#!/usr/bin/env python3
"""usage: gen_order.py <hot.text> <hot.rodata> <out.ld> [--v8 <libc_v8.a> <v8.txt>] <obj-or-archive>...

Builds a symbol -> (file, section) map from the objects and emits an INSERT
linker script that places the hot sections in .text.hot / .rodata.hot ahead of
.text / .rodata. Patterns are scoped to their object file (`*api.o(...)`):
LLD tests every input section against every unscoped pattern, which turns a
26k-pattern script into an 80s link; scoped, it is a few seconds.

With --v8, symbols defined in that archive are written to <v8.txt> for
mark_hot_sections.zig instead, and the script matches them with one
`.text.hot.*` / `.rodata.hot.*` glob (V8's sections are all named `.text`).
"""
import sys, subprocess, collections, re, os
hot_text, hot_rodata, out = sys.argv[1:4]
objs = sys.argv[4:]
v8_archive = v8_out = None
if objs and objs[0] == "--v8":
    v8_archive, v8_out = os.path.basename(objs[1]), objs[2]
    objs = [objs[1]] + objs[3:]
loc_of = collections.defaultdict(set)  # symbol -> {(archive, file, section)}
for o in objs:
    p = subprocess.run(["objdump", "-t", o], capture_output=True, text=True).stdout
    fname = archive = os.path.basename(o)
    for line in p.splitlines():
        m = re.match(r"^(.+?):\s+file format elf", line)
        if m:
            fname = os.path.basename(m.group(1))
            continue
        # value flags section<TAB>size name  -- the section can contain spaces (zig)
        if "\t" not in line: continue
        left, right = line.split("\t", 1)
        parts = right.split(None, 1)
        if len(parts) != 2: continue
        name = parts[1].strip()
        for vis in (".hidden ", ".protected ", ".internal "):
            if name.startswith(vis): name = name[len(vis):]
        fl = left[17:24]
        if "d" in fl: continue  # section symbol
        sec = left[24:].strip()
        if sec.startswith((".text", ".rodata")):
            loc_of[name].add((archive, fname, sec))
def esc(s):
    return re.sub(r'([*?\[\]\\])', r'\\\1', s)
stats = collections.Counter()
v8_syms = []
def emit(hotfile, prefix):
    by_file = collections.OrderedDict(); seen = set()
    for name in open(hotfile).read().split("\n"):
        if not name: continue
        locs = loc_of.get(name)
        if not locs: stats[prefix + " nomap"] += 1; continue
        for archive, fname, sec in sorted(locs):
            if not sec.startswith(prefix): continue
            if archive == v8_archive:
                # The embedded builtins blob is left cold (see
                # mark_hot_sections.zig); don't emit a pattern for it.
                if name.startswith("Builtins_"): stats[prefix + " v8-blob-skip"] += 1; continue
                v8_syms.append(name); stats[prefix + " v8"] += 1; continue
            if sec == prefix or sec.endswith("."): stats[prefix + " generic"] += 1; continue
            if '"' in sec or '"' in fname: stats[prefix + " quote-skip"] += 1; continue
            if (fname, sec) in seen: continue
            seen.add((fname, sec)); by_file.setdefault(fname, []).append(sec)
    stats[prefix + " sections"] = len(seen); stats[prefix + " files"] = len(by_file)
    # LLD unquotes section names but not the file pattern, so that one stays bare.
    lines = [f'    *{esc(f)}(\n' + "\n".join(f'      "{esc(s)}"' for s in secs) + "\n    )" for f, secs in by_file.items()]
    if v8_archive: lines.insert(0, f'    *("{prefix}.hot.*")')
    return lines
t = emit(hot_text, ".text"); r = emit(hot_rodata, ".rodata")
with open(out, "w") as f:
    f.write("/* Generated by orderfile/tools/gen_order.py, see orderfile/README.md. */\n")
    f.write("SECTIONS {\n  .text.hot : {\n" + "\n".join(t) + "\n  }\n} INSERT BEFORE .text;\n")
    f.write("SECTIONS {\n  .rodata.hot : {\n" + "\n".join(r) + "\n  }\n} INSERT BEFORE .rodata;\n")
if v8_out:
    with open(v8_out, "w") as f: f.write("\n".join(dict.fromkeys(v8_syms)) + "\n")
for k, v in sorted(stats.items()): print(f"{k}: {v}")