Merge pull request #2318 from lightpanda-io/atob_btoa

Fix atob/btoa

.github/actions/install/action.yml

@@ -13,7 +13,7 @@ inputs:
   zig-v8:
     description: 'zig v8 version to install'
     required: false
-    default: 'v0.4.2'
+    default: 'v0.4.3'
   v8:
     description: 'v8 version to install'
     required: false

Dockerfile

@@ -3,7 +3,7 @@ FROM debian:stable-slim
 ARG MINISIG=0.12
 ARG ZIG_MINISIG=RWSGOq2NVecA2UPNdBUZykf1CCb147pkmdtYxgb3Ti+JO/wCYvhbAb/U
 ARG V8=14.0.365.4
-ARG ZIG_V8=v0.4.2
+ARG ZIG_V8=v0.4.3
 ARG TARGETPLATFORM
 
 RUN apt-get update -yq && \

build.zig.zon

@@ -5,8 +5,8 @@
     .minimum_zig_version = "0.15.2",
     .dependencies = .{
         .v8 = .{
-            .url = "https://github.com/lightpanda-io/zig-v8-fork/archive/refs/tags/v0.4.2.tar.gz",
-            .hash = "v8-0.0.0-xddH672HBABNrbtyNk9o4QXxQJTlpjiCscmdEQuMvKnR",
+            .url = "https://github.com/lightpanda-io/zig-v8-fork/archive/refs/tags/v0.4.3.tar.gz",
+            .hash = "v8-0.0.0-xddH61GNBABFJ11FJ8KDYXITyjKh4jQ54taEenYek2xJ",
         },
         // .v8 = .{ .path = "../zig-v8-fork" },
         .brotli = .{

src/browser/js/Caller.zig

@@ -846,7 +846,23 @@ fn getArgs(comptime F: type, comptime offset: usize, local: *const Local, info:
             @field(args, tupleFieldName(field_index)) = null;
         } else {
             const js_val = info.getArg(@intCast(i), local);
-            @field(args, tupleFieldName(field_index)) = local.jsValueToZig(param.type.?, js_val) catch {
+            // Only fold errors we don't recognize into InvalidArgument; let
+            // domain-meaningful ones (e.g. InvalidCharacterError from a
+            // String.OneByte param) propagate so handleError can map them
+            // to the right DOMException. Compared by name because the per-
+            // type instantiation of jsValueToZig may not include such errors
+            // in its inferred error set.
+            @field(args, tupleFieldName(field_index)) = local.jsValueToZig(param.type.?, js_val) catch |err| {
+                const DOMException = @import("../webapi/DOMException.zig");
+                if (DOMException.fromError(err) != null) {
+                    // I don't love this. But we have [a few] cases when trying to
+                    // map a JS Value that we have a specific DOMException to throw.
+                    // Ideally we should only do this if dom_exception = true in the
+                    // bridge definition. But we don't have access to that here.
+                    // Instead, we just rely on the fact that local.jsValueToZig
+                    // only throws a DOMException-known error when it should.
+                    return err;
+                }
                 return error.InvalidArgument;
             };
         }

src/browser/js/Isolate.zig

@@ -73,6 +73,10 @@ pub fn initStringHandle(self: Isolate, str: []const u8) *const v8.String {
     return v8.v8__String__NewFromUtf8(self.handle, str.ptr, v8.kNormal, @as(c_int, @intCast(str.len))).?;
 }
 
+pub fn initOneByteStringHandle(self: Isolate, bytes: []const u8) *const v8.String {
+    return v8.v8__String__NewFromOneByte(self.handle, bytes.ptr, v8.kNormal, @as(c_int, @intCast(bytes.len))).?;
+}
+
 pub fn createError(self: Isolate, msg: []const u8) *const v8.Value {
     const message = self.initStringHandle(msg);
     return v8.v8__Exception__Error(message).?;

src/browser/js/Local.zig

@@ -62,6 +62,17 @@ pub fn newString(self: *const Local, str: []const u8) js.String {
     };
 }
 
+// Creates a JS string by mapping each input byte 0..255 directly to a JS
+// code unit, with no UTF-8 decoding. Use this when handing back binary data
+// (e.g. atob output) — passing those bytes through `newString` would treat
+// any byte 0x80..0xFF as malformed UTF-8 and replace it with U+FFFD.
+pub fn newOneByteString(self: *const Local, bytes: []const u8) js.String {
+    return .{
+        .local = self,
+        .handle = self.isolate.initOneByteStringHandle(bytes),
+    };
+}
+
 pub fn newObject(self: *const Local) js.Object {
     return .{
         .local = self,
@@ -745,6 +756,15 @@ fn jsValueToStruct(self: *const Local, comptime T: type, js_val: js.Value) !?T {
             };
         },
         js.String => return js_val.isString(),
+        js.String.OneByte => {
+            // Receives a "binary string": each JS code unit must fit in a byte
+            // (0..255). Throws InvalidCharacterError if any code unit is out
+            // of range, matching the WHATWG btoa spec — which is the main
+            // intended caller, but applicable to any binary-string input.
+            const js_str = js_val.isString() orelse return null;
+            if (!js_str.containsOnlyOneByte()) return error.InvalidCharacterError;
+            return .{ .bytes = try js_str.toOneByteSlice(self.call_arena) };
+        },
         string.String => {
             const js_str = js_val.isString() orelse return null;
             return try js_str.toSSO(false);

src/browser/js/String.zig

@@ -31,6 +31,22 @@ const String = @This();
 local: *const js.Local,
 handle: *const v8.String,
 
+// A byte slice that should be handed to JS as a "binary string" — each byte
+// 0..255 becomes a JS code unit 0..255 (Latin-1), with no UTF-8 decoding.
+// Return this from a Web API method whenever the contract is "one byte per
+// JS character" (atob, FileReader.readAsBinaryString, etc.). The framework
+// turns it into a V8 string via `String::NewFromOneByte`.
+pub const OneByte = struct {
+    bytes: []const u8,
+};
+
+pub fn toValue(self: String) js.Value {
+    return .{
+        .local = self.local,
+        .handle = @ptrCast(self.handle),
+    };
+}
+
 pub fn toSlice(self: String) ![]u8 {
     return self._toSlice(false, self.local.call_arena);
 }
@@ -114,3 +130,32 @@ pub fn format(self: String, writer: *std.Io.Writer) !void {
 pub fn len(self: String) usize {
     return @intCast(v8.v8__String__Utf8Length(self.handle, self.local.isolate.handle));
 }
+
+// JS-level character (code unit) count, independent of encoding. Equivalent
+// to `s.length` in JavaScript. Use this — not `len()` — when allocating a
+// buffer for one-byte / Latin-1 reads.
+pub fn lenChars(self: String) usize {
+    return @intCast(v8.v8__String__Length(self.handle));
+}
+
+// True iff every code unit in the string fits in a single byte (codepoint
+// <= 0xFF, i.e. Latin-1). Used by btoa to reject strings with codepoints
+// outside the binary-string range.
+pub fn containsOnlyOneByte(self: String) bool {
+    return v8.v8__String__ContainsOnlyOneByte(self.handle);
+}
+
+// Read the string as Latin-1 bytes — each output byte equals the
+// corresponding code unit. Caller must have already established (via
+// `containsOnlyOneByte`) that no code unit exceeds 0xFF; otherwise V8
+// silently truncates to the low byte.
+pub fn toOneByteSlice(self: String, allocator: Allocator) ![]u8 {
+    const handle = self.handle;
+    const isolate = self.local.isolate.handle;
+    const length: u32 = @intCast(v8.v8__String__Length(handle));
+    const buf = try allocator.alloc(u8, length);
+    if (length > 0) {
+        v8.v8__String__WriteOneByte(handle, isolate, 0, length, buf.ptr);
+    }
+    return buf;
+}

src/browser/js/js.zig

@@ -234,6 +234,7 @@ pub fn simpleZigValueToJs(isolate: Isolate, value: anytype, comptime fail: bool,
         .@"struct" => {
             switch (@TypeOf(value)) {
                 string.String => return isolate.initStringHandle(value.str()),
+                String.OneByte => return @ptrCast(isolate.initOneByteStringHandle(value.bytes)),
                 ArrayBuffer => {
                     const values = value.values;
                     const len = values.len;

src/browser/tests/event/message.html

@@ -40,6 +40,15 @@
 }
 </script>
 
+<script id=messageEventPorts>
+{
+  // ports must always be a (frozen) array, never undefined.
+  const evt = new MessageEvent('message');
+  testing.expectEqual(0, evt.ports.length);
+  testing.expectEqual(true, Array.isArray(evt.ports) || typeof evt.ports.length === 'number');
+}
+</script>
+
 <script id=messageEventInheritance>
 {
   const evt4 = new MessageEvent('custom');

src/browser/tests/window/window.html

@@ -88,6 +88,73 @@
   });
 </script>
 
+<script id=atob_binary_string>
+  // atob must produce a "binary string": each output byte (0-255) becomes
+  // a JS string character whose codepoint equals that byte. High-bit bytes
+  // must NOT be folded through UTF-8 (which would replace them with U+FFFD).
+
+  // /w== decodes to a single 0xFF byte.
+  const ff = atob('/w==');
+  testing.expectEqual(1, ff.length);
+  testing.expectEqual(0xff, ff.charCodeAt(0));
+
+  // 0x00..0x07
+  const lo = atob('AAECAwQFBgc=');
+  testing.expectEqual(8, lo.length);
+  for (let i = 0; i < 8; i++) testing.expectEqual(i, lo.charCodeAt(i));
+
+  // 0xFC 0xFD 0xFE 0xFF — the bytes that previously turned into U+FFFD.
+  const hi = atob('/P3+/w==');
+  testing.expectEqual(4, hi.length);
+  testing.expectEqual(0xfc, hi.charCodeAt(0));
+  testing.expectEqual(0xfd, hi.charCodeAt(1));
+  testing.expectEqual(0xfe, hi.charCodeAt(2));
+  testing.expectEqual(0xff, hi.charCodeAt(3));
+
+  // Boundary at 0x7F/0x80.
+  const boundary = atob('f4A=');
+  testing.expectEqual(2, boundary.length);
+  testing.expectEqual(0x7f, boundary.charCodeAt(0));
+  testing.expectEqual(0x80, boundary.charCodeAt(1));
+
+  // Ensure the result is usable as input to a Uint8Array, the way recaptcha
+  // and many libraries decode binary blobs.
+  const bytes = new Uint8Array(hi.length);
+  for (let i = 0; i < hi.length; i++) bytes[i] = hi.charCodeAt(i);
+  testing.expectEqual(0xfc, bytes[0]);
+  testing.expectEqual(0xfd, bytes[1]);
+  testing.expectEqual(0xfe, bytes[2]);
+  testing.expectEqual(0xff, bytes[3]);
+</script>
+
+<script id=btoa_binary_string>
+  // btoa must accept a "binary string": each char's codepoint (0-255) is one
+  // input byte. Chars > 0xFF must throw InvalidCharacterError.
+
+  testing.expectEqual('/w==', btoa(String.fromCharCode(0xff)));
+  testing.expectEqual('/P3+/w==', btoa(String.fromCharCode(0xfc, 0xfd, 0xfe, 0xff)));
+  testing.expectEqual('AAECAwQFBgc=', btoa(String.fromCharCode(0, 1, 2, 3, 4, 5, 6, 7)));
+  testing.expectEqual('f4A=', btoa(String.fromCharCode(0x7f, 0x80)));
+
+  // Codepoints above 0xFF are not representable as a binary string.
+  testing.expectError('InvalidCharacterError', () => {
+    btoa(String.fromCharCode(0x100));
+  });
+  testing.expectError('InvalidCharacterError', () => {
+    btoa('a' + String.fromCharCode(0x1ff) + 'b');
+  });
+</script>
+
+<script id=btoa_atob_binary_roundtrip>
+  // Roundtrip every byte value 0..255.
+  let all = '';
+  for (let i = 0; i < 256; i++) all += String.fromCharCode(i);
+  const encoded = btoa(all);
+  const decoded = atob(encoded);
+  testing.expectEqual(256, decoded.length);
+  for (let i = 0; i < 256; i++) testing.expectEqual(i, decoded.charCodeAt(i));
+</script>
+
 <script id=btoa_atob_roundtrip>
   const testStrings = [
     'Hello World!',

src/browser/webapi/FileReader.zig

@@ -59,6 +59,7 @@ const ReadyState = enum(u8) {
 
 const Result = union(enum) {
     string: []const u8,
+    binary_string: js.String.OneByte,
     arraybuffer: js.ArrayBuffer,
 };
 
@@ -209,7 +210,7 @@ fn readInternal(self: *FileReader, blob: *Blob, read_type: ReadType) !void {
     // Process the data based on read type
     self._result = switch (read_type) {
         .arraybuffer => .{ .arraybuffer = .{ .values = data } },
-        .binary_string => .{ .string = data },
+        .binary_string => .{ .binary_string = .{ .bytes = data } },
         .text => .{ .string = data },
         .data_url => blk: {
             // Create data URL with base64 encoding

src/browser/webapi/Window.zig

@@ -607,12 +607,13 @@ pub fn postMessage(self: *Window, message: js.Value.Temp, target_origin: ?[]cons
     });
 }
 
-pub fn btoa(_: *const Window, input: []const u8, frame: *Frame) ![]const u8 {
-    return @import("encoding/base64.zig").encode(frame.call_arena, input);
+pub fn btoa(_: *const Window, input: js.String.OneByte, frame: *Frame) ![]const u8 {
+    return @import("encoding/base64.zig").encode(frame.call_arena, input.bytes);
 }
 
-pub fn atob(_: *const Window, input: []const u8, frame: *Frame) ![]const u8 {
-    return @import("encoding/base64.zig").decode(frame.call_arena, input);
+pub fn atob(_: *const Window, input: js.String.OneByte, frame: *Frame) !js.String.OneByte {
+    const bytes = try @import("encoding/base64.zig").decode(frame.call_arena, input.bytes);
+    return .{ .bytes = bytes };
 }
 
 pub fn structuredClone(_: *const Window, value: js.Value) !js.Value {
@@ -1052,7 +1053,7 @@ pub const JsApi = struct {
     pub const cancelIdleCallback = bridge.function(Window.cancelIdleCallback, .{});
     pub const matchMedia = bridge.function(Window.matchMedia, .{});
     pub const postMessage = bridge.function(Window.postMessage, .{});
-    pub const btoa = bridge.function(Window.btoa, .{});
+    pub const btoa = bridge.function(Window.btoa, .{ .dom_exception = true });
     pub const atob = bridge.function(Window.atob, .{ .dom_exception = true });
     pub const reportError = bridge.function(Window.reportError, .{});
     pub const structuredClone = bridge.function(Window.structuredClone, .{});

src/browser/webapi/WorkerGlobalScope.zig

@@ -264,14 +264,13 @@ pub fn receiveMessage(self: *WorkerGlobalScope, data: JS.Value) !void {
     });
 }
 
-pub fn btoa(_: *const WorkerGlobalScope, input: []const u8, exec: *JS.Execution) ![]const u8 {
-    const base64 = @import("encoding/base64.zig");
-    return base64.encode(exec.call_arena, input);
+pub fn btoa(_: *const WorkerGlobalScope, input: JS.String.OneByte, exec: *JS.Execution) ![]const u8 {
+    return @import("encoding/base64.zig").encode(exec.call_arena, input.bytes);
 }
 
-pub fn atob(_: *const WorkerGlobalScope, input: []const u8, exec: *JS.Execution) ![]const u8 {
-    const base64 = @import("encoding/base64.zig");
-    return base64.decode(exec.call_arena, input);
+pub fn atob(_: *const WorkerGlobalScope, input: JS.String.OneByte, exec: *JS.Execution) !JS.String.OneByte {
+    const bytes = try @import("encoding/base64.zig").decode(exec.call_arena, input.bytes);
+    return .{ .bytes = bytes };
 }
 
 pub fn structuredClone(_: *const WorkerGlobalScope, value: JS.Value) !JS.Value {
@@ -449,7 +448,7 @@ pub const JsApi = struct {
     pub const onrejectionhandled = bridge.accessor(WorkerGlobalScope.getOnRejectionHandled, WorkerGlobalScope.setOnRejectionHandled, .{});
     pub const onunhandledrejection = bridge.accessor(WorkerGlobalScope.getOnUnhandledRejection, WorkerGlobalScope.setOnUnhandledRejection, .{});
 
-    pub const btoa = bridge.function(WorkerGlobalScope.btoa, .{});
+    pub const btoa = bridge.function(WorkerGlobalScope.btoa, .{ .dom_exception = true });
     pub const atob = bridge.function(WorkerGlobalScope.atob, .{ .dom_exception = true });
     pub const structuredClone = bridge.function(WorkerGlobalScope.structuredClone, .{});
     pub const postMessage = bridge.function(WorkerGlobalScope.postMessage, .{});

src/browser/webapi/encoding/base64.zig

@@ -16,29 +16,27 @@
 // You should have received a copy of the GNU Affero General Public License
 // along with this program.  If not, see <https://www.gnu.org/licenses/>.
 
-//! Base64 encoding/decoding helpers for btoa/atob.
-//! Used by both Window and WorkerGlobalScope.
+//! Pure-byte base64 helpers for btoa/atob. The "binary string" semantics
+//! (each JS code unit 0..255 = one byte) are handled at the JS boundary in
+//! Window.atob / Window.btoa via the one-byte string APIs — this module
+//! just deals in bytes.
 
 const std = @import("std");
 const Allocator = std.mem.Allocator;
 
-/// Encodes input to base64 (btoa).
 pub fn encode(alloc: Allocator, input: []const u8) ![]const u8 {
     const encoded_len = std.base64.standard.Encoder.calcSize(input.len);
     const encoded = try alloc.alloc(u8, encoded_len);
     return std.base64.standard.Encoder.encode(encoded, input);
 }
 
-/// Decodes base64 input (atob).
-/// Implements forgiving base64 decode per WHATWG spec.
+/// Forgiving base64 decode per WHATWG spec:
+/// https://infra.spec.whatwg.org/#forgiving-base64-decode
 pub fn decode(alloc: Allocator, input: []const u8) ![]const u8 {
     const trimmed = std.mem.trim(u8, input, &std.ascii.whitespace);
-    // Forgiving base64 decode per WHATWG spec:
-    // https://infra.spec.whatwg.org/#forgiving-base64-decode
-    // Remove trailing padding to use standard_no_pad decoder
     const unpadded = std.mem.trimRight(u8, trimmed, "=");
 
-    // Length % 4 == 1 is invalid (can't represent valid base64)
+    // Length % 4 == 1 is invalid (can't represent valid base64).
     if (unpadded.len % 4 == 1) {
         return error.InvalidCharacterError;
     }

src/browser/webapi/event/MessageEvent.zig

@@ -23,6 +23,7 @@ const js = @import("../../js/js.zig");
 const Page = @import("../../Page.zig");
 
 const Event = @import("../Event.zig");
+const MessagePort = @import("../MessagePort.zig");
 const Window = @import("../Window.zig");
 
 const String = lp.String;
@@ -116,6 +117,10 @@ pub fn getSource(self: *const MessageEvent) ?*Window {
     return self._source;
 }
 
+pub fn getPorts(_: *const MessageEvent) []*MessagePort {
+    return &.{};
+}
+
 pub const JsApi = struct {
     pub const bridge = js.Bridge(MessageEvent);
 
@@ -129,6 +134,7 @@ pub const JsApi = struct {
     pub const data = bridge.accessor(MessageEvent.getData, null, .{});
     pub const origin = bridge.accessor(MessageEvent.getOrigin, null, .{});
     pub const source = bridge.accessor(MessageEvent.getSource, null, .{});
+    pub const ports = bridge.accessor(MessageEvent.getPorts, null, .{});
 };
 
 const testing = @import("../../../testing.zig");