drops the table.
 func (w *tWriter) drop() {
-	w.w.Close()
+	w.close()
 	w.file.Remove()
 	w.t.s.reuseFileNum(w.file.Num())
-	w.w = nil
 	w.file = nil
 	w.tw = nil
 	w.first = nil

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/table/block_test.go

@@ -19,13 +19,18 @@ import (
 	"github.com/syndtr/goleveldb/leveldb/util"
 )
 
-func (b *block) TestNewIterator(slice *util.Range) iterator.Iterator {
-	return b.newIterator(slice, false, nil)
+type blockTesting struct {
+	tr *Reader
+	b  *block
+}
+
+func (t *blockTesting) TestNewIterator(slice *util.Range) iterator.Iterator {
+	return t.tr.newBlockIter(t.b, nil, slice, false)
 }
 
 var _ = testutil.Defer(func() {
 	Describe("Block", func() {
-		Build := func(kv *testutil.KeyValue, restartInterval int) *block {
+		Build := func(kv *testutil.KeyValue, restartInterval int) *blockTesting {
 			// Building the block.
 			bw := &blockWriter{
 				restartInterval: restartInterval,
@@ -39,11 +44,13 @@ var _ = testutil.Defer(func() {
 			// Opening the block.
 			data := bw.buf.Bytes()
 			restartsLen := int(binary.LittleEndian.Uint32(data[len(data)-4:]))
-			return &block{
-				tr:             &Reader{cmp: comparer.DefaultComparer},
-				data:           data,
-				restartsLen:    restartsLen,
-				restartsOffset: len(data) - (restartsLen+1)*4,
+			return &blockTesting{
+				tr: &Reader{cmp: comparer.DefaultComparer},
+				b: &block{
+					data:           data,
+					restartsLen:    restartsLen,
+					restartsOffset: len(data) - (restartsLen+1)*4,
+				},
 			}
 		}
 
@@ -102,11 +109,11 @@ var _ = testutil.Defer(func() {
 			for restartInterval := 1; restartInterval <= 5; restartInterval++ {
 				Describe(fmt.Sprintf("with restart interval of %d", restartInterval), func() {
 					// Make block.
-					br := Build(kv, restartInterval)
+					bt := Build(kv, restartInterval)
 
 					Test := func(r *util.Range) func(done Done) {
 						return func(done Done) {
-							iter := br.newIterator(r, false, nil)
+							iter := bt.TestNewIterator(r)
 							Expect(iter.Error()).ShouldNot(HaveOccurred())
 
 							t := testutil.IteratorTesting{
@@ -115,6 +122,7 @@ var _ = testutil.Defer(func() {
 							}
 
 							testutil.DoIteratorTesting(&t)
+							iter.Release()
 							done <- true
 						}
 					}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/table/reader.go

@@ -8,29 +8,41 @@ package table
 
 import (
 	"encoding/binary"
-	"errors"
 	"fmt"
 	"io"
 	"sort"
 	"strings"
 	"sync"
 
-	"code.google.com/p/snappy-go/snappy"
+	"github.com/syndtr/gosnappy/snappy"
 
 	"github.com/syndtr/goleveldb/leveldb/cache"
 	"github.com/syndtr/goleveldb/leveldb/comparer"
+	"github.com/syndtr/goleveldb/leveldb/errors"
 	"github.com/syndtr/goleveldb/leveldb/filter"
 	"github.com/syndtr/goleveldb/leveldb/iterator"
 	"github.com/syndtr/goleveldb/leveldb/opt"
+	"github.com/syndtr/goleveldb/leveldb/storage"
 	"github.com/syndtr/goleveldb/leveldb/util"
 )
 
 var (
-	ErrNotFound       = util.ErrNotFound
+	ErrNotFound       = errors.ErrNotFound
 	ErrReaderReleased = errors.New("leveldb/table: reader released")
 	ErrIterReleased   = errors.New("leveldb/table: iterator released")
 )
 
+type ErrCorrupted struct {
+	Pos    int64
+	Size   int64
+	Kind   string
+	Reason string
+}
+
+func (e *ErrCorrupted) Error() string {
+	return fmt.Sprintf("leveldb/table: corruption on %s (pos=%d): %s", e.Kind, e.Pos, e.Reason)
+}
+
 func max(x, y int) int {
 	if x > y {
 		return x
@@ -39,22 +51,21 @@ func max(x, y int) int {
 }
 
 type block struct {
-	tr             *Reader
+	bpool          *util.BufferPool
+	bh             blockHandle
 	data           []byte
 	restartsLen    int
 	restartsOffset int
-	// Whether checksum is verified and valid.
-	checksum bool
 }
 
-func (b *block) seek(rstart, rlimit int, key []byte) (index, offset int, err error) {
+func (b *block) seek(cmp comparer.Comparer, rstart, rlimit int, key []byte) (index, offset int, err error) {
 	index = sort.Search(b.restartsLen-rstart-(b.restartsLen-rlimit), func(i int) bool {
 		offset := int(binary.LittleEndian.Uint32(b.data[b.restartsOffset+4*(rstart+i):]))
 		offset += 1                                 // shared always zero, since this is a restart point
 		v1, n1 := binary.Uvarint(b.data[offset:])   // key length
 		_, n2 := binary.Uvarint(b.data[offset+n1:]) // value length
 		m := offset + n1 + n2
-		return b.tr.cmp.Compare(b.data[m:m+int(v1)], key) > 0
+		return cmp.Compare(b.data[m:m+int(v1)], key) > 0
 	}) + rstart - 1
 	if index < rstart {
 		// The smallest key is greater-than key sought.
@@ -77,7 +88,7 @@ func (b *block) restartOffset(index int) int {
 func (b *block) entry(offset int) (key, value []byte, nShared, n int, err error) {
 	if offset >= b.restartsOffset {
 		if offset != b.restartsOffset {
-			err = errors.New("leveldb/table: Reader: BlockEntry: invalid block (block entries offset not aligned)")
+			err = &ErrCorrupted{Reason: "entries offset not aligned"}
 		}
 		return
 	}
@@ -87,7 +98,7 @@ func (b *block) entry(offset int) (key, value []byte, nShared, n int, err error)
 	m := n0 + n1 + n2
 	n = m + int(v1) + int(v2)
 	if n0 <= 0 || n1 <= 0 || n2 <= 0 || offset+n > b.restartsOffset {
-		err = errors.New("leveldb/table: Reader: invalid block (block entries corrupted)")
+		err = &ErrCorrupted{Reason: "entries corrupted"}
 		return
 	}
 	key = b.data[offset+m : offset+m+int(v1)]
@@ -96,48 +107,9 @@ func (b *block) entry(offset int) (key, value []byte, nShared, n int, err error)
 	return
 }
 
-func (b *block) newIterator(slice *util.Range, inclLimit bool, cache util.Releaser) *blockIter {
-	bi := &blockIter{
-		block: b,
-		cache: cache,
-		// Valid key should never be nil.
-		key:             make([]byte, 0),
-		dir:             dirSOI,
-		riStart:         0,
-		riLimit:         b.restartsLen,
-		offsetStart:     0,
-		offsetRealStart: 0,
-		offsetLimit:     b.restartsOffset,
-	}
-	if slice != nil {
-		if slice.Start != nil {
-			if bi.Seek(slice.Start) {
-				bi.riStart = b.restartIndex(bi.restartIndex, b.restartsLen, bi.prevOffset)
-				bi.offsetStart = b.restartOffset(bi.riStart)
-				bi.offsetRealStart = bi.prevOffset
-			} else {
-				bi.riStart = b.restartsLen
-				bi.offsetStart = b.restartsOffset
-				bi.offsetRealStart = b.restartsOffset
-			}
-		}
-		if slice.Limit != nil {
-			if bi.Seek(slice.Limit) && (!inclLimit || bi.Next()) {
-				bi.offsetLimit = bi.prevOffset
-				bi.riLimit = bi.restartIndex + 1
-			}
-		}
-		bi.reset()
-		if bi.offsetStart > bi.offsetLimit {
-			bi.sErr(errors.New("leveldb/table: Reader: invalid slice range"))
-		}
-	}
-	return bi
-}
-
 func (b *block) Release() {
-	b.tr.bpool.Put(b.data)
-	b.tr = nil
+	b.bpool.Put(b.data)
+	b.bpool = nil
 	b.data = nil
 }
 
@@ -152,10 +124,12 @@ const (
 )
 
 type blockIter struct {
-	block           *block
-	cache, releaser util.Releaser
-	key, value      []byte
-	offset          int
+	tr            *Reader
+	block         *block
+	blockReleaser util.Releaser
+	releaser      util.Releaser
+	key, value    []byte
+	offset        int
 	// Previous offset, only filled by Next.
 	prevOffset   int
 	prevNode     []int
@@ -252,7 +226,7 @@ func (i *blockIter) Seek(key []byte) bool {
 		return false
 	}
 
-	ri, offset, err := i.block.seek(i.riStart, i.riLimit, key)
+	ri, offset, err := i.block.seek(i.tr.cmp, i.riStart, i.riLimit, key)
 	if err != nil {
 		i.sErr(err)
 		return false
@@ -263,7 +237,7 @@ func (i *blockIter) Seek(key []byte) bool {
 		i.dir = dirForward
 	}
 	for i.Next() {
-		if i.block.tr.cmp.Compare(i.key, key) >= 0 {
+		if i.tr.cmp.Compare(i.key, key) >= 0 {
 			return true
 		}
 	}
@@ -288,7 +262,7 @@ func (i *blockIter) Next() bool {
 	for i.offset < i.offsetRealStart {
 		key, value, nShared, n, err := i.block.entry(i.offset)
 		if err != nil {
-			i.sErr(err)
+			i.sErr(i.tr.fixErrCorruptedBH(i.block.bh, err))
 			return false
 		}
 		if n == 0 {
@@ -302,13 +276,13 @@ func (i *blockIter) Next() bool {
 	if i.offset >= i.offsetLimit {
 		i.dir = dirEOI
 		if i.offset != i.offsetLimit {
-			i.sErr(errors.New("leveldb/table: Reader: Next: invalid block (block entries offset not aligned)"))
+			i.sErr(i.tr.newErrCorruptedBH(i.block.bh, "entries offset not aligned"))
 		}
 		return false
 	}
 	key, value, nShared, n, err := i.block.entry(i.offset)
 	if err != nil {
-		i.sErr(err)
+		i.sErr(i.tr.fixErrCorruptedBH(i.block.bh, err))
 		return false
 	}
 	if n == 0 {
@@ -393,7 +367,7 @@ func (i *blockIter) Prev() bool {
 	for {
 		key, value, nShared, n, err := i.block.entry(offset)
 		if err != nil {
-			i.sErr(err)
+			i.sErr(i.tr.fixErrCorruptedBH(i.block.bh, err))
 			return false
 		}
 		if offset >= i.offsetRealStart {
@@ -412,7 +386,7 @@ func (i *blockIter) Prev() bool {
 		// Stop if target offset reached.
 		if offset >= i.offset {
 			if offset != i.offset {
-				i.sErr(errors.New("leveldb/table: Reader: Prev: invalid block (block entries offset not aligned)"))
+				i.sErr(i.tr.newErrCorruptedBH(i.block.bh, "entries offset not aligned"))
 				return false
 			}
 
@@ -440,15 +414,16 @@ func (i *blockIter) Value() []byte {
 
 func (i *blockIter) Release() {
 	if i.dir != dirReleased {
+		i.tr = nil
 		i.block = nil
 		i.prevNode = nil
 		i.prevKeys = nil
 		i.key = nil
 		i.value = nil
 		i.dir = dirReleased
-		if i.cache != nil {
-			i.cache.Release()
-			i.cache = nil
+		if i.blockReleaser != nil {
+			i.blockReleaser.Release()
+			i.blockReleaser = nil
 		}
 		if i.releaser != nil {
 			i.releaser.Release()
@@ -476,21 +451,21 @@ func (i *blockIter) Error() error {
 }
 
 type filterBlock struct {
-	tr         *Reader
+	bpool      *util.BufferPool
 	data       []byte
 	oOffset    int
 	baseLg     uint
 	filtersNum int
 }
 
-func (b *filterBlock) contains(offset uint64, key []byte) bool {
+func (b *filterBlock) contains(filter filter.Filter, offset uint64, key []byte) bool {
 	i := int(offset >> b.baseLg)
 	if i < b.filtersNum {
 		o := b.data[b.oOffset+i*4:]
 		n := int(binary.LittleEndian.Uint32(o))
 		m := int(binary.LittleEndian.Uint32(o[4:]))
 		if n < m && m <= b.oOffset {
-			return b.tr.filter.Contains(b.data[n:m], key)
+			return filter.Contains(b.data[n:m], key)
 		} else if n == m {
 			return false
 		}
@@ -499,16 +474,16 @@ func (b *filterBlock) contains(offset uint64, key []byte) bool {
 }
 
 func (b *filterBlock) Release() {
-	b.tr.bpool.Put(b.data)
-	b.tr = nil
+	b.bpool.Put(b.data)
+	b.bpool = nil
 	b.data = nil
 }
 
 type indexIter struct {
 	*blockIter
+	tr    *Reader
 	slice *util.Range
 	// Options
-	checksum  bool
 	fillCache bool
 }
 
@@ -519,116 +494,146 @@ func (i *indexIter) Get() iterator.Iterator {
 	}
 	dataBH, n := decodeBlockHandle(value)
 	if n == 0 {
-		return iterator.NewEmptyIterator(errors.New("leveldb/table: Reader: invalid table (bad data block handle)"))
+		return iterator.NewEmptyIterator(i.tr.newErrCorruptedBH(i.tr.indexBH, "bad data block handle"))
 	}
 
 	var slice *util.Range
 	if i.slice != nil && (i.blockIter.isFirst() || i.blockIter.isLast()) {
 		slice = i.slice
 	}
-	return i.blockIter.block.tr.getDataIterErr(dataBH, slice, i.checksum, i.fillCache)
+	return i.tr.getDataIterErr(dataBH, slice, i.tr.verifyChecksum, i.fillCache)
 }
 
 // Reader is a table reader.
 type Reader struct {
 	mu     sync.RWMutex
+	fi     *storage.FileInfo
 	reader io.ReaderAt
-	cache  cache.Namespace
+	cache  *cache.CacheGetter
 	err    error
 	bpool  *util.BufferPool
 	// Options
-	cmp        comparer.Comparer
-	filter     filter.Filter
-	checksum   bool
-	strictIter bool
+	o              *opt.Options
+	cmp            comparer.Comparer
+	filter         filter.Filter
+	verifyChecksum bool
 
-	dataEnd           int64
-	indexBH, filterBH blockHandle
-	indexBlock        *block
-	filterBlock       *filterBlock
+	dataEnd                   int64
+	metaBH, indexBH, filterBH blockHandle
+	indexBlock                *block
+	filterBlock               *filterBlock
 }
 
-func verifyChecksum(data []byte) bool {
-	n := len(data) - 4
-	checksum0 := binary.LittleEndian.Uint32(data[n:])
-	checksum1 := util.NewCRC(data[:n]).Value()
-	return checksum0 == checksum1
+func (r *Reader) blockKind(bh blockHandle) string {
+	switch bh.offset {
+	case r.metaBH.offset:
+		return "meta-block"
+	case r.indexBH.offset:
+		return "index-block"
+	case r.filterBH.offset:
+		if r.filterBH.length > 0 {
+			return "filter-block"
+		}
+	}
+	return "data-block"
 }
 
-func (r *Reader) readRawBlock(bh blockHandle, checksum bool) ([]byte, error) {
+func (r *Reader) newErrCorrupted(pos, size int64, kind, reason string) error {
+	return &errors.ErrCorrupted{File: r.fi, Err: &ErrCorrupted{Pos: pos, Size: size, Kind: kind, Reason: reason}}
+}
+
+func (r *Reader) newErrCorruptedBH(bh blockHandle, reason string) error {
+	return r.newErrCorrupted(int64(bh.offset), int64(bh.length), r.blockKind(bh), reason)
+}
+
+func (r *Reader) fixErrCorruptedBH(bh blockHandle, err error) error {
+	if cerr, ok := err.(*ErrCorrupted); ok {
+		cerr.Pos = int64(bh.offset)
+		cerr.Size = int64(bh.length)
+		cerr.Kind = r.blockKind(bh)
+		return &errors.ErrCorrupted{File: r.fi, Err: cerr}
+	}
+	return err
+}
+
+func (r *Reader) readRawBlock(bh blockHandle, verifyChecksum bool) ([]byte, error) {
 	data := r.bpool.Get(int(bh.length + blockTrailerLen))
 	if _, err := r.reader.ReadAt(data, int64(bh.offset)); err != nil && err != io.EOF {
 		return nil, err
 	}
-	if checksum || r.checksum {
-		if !verifyChecksum(data) {
+
+	if verifyChecksum {
+		n := bh.length + 1
+		checksum0 := binary.LittleEndian.Uint32(data[n:])
+		checksum1 := util.NewCRC(data[:n]).Value()
+		if checksum0 != checksum1 {
 			r.bpool.Put(data)
-			return nil, errors.New("leveldb/table: Reader: invalid block (checksum mismatch)")
+			return nil, r.newErrCorruptedBH(bh, fmt.Sprintf("checksum mismatch, want=%#x got=%#x", checksum0, checksum1))
 		}
 	}
+
 	switch data[bh.length] {
 	case blockTypeNoCompression:
 		data = data[:bh.length]
 	case blockTypeSnappyCompression:
 		decLen, err := snappy.DecodedLen(data[:bh.length])
 		if err != nil {
-			return nil, err
+			return nil, r.newErrCorruptedBH(bh, err.Error())
 		}
-		tmp := data
-		data, err = snappy.Decode(r.bpool.Get(decLen), tmp[:bh.length])
-		r.bpool.Put(tmp)
+		decData := r.bpool.Get(decLen)
+		decData, err = snappy.Decode(decData, data[:bh.length])
+		r.bpool.Put(data)
 		if err != nil {
-			return nil, err
+			r.bpool.Put(decData)
+			return nil, r.newErrCorruptedBH(bh, err.Error())
 		}
+		data = decData
 	default:
 		r.bpool.Put(data)
-		return nil, fmt.Errorf("leveldb/table: Reader: unknown block compression type: %d", data[bh.length])
+		return nil, r.newErrCorruptedBH(bh, fmt.Sprintf("unknown compression type %#x", data[bh.length]))
 	}
 	return data, nil
 }
 
-func (r *Reader) readBlock(bh blockHandle, checksum bool) (*block, error) {
-	data, err := r.readRawBlock(bh, checksum)
+func (r *Reader) readBlock(bh blockHandle, verifyChecksum bool) (*block, error) {
+	data, err := r.readRawBlock(bh, verifyChecksum)
 	if err != nil {
 		return nil, err
 	}
 	restartsLen := int(binary.LittleEndian.Uint32(data[len(data)-4:]))
 	b := &block{
-		tr:             r,
+		bpool:          r.bpool,
+		bh:             bh,
 		data:           data,
 		restartsLen:    restartsLen,
 		restartsOffset: len(data) - (restartsLen+1)*4,
-		checksum:       checksum || r.checksum,
 	}
 	return b, nil
 }
 
-func (r *Reader) readBlockCached(bh blockHandle, checksum, fillCache bool) (*block, util.Releaser, error) {
+func (r *Reader) readBlockCached(bh blockHandle, verifyChecksum, fillCache bool) (*block, util.Releaser, error) {
 	if r.cache != nil {
-		var err error
-		ch := r.cache.Get(bh.offset, func() (charge int, value interface{}) {
-			if !fillCache {
-				return 0, nil
-			}
-			var b *block
-			b, err = r.readBlock(bh, checksum)
-			if err != nil {
-				return 0, nil
-			}
-			return cap(b.data), b
-		})
+		var (
+			err error
+			ch  *cache.Handle
+		)
+		if fillCache {
+			ch = r.cache.Get(bh.offset, func() (size int, value cache.Value) {
+				var b *block
+				b, err = r.readBlock(bh, verifyChecksum)
+				if err != nil {
+					return 0, nil
+				}
+				return cap(b.data), b
+			})
+		} else {
+			ch = r.cache.Get(bh.offset, nil)
+		}
 		if ch != nil {
 			b, ok := ch.Value().(*block)
 			if !ok {
 				ch.Release()
-				return nil, nil, errors.New("leveldb/table: Reader: inconsistent block type")
-			}
-			if !b.checksum && (r.checksum || checksum) {
-				if !verifyChecksum(b.data) {
-					ch.Release()
-					return nil, nil, errors.New("leveldb/table: Reader: invalid block (checksum mismatch)")
-				}
-				b.checksum = true
+				return nil, nil, errors.New("leveldb/table: inconsistent block type")
 			}
 			return b, ch, err
 		} else if err != nil {
@@ -636,7 +641,7 @@ func (r *Reader) readBlockCached(bh blockHandle, checksum, fillCache bool) (*blo
 		}
 	}
 
-	b, err := r.readBlock(bh, checksum)
+	b, err := r.readBlock(bh, verifyChecksum)
 	return b, b, err
 }
 
@@ -647,15 +652,15 @@ func (r *Reader) readFilterBlock(bh blockHandle) (*filterBlock, error) {
 	}
 	n := len(data)
 	if n < 5 {
-		return nil, errors.New("leveldb/table: Reader: invalid filter block (too short)")
+		return nil, r.newErrCorruptedBH(bh, "too short")
 	}
 	m := n - 5
 	oOffset := int(binary.LittleEndian.Uint32(data[m:]))
 	if oOffset > m {
-		return nil, errors.New("leveldb/table: Reader: invalid filter block (invalid offset)")
+		return nil, r.newErrCorruptedBH(bh, "invalid data-offsets offset")
 	}
 	b := &filterBlock{
-		tr:         r,
+		bpool:      r.bpool,
 		data:       data,
 		oOffset:    oOffset,
 		baseLg:     uint(data[n-1]),
@@ -666,23 +671,27 @@ func (r *Reader) readFilterBlock(bh blockHandle) (*filterBlock, error) {
 
 func (r *Reader) readFilterBlockCached(bh blockHandle, fillCache bool) (*filterBlock, util.Releaser, error) {
 	if r.cache != nil {
-		var err error
-		ch := r.cache.Get(bh.offset, func() (charge int, value interface{}) {
-			if !fillCache {
-				return 0, nil
-			}
-			var b *filterBlock
-			b, err = r.readFilterBlock(bh)
-			if err != nil {
-				return 0, nil
-			}
-			return cap(b.data), b
-		})
+		var (
+			err error
+			ch  *cache.Handle
+		)
+		if fillCache {
+			ch = r.cache.Get(bh.offset, func() (size int, value cache.Value) {
+				var b *filterBlock
+				b, err = r.readFilterBlock(bh)
+				if err != nil {
+					return 0, nil
+				}
+				return cap(b.data), b
+			})
+		} else {
+			ch = r.cache.Get(bh.offset, nil)
+		}
 		if ch != nil {
 			b, ok := ch.Value().(*filterBlock)
 			if !ok {
 				ch.Release()
-				return nil, nil, errors.New("leveldb/table: Reader: inconsistent block type")
+				return nil, nil, errors.New("leveldb/table: inconsistent block type")
 			}
 			return b, ch, err
 		} else if err != nil {
@@ -708,15 +717,55 @@ func (r *Reader) getFilterBlock(fillCache bool) (*filterBlock, util.Releaser, er
 	return r.filterBlock, util.NoopReleaser{}, nil
 }
 
-func (r *Reader) getDataIter(dataBH blockHandle, slice *util.Range, checksum, fillCache bool) iterator.Iterator {
-	b, rel, err := r.readBlockCached(dataBH, checksum, fillCache)
+func (r *Reader) newBlockIter(b *block, bReleaser util.Releaser, slice *util.Range, inclLimit bool) *blockIter {
+	bi := &blockIter{
+		tr:            r,
+		block:         b,
+		blockReleaser: bReleaser,
+		// Valid key should never be nil.
+		key:             make([]byte, 0),
+		dir:             dirSOI,
+		riStart:         0,
+		riLimit:         b.restartsLen,
+		offsetStart:     0,
+		offsetRealStart: 0,
+		offsetLimit:     b.restartsOffset,
+	}
+	if slice != nil {
+		if slice.Start != nil {
+			if bi.Seek(slice.Start) {
+				bi.riStart = b.restartIndex(bi.restartIndex, b.restartsLen, bi.prevOffset)
+				bi.offsetStart = b.restartOffset(bi.riStart)
+				bi.offsetRealStart = bi.prevOffset
+			} else {
+				bi.riStart = b.restartsLen
+				bi.offsetStart = b.restartsOffset
+				bi.offsetRealStart = b.restartsOffset
+			}
+		}
+		if slice.Limit != nil {
+			if bi.Seek(slice.Limit) && (!inclLimit || bi.Next()) {
+				bi.offsetLimit = bi.prevOffset
+				bi.riLimit = bi.restartIndex + 1
+			}
+		}
+		bi.reset()
+		if bi.offsetStart > bi.offsetLimit {
+			bi.sErr(errors.New("leveldb/table: invalid slice range"))
+		}
+	}
+	return bi
+}
+
+func (r *Reader) getDataIter(dataBH blockHandle, slice *util.Range, verifyChecksum, fillCache bool) iterator.Iterator {
+	b, rel, err := r.readBlockCached(dataBH, verifyChecksum, fillCache)
 	if err != nil {
 		return iterator.NewEmptyIterator(err)
 	}
-	return b.newIterator(slice, false, rel)
+	return r.newBlockIter(b, rel, slice, false)
 }
 
-func (r *Reader) getDataIterErr(dataBH blockHandle, slice *util.Range, checksum, fillCache bool) iterator.Iterator {
+func (r *Reader) getDataIterErr(dataBH blockHandle, slice *util.Range, verifyChecksum, fillCache bool) iterator.Iterator {
 	r.mu.RLock()
 	defer r.mu.RUnlock()
 
@@ -724,7 +773,7 @@ func (r *Reader) getDataIterErr(dataBH blockHandle, slice *util.Range, checksum,
 		return iterator.NewEmptyIterator(r.err)
 	}
 
-	return r.getDataIter(dataBH, slice, checksum, fillCache)
+	return r.getDataIter(dataBH, slice, verifyChecksum, fillCache)
 }
 
 // NewIterator creates an iterator from the table.
@@ -752,21 +801,15 @@ func (r *Reader) NewIterator(slice *util.Range, ro *opt.ReadOptions) iterator.It
 		return iterator.NewEmptyIterator(err)
 	}
 	index := &indexIter{
-		blockIter: indexBlock.newIterator(slice, true, rel),
+		blockIter: r.newBlockIter(indexBlock, rel, slice, true),
+		tr:        r,
 		slice:     slice,
-		checksum:  ro.GetStrict(opt.StrictBlockChecksum),
 		fillCache: !ro.GetDontFillCache(),
 	}
-	return iterator.NewIndexedIterator(index, r.strictIter || ro.GetStrict(opt.StrictIterator), true)
+	return iterator.NewIndexedIterator(index, opt.GetStrict(r.o, ro, opt.StrictReader))
 }
 
-// Find finds key/value pair whose key is greater than or equal to the
-// given key. It returns ErrNotFound if the table doesn't contain
-// such pair.
-//
-// The caller should not modify the contents of the returned slice, but
-// it is safe to modify the contents of the argument after Find returns.
-func (r *Reader) Find(key []byte, ro *opt.ReadOptions) (rkey, value []byte, err error) {
+func (r *Reader) find(key []byte, filtered bool, ro *opt.ReadOptions, noValue bool) (rkey, value []byte, err error) {
 	r.mu.RLock()
 	defer r.mu.RUnlock()
 
@@ -781,7 +824,7 @@ func (r *Reader) Find(key []byte, ro *opt.ReadOptions) (rkey, value []byte, err
 	}
 	defer rel.Release()
 
-	index := indexBlock.newIterator(nil, true, nil)
+	index := r.newBlockIter(indexBlock, nil, nil, true)
 	defer index.Release()
 	if !index.Seek(key) {
 		err = index.Error()
@@ -792,20 +835,23 @@ func (r *Reader) Find(key []byte, ro *opt.ReadOptions) (rkey, value []byte, err
 	}
 	dataBH, n := decodeBlockHandle(index.Value())
 	if n == 0 {
-		err = errors.New("leveldb/table: Reader: invalid table (bad data block handle)")
+		r.err = r.newErrCorruptedBH(r.indexBH, "bad data block handle")
 		return
 	}
-	if r.filter != nil {
-		filterBlock, rel, ferr := r.getFilterBlock(true)
+	if filtered && r.filter != nil {
+		filterBlock, frel, ferr := r.getFilterBlock(true)
 		if ferr == nil {
-			if !filterBlock.contains(dataBH.offset, key) {
-				rel.Release()
+			if !filterBlock.contains(r.filter, dataBH.offset, key) {
+				frel.Release()
 				return nil, nil, ErrNotFound
 			}
-			rel.Release()
+			frel.Release()
+		} else if !errors.IsCorrupted(ferr) {
+			err = ferr
+			return
 		}
 	}
-	data := r.getDataIter(dataBH, nil, ro.GetStrict(opt.StrictBlockChecksum), !ro.GetDontFillCache())
+	data := r.getDataIter(dataBH, nil, r.verifyChecksum, !ro.GetDontFillCache())
 	defer data.Release()
 	if !data.Seek(key) {
 		err = data.Error()
@@ -816,21 +862,52 @@ func (r *Reader) Find(key []byte, ro *opt.ReadOptions) (rkey, value []byte, err
 	}
 	// Don't use block buffer, no need to copy the buffer.
 	rkey = data.Key()
-	if r.bpool == nil {
-		value = data.Value()
-	} else {
-		// Use block buffer, and since the buffer will be recycled, the buffer
-		// need to be copied.
-		value = append([]byte{}, data.Value()...)
+	if !noValue {
+		if r.bpool == nil {
+			value = data.Value()
+		} else {
+			// Use block buffer, and since the buffer will be recycled, the buffer
+			// need to be copied.
+			value = append([]byte{}, data.Value()...)
+		}
 	}
 	return
 }
 
+// Find finds key/value pair whose key is greater than or equal to the
+// given key. It returns ErrNotFound if the table doesn't contain
+// such pair.
+// If filtered is true then the nearest 'block' will be checked against
+// 'filter data' (if present) and will immediately return ErrNotFound if
+// 'filter data' indicates that such pair doesn't exist.
+//
+// The caller may modify the contents of the returned slice as it is its
+// own copy.
+// It is safe to modify the contents of the argument after Find returns.
+func (r *Reader) Find(key []byte, filtered bool, ro *opt.ReadOptions) (rkey, value []byte, err error) {
+	return r.find(key, filtered, ro, false)
+}
+
+// Find finds key that is greater than or equal to the given key.
+// It returns ErrNotFound if the table doesn't contain such key.
+// If filtered is true then the nearest 'block' will be checked against
+// 'filter data' (if present) and will immediately return ErrNotFound if
+// 'filter data' indicates that such key doesn't exist.
+//
+// The caller may modify the contents of the returned slice as it is its
+// own copy.
+// It is safe to modify the contents of the argument after Find returns.
+func (r *Reader) FindKey(key []byte, filtered bool, ro *opt.ReadOptions) (rkey []byte, err error) {
+	rkey, _, err = r.find(key, filtered, ro, true)
+	return
+}
+
 // Get gets the value for the given key. It returns errors.ErrNotFound
 // if the table does not contain the key.
 //
-// The caller should not modify the contents of the returned slice, but
-// it is safe to modify the contents of the argument after Get returns.
+// The caller may modify the contents of the returned slice as it is its
+// own copy.
+// It is safe to modify the contents of the argument after Find returns.
 func (r *Reader) Get(key []byte, ro *opt.ReadOptions) (value []byte, err error) {
 	r.mu.RLock()
 	defer r.mu.RUnlock()
@@ -840,7 +917,7 @@ func (r *Reader) Get(key []byte, ro *opt.ReadOptions) (value []byte, err error)
 		return
 	}
 
-	rkey, value, err := r.Find(key, ro)
+	rkey, value, err := r.find(key, false, ro, false)
 	if err == nil && r.cmp.Compare(rkey, key) != 0 {
 		value = nil
 		err = ErrNotFound
@@ -866,12 +943,12 @@ func (r *Reader) OffsetOf(key []byte) (offset int64, err error) {
 	}
 	defer rel.Release()
 
-	index := indexBlock.newIterator(nil, true, nil)
+	index := r.newBlockIter(indexBlock, nil, nil, true)
 	defer index.Release()
 	if index.Seek(key) {
 		dataBH, n := decodeBlockHandle(index.Value())
 		if n == 0 {
-			err = errors.New("leveldb/table: Reader: invalid table (bad data block handle)")
+			r.err = r.newErrCorruptedBH(r.indexBH, "bad data block handle")
 			return
 		}
 		offset = int64(dataBH.offset)
@@ -908,55 +985,70 @@ func (r *Reader) Release() {
 }
 
 // NewReader creates a new initialized table reader for the file.
-// The cache and bpool is optional and can be nil.
+// The fi, cache and bpool is optional and can be nil.
 //
 // The returned table reader instance is goroutine-safe.
-func NewReader(f io.ReaderAt, size int64, cache cache.Namespace, bpool *util.BufferPool, o *opt.Options) *Reader {
-	r := &Reader{
-		reader:     f,
-		cache:      cache,
-		bpool:      bpool,
-		cmp:        o.GetComparer(),
-		checksum:   o.GetStrict(opt.StrictBlockChecksum),
-		strictIter: o.GetStrict(opt.StrictIterator),
-	}
+func NewReader(f io.ReaderAt, size int64, fi *storage.FileInfo, cache *cache.CacheGetter, bpool *util.BufferPool, o *opt.Options) (*Reader, error) {
 	if f == nil {
-		r.err = errors.New("leveldb/table: Reader: nil file")
-		return r
+		return nil, errors.New("leveldb/table: nil file")
 	}
+
+	r := &Reader{
+		fi:             fi,
+		reader:         f,
+		cache:          cache,
+		bpool:          bpool,
+		o:              o,
+		cmp:            o.GetComparer(),
+		verifyChecksum: o.GetStrict(opt.StrictBlockChecksum),
+	}
+
 	if size < footerLen {
-		r.err = errors.New("leveldb/table: Reader: invalid table (file size is too small)")
-		return r
+		r.err = r.newErrCorrupted(0, size, "table", "too small")
+		return r, nil
 	}
+
+	footerPos := size - footerLen
 	var footer [footerLen]byte
-	if _, err := r.reader.ReadAt(footer[:], size-footerLen); err != nil && err != io.EOF {
-		r.err = fmt.Errorf("leveldb/table: Reader: invalid table (could not read footer): %v", err)
+	if _, err := r.reader.ReadAt(footer[:], footerPos); err != nil && err != io.EOF {
+		return nil, err
 	}
 	if string(footer[footerLen-len(magic):footerLen]) != magic {
-		r.err = errors.New("leveldb/table: Reader: invalid table (bad magic number)")
-		return r
+		r.err = r.newErrCorrupted(footerPos, footerLen, "table-footer", "bad magic number")
+		return r, nil
 	}
+
+	var n int
 	// Decode the metaindex block handle.
-	metaBH, n := decodeBlockHandle(footer[:])
+	r.metaBH, n = decodeBlockHandle(footer[:])
 	if n == 0 {
-		r.err = errors.New("leveldb/table: Reader: invalid table (bad metaindex block handle)")
-		return r
+		r.err = r.newErrCorrupted(footerPos, footerLen, "table-footer", "bad metaindex block handle")
+		return r, nil
 	}
+
 	// Decode the index block handle.
 	r.indexBH, n = decodeBlockHandle(footer[n:])
 	if n == 0 {
-		r.err = errors.New("leveldb/table: Reader: invalid table (bad index block handle)")
-		return r
+		r.err = r.newErrCorrupted(footerPos, footerLen, "table-footer", "bad index block handle")
+		return r, nil
 	}
+
 	// Read metaindex block.
-	metaBlock, err := r.readBlock(metaBH, true)
+	metaBlock, err := r.readBlock(r.metaBH, true)
 	if err != nil {
-		r.err = err
-		return r
+		if errors.IsCorrupted(err) {
+			r.err = err
+			return r, nil
+		} else {
+			return nil, err
+		}
 	}
+
 	// Set data end.
-	r.dataEnd = int64(metaBH.offset)
-	metaIter := metaBlock.newIterator(nil, false, nil)
+	r.dataEnd = int64(r.metaBH.offset)
+
+	// Read metaindex.
+	metaIter := r.newBlockIter(metaBlock, nil, nil, true)
 	for metaIter.Next() {
 		key := string(metaIter.Key())
 		if !strings.HasPrefix(key, "filter.") {
@@ -989,19 +1081,27 @@ func NewReader(f io.ReaderAt, size int64, cache cache.Namespace, bpool *util.Buf
 
 	// Cache index and filter block locally, since we don't have global cache.
 	if cache == nil {
-		r.indexBlock, r.err = r.readBlock(r.indexBH, true)
-		if r.err != nil {
-			return r
+		r.indexBlock, err = r.readBlock(r.indexBH, true)
+		if err != nil {
+			if errors.IsCorrupted(err) {
+				r.err = err
+				return r, nil
+			} else {
+				return nil, err
+			}
 		}
 		if r.filter != nil {
 			r.filterBlock, err = r.readFilterBlock(r.filterBH)
 			if err != nil {
+				if !errors.IsCorrupted(err) {
+					return nil, err
+				}
+
 				// Don't use filter then.
 				r.filter = nil
-				r.filterBH = blockHandle{}
 			}
 		}
 	}
 
-	return r
+	return r, nil
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/table/table.go

@@ -133,9 +133,9 @@ Filter block trailer:
 
       +- 4-bytes -+
      /             \
-    +---------------+---------------+---------------+-------------------------+------------------+
-    |    offset 1   |      ....     |    offset n   | filter offset (4-bytes) | base Lg (1-byte) |
-    +-------------- +---------------+---------------+-------------------------+------------------+
+    +---------------+---------------+---------------+-------------------------------+------------------+
+    | data 1 offset |      ....     | data n offset | data-offsets offset (4-bytes) | base Lg (1-byte) |
+    +-------------- +---------------+---------------+-------------------------------+------------------+
 
 
 NOTE: All fixed-length integer are little-endian.

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/table/table_suite_test.go

@@ -3,15 +3,9 @@ package table
 import (
 	"testing"
 
-	. "github.com/onsi/ginkgo"
-	. "github.com/onsi/gomega"
-
 	"github.com/syndtr/goleveldb/leveldb/testutil"
 )
 
 func TestTable(t *testing.T) {
-	testutil.RunDefer()
-
-	RegisterFailHandler(Fail)
-	RunSpecs(t, "Table Suite")
+	testutil.RunSuite(t, "Table Suite")
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/table/table_test.go

@@ -23,7 +23,7 @@ type tableWrapper struct {
 }
 
 func (t tableWrapper) TestFind(key []byte) (rkey, rvalue []byte, err error) {
-	return t.Reader.Find(key, nil)
+	return t.Reader.Find(key, false, nil)
 }
 
 func (t tableWrapper) TestGet(key []byte) (value []byte, err error) {
@@ -59,7 +59,8 @@ var _ = testutil.Defer(func() {
 			It("Should be able to approximate offset of a key correctly", func() {
 				Expect(err).ShouldNot(HaveOccurred())
 
-				tr := NewReader(bytes.NewReader(buf.Bytes()), int64(buf.Len()), nil, nil, o)
+				tr, err := NewReader(bytes.NewReader(buf.Bytes()), int64(buf.Len()), nil, nil, nil, o)
+				Expect(err).ShouldNot(HaveOccurred())
 				CheckOffset := func(key string, expect, threshold int) {
 					offset, err := tr.OffsetOf([]byte(key))
 					Expect(err).ShouldNot(HaveOccurred())
@@ -95,7 +96,7 @@ var _ = testutil.Defer(func() {
 				tw.Close()
 
 				// Opening the table.
-				tr := NewReader(bytes.NewReader(buf.Bytes()), int64(buf.Len()), nil, nil, o)
+				tr, _ := NewReader(bytes.NewReader(buf.Bytes()), int64(buf.Len()), nil, nil, nil, o)
 				return tableWrapper{tr}
 			}
 			Test := func(kv *testutil.KeyValue, body func(r *Reader)) func() {

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/table/writer.go

@@ -12,7 +12,7 @@ import (
 	"fmt"
 	"io"
 
-	"code.google.com/p/snappy-go/snappy"
+	"github.com/syndtr/gosnappy/snappy"
 
 	"github.com/syndtr/goleveldb/leveldb/comparer"
 	"github.com/syndtr/goleveldb/leveldb/filter"

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/testutil/db.go

@@ -12,6 +12,7 @@ import (
 
 	. "github.com/onsi/gomega"
 
+	"github.com/syndtr/goleveldb/leveldb/errors"
 	"github.com/syndtr/goleveldb/leveldb/iterator"
 	"github.com/syndtr/goleveldb/leveldb/util"
 )
@@ -34,6 +35,10 @@ type Get interface {
 	TestGet(key []byte) (value []byte, err error)
 }
 
+type Has interface {
+	TestHas(key []byte) (ret bool, err error)
+}
+
 type NewIterator interface {
 	TestNewIterator(slice *util.Range) iterator.Iterator
 }
@@ -110,7 +115,7 @@ func (t *DBTesting) TestAllPresent() {
 
 func (t *DBTesting) TestDeletedKey(key []byte) {
 	_, err := t.DB.TestGet(key)
-	Expect(err).Should(Equal(util.ErrNotFound), "Get on deleted key %q, %s", key, t.text())
+	Expect(err).Should(Equal(errors.ErrNotFound), "Get on deleted key %q, %s", key, t.text())
 }
 
 func (t *DBTesting) TestAllDeleted() {
@@ -212,5 +217,6 @@ func DoDBTesting(t *DBTesting) {
 		}
 
 		DoIteratorTesting(&it)
+		iter.Release()
 	}
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/testutil/ginkgo.go

@@ -0,0 +1,21 @@
+package testutil
+
+import (
+	. "github.com/onsi/ginkgo"
+	. "github.com/onsi/gomega"
+)
+
+func RunSuite(t GinkgoTestingT, name string) {
+	RunDefer()
+
+	SynchronizedBeforeSuite(func() []byte {
+		RunDefer("setup")
+		return nil
+	}, func(data []byte) {})
+	SynchronizedAfterSuite(func() {
+		RunDefer("teardown")
+	}, func() {})
+
+	RegisterFailHandler(Fail)
+	RunSpecs(t, name)
+}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/testutil/kvtest.go

@@ -13,6 +13,7 @@ import (
 	. "github.com/onsi/ginkgo"
 	. "github.com/onsi/gomega"
 
+	"github.com/syndtr/goleveldb/leveldb/errors"
 	"github.com/syndtr/goleveldb/leveldb/util"
 )
 
@@ -25,9 +26,11 @@ func KeyValueTesting(rnd *rand.Rand, kv KeyValue, p DB, setup func(KeyValue) DB,
 		BeforeEach(func() {
 			p = setup(kv)
 		})
-		AfterEach(func() {
-			teardown(p)
-		})
+		if teardown != nil {
+			AfterEach(func() {
+				teardown(p)
+			})
+		}
 	}
 
 	It("Should find all keys with Find", func() {
@@ -59,7 +62,7 @@ func KeyValueTesting(rnd *rand.Rand, kv KeyValue, p DB, setup func(KeyValue) DB,
 			}
 			rkey, _, err := db.TestFind(key)
 			Expect(err).Should(HaveOccurred(), "Find for key %q yield key %q", key, rkey)
-			Expect(err).Should(Equal(util.ErrNotFound))
+			Expect(err).Should(Equal(errors.ErrNotFound))
 		}
 	})
 
@@ -77,7 +80,27 @@ func KeyValueTesting(rnd *rand.Rand, kv KeyValue, p DB, setup func(KeyValue) DB,
 				if len(key_) > 0 {
 					_, err = db.TestGet(key_)
 					Expect(err).Should(HaveOccurred(), "Error for key %q", key_)
-					Expect(err).Should(Equal(util.ErrNotFound))
+					Expect(err).Should(Equal(errors.ErrNotFound))
+				}
+			})
+		}
+	})
+
+	It("Should only find present key with Has", func() {
+		if db, ok := p.(Has); ok {
+			ShuffledIndex(nil, kv.Len(), 1, func(i int) {
+				key_, key, _ := kv.IndexInexact(i)
+
+				// Using exact key.
+				ret, err := db.TestHas(key)
+				Expect(err).ShouldNot(HaveOccurred(), "Error for key %q", key)
+				Expect(ret).Should(BeTrue(), "False for key %q", key)
+
+				// Using inexact key.
+				if len(key_) > 0 {
+					ret, err = db.TestHas(key_)
+					Expect(err).ShouldNot(HaveOccurred(), "Error for key %q", key_)
+					Expect(ret).ShouldNot(BeTrue(), "True for key %q", key)
 				}
 			})
 		}
@@ -94,6 +117,7 @@ func KeyValueTesting(rnd *rand.Rand, kv KeyValue, p DB, setup func(KeyValue) DB,
 			}
 
 			DoIteratorTesting(&t)
+			iter.Release()
 		}
 	}
 
@@ -102,7 +126,7 @@ func KeyValueTesting(rnd *rand.Rand, kv KeyValue, p DB, setup func(KeyValue) DB,
 		done <- true
 	}, 3.0)
 
-	RandomIndex(rnd, kv.Len(), kv.Len(), func(i int) {
+	RandomIndex(rnd, kv.Len(), Min(kv.Len(), 50), func(i int) {
 		type slice struct {
 			r            *util.Range
 			start, limit int
@@ -120,7 +144,7 @@ func KeyValueTesting(rnd *rand.Rand, kv KeyValue, p DB, setup func(KeyValue) DB,
 		}
 	})
 
-	RandomRange(rnd, kv.Len(), kv.Len(), func(start, limit int) {
+	RandomRange(rnd, kv.Len(), Min(kv.Len(), 50), func(start, limit int) {
 		It(fmt.Sprintf("Should iterates and seeks correctly of a slice %d .. %d", start, limit), func(done Done) {
 			r := kv.Range(start, limit)
 			TestIter(&r, kv.Slice(start, limit))
@@ -133,10 +157,22 @@ func AllKeyValueTesting(rnd *rand.Rand, body, setup func(KeyValue) DB, teardown
 	Test := func(kv *KeyValue) func() {
 		return func() {
 			var p DB
+			if setup != nil {
+				Defer("setup", func() {
+					p = setup(*kv)
+				})
+			}
+			if teardown != nil {
+				Defer("teardown", func() {
+					teardown(p)
+				})
+			}
 			if body != nil {
 				p = body(*kv)
 			}
-			KeyValueTesting(rnd, *kv, p, setup, teardown)
+			KeyValueTesting(rnd, *kv, p, func(KeyValue) DB {
+				return p
+			}, nil)
 		}
 	}
 
@@ -147,4 +183,5 @@ func AllKeyValueTesting(rnd *rand.Rand, body, setup func(KeyValue) DB, teardown
 	Describe("with big value", Test(KeyValue_BigValue()))
 	Describe("with special key", Test(KeyValue_SpecialKey()))
 	Describe("with multiple key/value", Test(KeyValue_MultipleKeyValue()))
+	Describe("with generated key/value", Test(KeyValue_Generate(nil, 120, 1, 50, 10, 120)))
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/testutil/storage.go

@@ -397,6 +397,7 @@ func (s *Storage) logI(format string, args ...interface{}) {
 
 func (s *Storage) Log(str string) {
 	s.log(1, "Log: "+str)
+	s.Storage.Log(str)
 }
 
 func (s *Storage) Lock() (r util.Releaser, err error) {

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/testutil/util.go

@@ -155,3 +155,17 @@ func RandomRange(rnd *rand.Rand, n, round int, fn func(start, limit int)) {
 	}
 	return
 }
+
+func Max(x, y int) int {
+	if x > y {
+		return x
+	}
+	return y
+}
+
+func Min(x, y int) int {
+	if x < y {
+		return x
+	}
+	return y
+}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/testutil_test.go

@@ -34,6 +34,10 @@ func (t *testingDB) TestGet(key []byte) (value []byte, err error) {
 	return t.Get(key, t.ro)
 }
 
+func (t *testingDB) TestHas(key []byte) (ret bool, err error) {
+	return t.Has(key, t.ro)
+}
+
 func (t *testingDB) TestNewIterator(slice *util.Range) iterator.Iterator {
 	return t.NewIterator(slice, t.ro)
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/util.go

@@ -14,10 +14,10 @@ import (
 )
 
 func shorten(str string) string {
-	if len(str) <= 4 {
+	if len(str) <= 8 {
 		return str
 	}
-	return str[:1] + ".." + str[len(str)-1:]
+	return str[:3] + ".." + str[len(str)-3:]
 }
 
 var bunits = [...]string{"", "Ki", "Mi", "Gi"}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/util/buffer_pool.go

@@ -8,6 +8,7 @@ package util
 
 import (
 	"fmt"
+	"sync"
 	"sync/atomic"
 	"time"
 )
@@ -19,17 +20,16 @@ type buffer struct {
 
 // BufferPool is a 'buffer pool'.
 type BufferPool struct {
-	pool       [6]chan []byte
-	size       [5]uint32
-	sizeMiss   [5]uint32
-	sizeHalf   [5]uint32
-	baseline   [4]int
-	baselinex0 int
-	baselinex1 int
-	baseline0  int
-	baseline1  int
-	baseline2  int
-	close      chan struct{}
+	pool      [6]chan []byte
+	size      [5]uint32
+	sizeMiss  [5]uint32
+	sizeHalf  [5]uint32
+	baseline  [4]int
+	baseline0 int
+
+	mu     sync.RWMutex
+	closed bool
+	closeC chan struct{}
 
 	get     uint32
 	put     uint32
@@ -58,6 +58,13 @@ func (p *BufferPool) Get(n int) []byte {
 		return make([]byte, n)
 	}
 
+	p.mu.RLock()
+	defer p.mu.RUnlock()
+
+	if p.closed {
+		return make([]byte, n)
+	}
+
 	atomic.AddUint32(&p.get, 1)
 
 	poolNum := p.poolNum(n)
@@ -153,6 +160,13 @@ func (p *BufferPool) Put(b []byte) {
 		return
 	}
 
+	p.mu.RLock()
+	defer p.mu.RUnlock()
+
+	if p.closed {
+		return
+	}
+
 	atomic.AddUint32(&p.put, 1)
 
 	pool := p.pool[p.poolNum(cap(b))]
@@ -168,10 +182,12 @@ func (p *BufferPool) Close() {
 		return
 	}
 
-	select {
-	case p.close <- struct{}{}:
-	default:
+	p.mu.Lock()
+	if !p.closed {
+		p.closed = true
+		p.closeC <- struct{}{}
 	}
+	p.mu.Unlock()
 }
 
 func (p *BufferPool) String() string {
@@ -194,7 +210,8 @@ func (p *BufferPool) drain() {
 				default:
 				}
 			}
-		case <-p.close:
+		case <-p.closeC:
+			close(p.closeC)
 			for _, ch := range p.pool {
 				close(ch)
 			}
@@ -211,7 +228,7 @@ func NewBufferPool(baseline int) *BufferPool {
 	p := &BufferPool{
 		baseline0: baseline,
 		baseline:  [...]int{baseline / 4, baseline / 2, baseline * 2, baseline * 4},
-		close:     make(chan struct{}, 1),
+		closeC:    make(chan struct{}, 1),
 	}
 	for i, cap := range []int{2, 2, 4, 4, 2, 1} {
 		p.pool[i] = make(chan []byte, cap)

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/util/util.go

@@ -12,7 +12,6 @@ import (
 )
 
 var (
-	ErrNotFound    = errors.New("leveldb: not found")
 	ErrReleased    = errors.New("leveldb: resource already relesed")
 	ErrHasReleaser = errors.New("leveldb: releaser already defined")
 )

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/version.go

@@ -7,7 +7,6 @@
 package leveldb
 
 import (
-	"errors"
 	"sync/atomic"
 	"unsafe"
 
@@ -16,19 +15,6 @@ import (
 	"github.com/syndtr/goleveldb/leveldb/util"
 )
 
-var levelMaxSize [kNumLevels]float64
-
-func init() {
-	// Precompute max size of each level
-	for level := range levelMaxSize {
-		res := float64(10 * 1048576)
-		for n := level; n > 1; n-- {
-			res *= 10
-		}
-		levelMaxSize[level] = res
-	}
-}
-
 type tSet struct {
 	level int
 	table *tFile
@@ -37,7 +23,7 @@ type tSet struct {
 type version struct {
 	s *session
 
-	tables [kNumLevels]tFiles
+	tables []tFiles
 
 	// Level that should be compacted next and its compaction score.
 	// Score < 1 means compaction is not strictly needed. These fields
@@ -47,11 +33,16 @@ type version struct {
 
 	cSeek unsafe.Pointer
 
-	ref  int
+	ref int
+	// Succeeding version.
 	next *version
 }
 
-func (v *version) release_NB() {
+func newVersion(s *session) *version {
+	return &version{s: s, tables: make([]tFiles, s.o.GetNumLevel())}
+}
+
+func (v *version) releaseNB() {
 	v.ref--
 	if v.ref > 0 {
 		return
@@ -77,13 +68,13 @@ func (v *version) release_NB() {
 		}
 	}
 
-	v.next.release_NB()
+	v.next.releaseNB()
 	v.next = nil
 }
 
 func (v *version) release() {
 	v.s.vmu.Lock()
-	v.release_NB()
+	v.releaseNB()
 	v.s.vmu.Unlock()
 }
 
@@ -123,17 +114,18 @@ func (v *version) walkOverlapping(ikey iKey, f func(level int, t *tFile) bool, l
 	}
 }
 
-func (v *version) get(ikey iKey, ro *opt.ReadOptions) (value []byte, tcomp bool, err error) {
+func (v *version) get(ikey iKey, ro *opt.ReadOptions, noValue bool) (value []byte, tcomp bool, err error) {
 	ukey := ikey.ukey()
 
 	var (
 		tset  *tSet
 		tseek bool
 
-		l0found bool
-		l0seq   uint64
-		l0vt    vType
-		l0val   []byte
+		// Level-0.
+		zfound bool
+		zseq   uint64
+		zkt    kType
+		zval   []byte
 	)
 
 	err = ErrNotFound
@@ -150,55 +142,60 @@ func (v *version) get(ikey iKey, ro *opt.ReadOptions) (value []byte, tcomp bool,
 			}
 		}
 
-		ikey__, val_, err_ := v.s.tops.find(t, ikey, ro)
-		switch err_ {
+		var (
+			fikey, fval []byte
+			ferr        error
+		)
+		if noValue {
+			fikey, ferr = v.s.tops.findKey(t, ikey, ro)
+		} else {
+			fikey, fval, ferr = v.s.tops.find(t, ikey, ro)
+		}
+		switch ferr {
 		case nil:
 		case ErrNotFound:
 			return true
 		default:
-			err = err_
+			err = ferr
 			return false
 		}
 
-		ikey_ := iKey(ikey__)
-		if seq, vt, ok := ikey_.parseNum(); ok {
-			if v.s.icmp.uCompare(ukey, ikey_.ukey()) != 0 {
-				return true
-			}
-
-			if level == 0 {
-				if seq >= l0seq {
-					l0found = true
-					l0seq = seq
-					l0vt = vt
-					l0val = val_
+		if fukey, fseq, fkt, fkerr := parseIkey(fikey); fkerr == nil {
+			if v.s.icmp.uCompare(ukey, fukey) == 0 {
+				if level == 0 {
+					if fseq >= zseq {
+						zfound = true
+						zseq = fseq
+						zkt = fkt
+						zval = fval
+					}
+				} else {
+					switch fkt {
+					case ktVal:
+						value = fval
+						err = nil
+					case ktDel:
+					default:
+						panic("leveldb: invalid iKey type")
+					}
+					return false
 				}
-			} else {
-				switch vt {
-				case tVal:
-					value = val_
-					err = nil
-				case tDel:
-				default:
-					panic("leveldb: invalid internal key type")
-				}
-				return false
 			}
 		} else {
-			err = errors.New("leveldb: internal key corrupted")
+			err = fkerr
 			return false
 		}
 
 		return true
 	}, func(level int) bool {
-		if l0found {
-			switch l0vt {
-			case tVal:
-				value = l0val
+		if zfound {
+			switch zkt {
+			case ktVal:
+				value = zval
 				err = nil
-			case tDel:
+			case ktDel:
 			default:
-				panic("leveldb: invalid internal key type")
+				panic("leveldb: invalid iKey type")
 			}
 			return false
 		}
@@ -216,13 +213,13 @@ func (v *version) getIterators(slice *util.Range, ro *opt.ReadOptions) (its []it
 		its = append(its, it)
 	}
 
-	strict := v.s.o.GetStrict(opt.StrictIterator) || ro.GetStrict(opt.StrictIterator)
+	strict := opt.GetStrict(v.s.o.Options, ro, opt.StrictReader)
 	for _, tables := range v.tables[1:] {
 		if len(tables) == 0 {
 			continue
 		}
 
-		it := iterator.NewIndexedIterator(tables.newIndexIterator(v.s.tops, v.s.icmp, slice, ro), strict, true)
+		it := iterator.NewIndexedIterator(tables.newIndexIterator(v.s.tops, v.s.icmp, slice, ro), strict)
 		its = append(its, it)
 	}
 
@@ -230,7 +227,7 @@ func (v *version) getIterators(slice *util.Range, ro *opt.ReadOptions) (its []it
 }
 
 func (v *version) newStaging() *versionStaging {
-	return &versionStaging{base: v}
+	return &versionStaging{base: v, tables: make([]tablesScratch, v.s.o.GetNumLevel())}
 }
 
 // Spawn a new version based on this version.
@@ -285,12 +282,13 @@ func (v *version) offsetOf(ikey iKey) (n uint64, err error) {
 func (v *version) pickLevel(umin, umax []byte) (level int) {
 	if !v.tables[0].overlaps(v.s.icmp, umin, umax, true) {
 		var overlaps tFiles
-		for ; level < kMaxMemCompactLevel; level++ {
+		maxLevel := v.s.o.GetMaxMemCompationLevel()
+		for ; level < maxLevel; level++ {
 			if v.tables[level+1].overlaps(v.s.icmp, umin, umax, false) {
 				break
 			}
 			overlaps = v.tables[level+2].getOverlaps(overlaps, v.s.icmp, umin, umax, false)
-			if overlaps.size() > kMaxGrandParentOverlapBytes {
+			if overlaps.size() > uint64(v.s.o.GetCompactionGPOverlaps(level)) {
 				break
 			}
 		}
@@ -318,9 +316,9 @@ func (v *version) computeCompaction() {
 			// file size is small (perhaps because of a small write-buffer
 			// setting, or very high compression ratios, or lots of
 			// overwrites/deletions).
-			score = float64(len(tables)) / kL0_CompactionTrigger
+			score = float64(len(tables)) / float64(v.s.o.GetCompactionL0Trigger())
 		} else {
-			score = float64(tables.size()) / levelMaxSize[level]
+			score = float64(tables.size()) / float64(v.s.o.GetCompactionTotalSize(level))
 		}
 
 		if score > bestScore {
@@ -337,12 +335,14 @@ func (v *version) needCompaction() bool {
 	return v.cScore >= 1 || atomic.LoadPointer(&v.cSeek) != nil
 }
 
+type tablesScratch struct {
+	added   map[uint64]atRecord
+	deleted map[uint64]struct{}
+}
+
 type versionStaging struct {
 	base   *version
-	tables [kNumLevels]struct {
-		added   map[uint64]ntRecord
-		deleted map[uint64]struct{}
-	}
+	tables []tablesScratch
 }
 
 func (p *versionStaging) commit(r *sessionRecord) {
@@ -367,7 +367,7 @@ func (p *versionStaging) commit(r *sessionRecord) {
 		tm := &(p.tables[r.level])
 
 		if tm.added == nil {
-			tm.added = make(map[uint64]ntRecord)
+			tm.added = make(map[uint64]atRecord)
 		}
 		tm.added[r.num] = r
 
@@ -379,7 +379,7 @@ func (p *versionStaging) commit(r *sessionRecord) {
 
 func (p *versionStaging) finish() *version {
 	// Build new version.
-	nv := &version{s: p.base.s}
+	nv := newVersion(p.base.s)
 	for level, tm := range p.tables {
 		btables := p.base.tables[level]
 
@@ -402,7 +402,7 @@ func (p *versionStaging) finish() *version {
 
 		// New tables.
 		for _, r := range tm.added {
-			nt = append(nt, r.makeFile(p.base.s))
+			nt = append(nt, p.base.s.tableFileFromRecord(r))
 		}
 
 		// Sort tables.
@@ -429,7 +429,7 @@ func (vr *versionReleaser) Release() {
 	v := vr.v
 	v.s.vmu.Lock()
 	if !vr.once {
-		v.release_NB()
+		v.releaseNB()
 		vr.once = true
 	}
 	v.s.vmu.Unlock()

Godeps/_workspace/src/golang.org/x/crypto/bcrypt/bcrypt.go

@@ -8,11 +8,11 @@ package bcrypt
 
 // The code is a port of Provos and Mazières's C implementation.
 import (
-	"code.google.com/p/go.crypto/blowfish"
 	"crypto/rand"
 	"crypto/subtle"
 	"errors"
 	"fmt"
+	"golang.org/x/crypto/blowfish"
 	"io"
 	"strconv"
 )

Godeps/_workspace/src/golang.org/x/text/transform/examples_test.go

@@ -8,8 +8,8 @@ import (
 	"fmt"
 	"unicode"
 
-	"code.google.com/p/go.text/transform"
-	"code.google.com/p/go.text/unicode/norm"
+	"golang.org/x/text/transform"
+	"golang.org/x/text/unicode/norm"
 )
 
 func ExampleRemoveFunc() {

Godeps/_workspace/src/golang.org/x/text/transform/transform.go

@@ -55,10 +55,17 @@ type Transformer interface {
 	// either error may be returned. Other than the error conditions listed
 	// here, implementations are free to report other errors that arise.
 	Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err error)
+
+	// Reset resets the state and allows a Transformer to be reused.
+	Reset()
 }
 
-// TODO: Do we require that a Transformer be reusable if it returns a nil error
-// or do we always require a reset after use?  Is Reset mandatory or optional?
+// NopResetter can be embedded by implementations of Transformer to add a nop
+// Reset method.
+type NopResetter struct{}
+
+// Reset implements the Reset method of the Transformer interface.
+func (NopResetter) Reset() {}
 
 // Reader wraps another io.Reader by transforming the bytes read.
 type Reader struct {
@@ -84,8 +91,9 @@ type Reader struct {
 const defaultBufSize = 4096
 
 // NewReader returns a new Reader that wraps r by transforming the bytes read
-// via t.
+// via t. It calls Reset on t.
 func NewReader(r io.Reader, t Transformer) *Reader {
+	t.Reset()
 	return &Reader{
 		r:   r,
 		t:   t,
@@ -170,8 +178,9 @@ type Writer struct {
 }
 
 // NewWriter returns a new Writer that wraps w by transforming the bytes written
-// via t.
+// via t. It calls Reset on t.
 func NewWriter(w io.Writer, t Transformer) *Writer {
+	t.Reset()
 	return &Writer{
 		w:   w,
 		t:   t,
@@ -247,7 +256,7 @@ func (w *Writer) Close() error {
 	return nil
 }
 
-type nop struct{}
+type nop struct{ NopResetter }
 
 func (nop) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err error) {
 	n := copy(dst, src)
@@ -257,7 +266,7 @@ func (nop) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err error) {
 	return n, n, err
 }
 
-type discard struct{}
+type discard struct{ NopResetter }
 
 func (discard) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err error) {
 	return 0, len(src), nil
@@ -327,6 +336,16 @@ func Chain(t ...Transformer) Transformer {
 	return c
 }
 
+// Reset resets the state of Chain. It calls Reset on all the Transformers.
+func (c *chain) Reset() {
+	for i, l := range c.link {
+		if l.t != nil {
+			l.t.Reset()
+		}
+		c.link[i].p, c.link[i].n = 0, 0
+	}
+}
+
 // Transform applies the transformers of c in sequence.
 func (c *chain) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err error) {
 	// Set up src and dst in the chain.
@@ -425,6 +444,8 @@ func RemoveFunc(f func(r rune) bool) Transformer {
 
 type removeF func(r rune) bool
 
+func (removeF) Reset() {}
+
 // Transform implements the Transformer interface.
 func (t removeF) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err error) {
 	for r, sz := rune(0), 0; len(src) > 0; src = src[sz:] {
@@ -436,7 +457,7 @@ func (t removeF) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err err
 
 			if sz == 1 {
 				// Invalid rune.
-				if !atEOF && !utf8.FullRune(src[nSrc:]) {
+				if !atEOF && !utf8.FullRune(src) {
 					err = ErrShortSrc
 					break
 				}
@@ -485,12 +506,14 @@ func grow(b []byte, n int) []byte {
 const initialBufSize = 128
 
 // String returns a string with the result of converting s[:n] using t, where
-// n <= len(s). If err == nil, n will be len(s).
+// n <= len(s). If err == nil, n will be len(s). It calls Reset on t.
 func String(t Transformer, s string) (result string, n int, err error) {
 	if s == "" {
 		return "", 0, nil
 	}
 
+	t.Reset()
+
 	// Allocate only once. Note that both dst and src escape when passed to
 	// Transform.
 	buf := [2 * initialBufSize]byte{}
@@ -571,8 +594,9 @@ func String(t Transformer, s string) (result string, n int, err error) {
 }
 
 // Bytes returns a new byte slice with the result of converting b[:n] using t,
-// where n <= len(b). If err == nil, n will be len(b).
+// where n <= len(b). If err == nil, n will be len(b). It calls Reset on t.
 func Bytes(t Transformer, b []byte) (result []byte, n int, err error) {
+	t.Reset()
 	dst := make([]byte, len(b))
 	pDst, pSrc := 0, 0
 	for {

Godeps/_workspace/src/golang.org/x/text/transform/transform_test.go

@@ -16,7 +16,7 @@ import (
 	"unicode/utf8"
 )
 
-type lowerCaseASCII struct{}
+type lowerCaseASCII struct{ NopResetter }
 
 func (lowerCaseASCII) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err error) {
 	n := len(src)
@@ -34,7 +34,7 @@ func (lowerCaseASCII) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, er
 
 var errYouMentionedX = errors.New("you mentioned X")
 
-type dontMentionX struct{}
+type dontMentionX struct{ NopResetter }
 
 func (dontMentionX) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err error) {
 	n := len(src)
@@ -52,7 +52,7 @@ func (dontMentionX) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err
 
 // doublerAtEOF is a strange Transformer that transforms "this" to "tthhiiss",
 // but only if atEOF is true.
-type doublerAtEOF struct{}
+type doublerAtEOF struct{ NopResetter }
 
 func (doublerAtEOF) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err error) {
 	if !atEOF {
@@ -71,7 +71,7 @@ func (doublerAtEOF) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err
 // rleDecode and rleEncode implement a toy run-length encoding: "aabbbbbbbbbb"
 // is encoded as "2a10b". The decoding is assumed to not contain any numbers.
 
-type rleDecode struct{}
+type rleDecode struct{ NopResetter }
 
 func (rleDecode) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err error) {
 loop:
@@ -104,6 +104,8 @@ loop:
 }
 
 type rleEncode struct {
+	NopResetter
+
 	// allowStutter means that "xxxxxxxx" can be encoded as "5x3x"
 	// instead of always as "8x".
 	allowStutter bool
@@ -136,6 +138,10 @@ func (e rleEncode) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err e
 // trickler consumes all input bytes, but writes a single byte at a time to dst.
 type trickler []byte
 
+func (t *trickler) Reset() {
+	*t = nil
+}
+
 func (t *trickler) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err error) {
 	*t = append(*t, src...)
 	if len(*t) == 0 {
@@ -157,6 +163,9 @@ func (t *trickler) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err e
 // to have some tolerance as long as progress can be detected.
 type delayedTrickler []byte
 
+func (t *delayedTrickler) Reset() {
+	*t = nil
+}
 func (t *delayedTrickler) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err error) {
 	if len(*t) > 0 && len(dst) > 0 {
 		dst[0] = (*t)[0]
@@ -447,7 +456,6 @@ var testCases = []testCase{
 
 func TestReader(t *testing.T) {
 	for _, tc := range testCases {
-		reset(tc.t)
 		r := NewReader(strings.NewReader(tc.src), tc.t)
 		// Differently sized dst and src buffers are not part of the
 		// exported API. We override them manually.
@@ -461,13 +469,6 @@ func TestReader(t *testing.T) {
 	}
 }
 
-func reset(t Transformer) {
-	var dst [128]byte
-	for err := ErrShortDst; err != nil; {
-		_, _, err = t.Transform(dst[:], nil, true)
-	}
-}
-
 func TestWriter(t *testing.T) {
 	tests := append(testCases, chainTests()...)
 	for _, tc := range tests {
@@ -477,7 +478,6 @@ func TestWriter(t *testing.T) {
 		}
 		for _, sz := range sizes {
 			bb := &bytes.Buffer{}
-			reset(tc.t)
 			w := NewWriter(bb, tc.t)
 			// Differently sized dst and src buffers are not part of the
 			// exported API. We override them manually.
@@ -735,7 +735,7 @@ func chainTests() []testCase {
 }
 
 func doTransform(tc testCase) (res string, iter int, err error) {
-	reset(tc.t)
+	tc.t.Reset()
 	dst := make([]byte, tc.dstSize)
 	out, in := make([]byte, 0, 2*len(tc.src)), []byte(tc.src)
 	for {
@@ -884,6 +884,14 @@ func TestRemoveFunc(t *testing.T) {
 			wantIter: 2,
 		},
 
+		{
+			// Test a long buffer greater than the internal buffer size
+			src:      "hello\xcc\xcc\xccworld",
+			srcSize:  13,
+			wantStr:  "hello\uFFFD\uFFFD\uFFFDworld",
+			wantIter: 1,
+		},
+
 		{
 			src:     "\u2345",
 			dstSize: 2,
@@ -947,7 +955,6 @@ func testString(t *testing.T, f func(Transformer, string) (string, int, error))
 			// that depend on a specific buffer size being set.
 			continue
 		}
-		reset(tt.t)
 		if tt.wantErr == ErrShortDst || tt.wantErr == ErrShortSrc {
 			// The result string will be different.
 			continue

Godeps/_workspace/src/golang.org/x/text/unicode/norm/Makefile

@@ -5,9 +5,6 @@
 maketables: maketables.go triegen.go
 	go build $^
 
-maketesttables: maketesttables.go triegen.go
-	go build $^
-
 normregtest: normregtest.go
 	go build $^
 
@@ -15,10 +12,6 @@ tables:	maketables
 	./maketables > tables.go
 	gofmt -w tables.go
 
-trietesttables: maketesttables
-	./maketesttables > triedata_test.go
-	gofmt -w triedata_test.go
-
 # Downloads from www.unicode.org, so not part
 # of standard test scripts.
 test: testtables regtest

Godeps/_workspace/src/golang.org/x/text/unicode/norm/example_iter_test.go

@@ -9,7 +9,7 @@ import (
 	"fmt"
 	"unicode/utf8"
 
-	"code.google.com/p/go.text/unicode/norm"
+	"golang.org/x/text/unicode/norm"
 )
 
 // EqualSimple uses a norm.Iter to compare two non-normalized

Godeps/_workspace/src/golang.org/x/text/unicode/norm/forminfo.go

@@ -201,17 +201,17 @@ func lookupInfoNFKC(b input, i int) Properties {
 // Properties returns properties for the first rune in s.
 func (f Form) Properties(s []byte) Properties {
 	if f == NFC || f == NFD {
-		return compInfo(nfcTrie.lookup(s))
+		return compInfo(nfcData.lookup(s))
 	}
-	return compInfo(nfkcTrie.lookup(s))
+	return compInfo(nfkcData.lookup(s))
 }
 
 // PropertiesString returns properties for the first rune in s.
 func (f Form) PropertiesString(s string) Properties {
 	if f == NFC || f == NFD {
-		return compInfo(nfcTrie.lookupString(s))
+		return compInfo(nfcData.lookupString(s))
 	}
-	return compInfo(nfkcTrie.lookupString(s))
+	return compInfo(nfkcData.lookupString(s))
 }
 
 // compInfo converts the information contained in v and sz

Godeps/_workspace/src/golang.org/x/text/unicode/norm/input.go

@@ -77,16 +77,16 @@ func (in *input) copySlice(buf []byte, b, e int) int {
 
 func (in *input) charinfoNFC(p int) (uint16, int) {
 	if in.bytes == nil {
-		return nfcTrie.lookupString(in.str[p:])
+		return nfcData.lookupString(in.str[p:])
 	}
-	return nfcTrie.lookup(in.bytes[p:])
+	return nfcData.lookup(in.bytes[p:])
 }
 
 func (in *input) charinfoNFKC(p int) (uint16, int) {
 	if in.bytes == nil {
-		return nfkcTrie.lookupString(in.str[p:])
+		return nfkcData.lookupString(in.str[p:])
 	}
-	return nfkcTrie.lookup(in.bytes[p:])
+	return nfkcData.lookup(in.bytes[p:])
 }
 
 func (in *input) hangul(p int) (r rune) {

Godeps/_workspace/src/golang.org/x/text/unicode/norm/iter.go

@@ -9,6 +9,8 @@ import (
 	"unicode/utf8"
 )
 
+// MaxSegmentSize is the maximum size of a byte buffer needed to consider any
+// sequence of starter and non-starter runes for the purpose of normalization.
 const MaxSegmentSize = maxByteBufferSize
 
 // An Iter iterates over a string or byte slice, while normalizing it

Godeps/_workspace/src/golang.org/x/text/unicode/norm/maketables.go

@@ -24,7 +24,8 @@ import (
 	"strings"
 	"unicode"
 
-	"code.google.com/p/go.text/internal/ucd"
+	"golang.org/x/text/internal/triegen"
+	"golang.org/x/text/internal/ucd"
 )
 
 func main() {
@@ -714,13 +715,14 @@ func printCharInfoTables() int {
 
 	varnames := []string{"nfc", "nfkc"}
 	for i := 0; i < FNumberOfFormTypes; i++ {
-		trie := newNode()
+		trie := triegen.NewTrie(varnames[i])
+
 		for r, c := range chars {
 			f := c.forms[i]
 			d := f.expandedDecomp
 			if len(d) != 0 {
 				_, key := mkstr(c.codePoint, &f)
-				trie.insert(rune(r), positionMap[key])
+				trie.Insert(rune(r), uint64(positionMap[key]))
 				if c.ccc != ccc(d[0]) {
 					// We assume the lead ccc of a decomposition !=0 in this case.
 					if ccc(d[0]) == 0 {
@@ -730,12 +732,16 @@ func printCharInfoTables() int {
 			} else if c.nLeadingNonStarters > 0 && len(f.expandedDecomp) == 0 && c.ccc == 0 && !f.combinesBackward {
 				// Handle cases where it can't be detected that the nLead should be equal
 				// to nTrail.
-				trie.insert(c.codePoint, positionMap[nLeadStr])
+				trie.Insert(c.codePoint, uint64(positionMap[nLeadStr]))
 			} else if v := makeEntry(&f, &c)<<8 | uint16(c.ccc); v != 0 {
-				trie.insert(c.codePoint, 0x8000|v)
+				trie.Insert(c.codePoint, uint64(0x8000|v))
 			}
 		}
-		size += trie.printTables(varnames[i])
+		sz, err := trie.Gen(os.Stdout, triegen.Compact(&normCompacter{name: varnames[i]}))
+		if err != nil {
+			logger.Fatal(err)
+		}
+		size += sz
 	}
 	return size
 }

Godeps/_workspace/src/golang.org/x/text/unicode/norm/normregtest.go

@@ -23,7 +23,7 @@ import (
 	"unicode"
 	"unicode/utf8"
 
-	"code.google.com/p/go.text/unicode/norm"
+	"golang.org/x/text/unicode/norm"
 )
 
 func main() {