lib: implement WHATWG-spec Big5 decoder in js

TextDecoder('big5') was routed through ICU's Big5 converter, whose
table includes vendor/PUA mappings the WHATWG Encoding Standard does
not define. Byte sequences the standard defines as invalid (and
which must decode to U+FFFD, or throw when fatal) instead decoded to
those extra ICU-only characters, e.g. bytes 0x83 0x5C decoded to
U+F00E instead of U+FFFD U+005C, even with `fatal: true`.

This adds a small, self-contained decoder for the 'big5' label (and
its aliases, which the standard maps to the same decoder) that
implements the algorithm and index table from the Encoding Standard
directly, mirroring how single-byte.js already reimplements the
legacy single-byte encodings instead of relying on ICU for them. It
does not touch the ICU-backed path used by any other encoding.

The decoder module is required lazily from internal/encoding.js, only
when a Big5 TextDecoder is constructed, so its large base64 index
table is not pulled into the startup snapshot; internal/encoding.js
itself is loaded during bootstrap.

Refs: https://encoding.spec.whatwg.org/#big5-decoder
Refs: #61041
Fixes: #40091
Signed-off-by: agape1225 <49804691+agape1225@users.noreply.github.com>

Author: agape1225

lib/internal/encoding.js

@@ -24,6 +24,7 @@ const {
   ERR_NO_ICU,
 } = require('internal/errors').codes;
 const kSingleByte = Symbol('single-byte');
+const kBig5 = Symbol('big5');
 const kHandle = Symbol('handle');
 const kFlags = Symbol('flags');
 const kEncoding = Symbol('encoding');
@@ -449,13 +450,24 @@ class TextDecoder {
     this[kUTF8FastPath] = false;
     this[kHandle] = undefined;
     this[kSingleByte] = undefined; // Does not care about streaming or BOM
+    this[kBig5] = false;
     this[kChunk] = null; // A copy of previous streaming tail or null
 
     if (enc === 'utf-8') {
       this[kUTF8FastPath] = true;
       this[kBOMSeen] = false;
     } else if (isSinglebyteEncoding(enc)) {
       this[kSingleByte] = createSinglebyteDecoder(enc, this[kFatal]);
+    } else if (enc === 'big5') {
+      // Not routed through ICU: ICU's own Big5 conversion table includes
+      // vendor/PUA mappings the WHATWG Encoding Standard does not, so byte
+      // sequences the standard defines as invalid would otherwise decode to
+      // those extra characters instead of U+FFFD.
+      // Loaded lazily: the module carries a large index table that should not
+      // be pulled into the startup snapshot for the common non-Big5 case.
+      const { createBig5Decoder } = require('internal/encoding/big5');
+      this[kBig5] = true;
+      this[kHandle] = createBig5Decoder(this[kFatal]);
     } else {
       this.#prepareConverter(); // Need to throw early if we don't support the encoding
     }
@@ -485,6 +497,12 @@ class TextDecoder {
     if (this[kSingleByte]) return this[kSingleByte](parseInput(input));
 
     const stream = options?.stream;
+
+    if (this[kBig5]) {
+      input = parseInput(input);
+      return stream ? this[kHandle].write(input) : this[kHandle].end(input);
+    }
+
     if (this[kUTF8FastPath]) {
       const chunk = this[kChunk];
       const ignoreBom = this[kIgnoreBOM] || this[kBOMSeen];

lib/internal/encoding/big5.js

@@ -0,0 +1,102 @@
+'use strict';
+
+// Big5 is decoded by a WHATWG-spec-faithful implementation (lib/internal/encoding/big5.js)
+// rather than through ICU, because ICU's own Big5 conversion table includes
+// vendor/PUA mappings the WHATWG Encoding Standard does not: byte sequences
+// the standard defines as invalid decode to those extra characters instead
+// of U+FFFD when routed through ICU.
+// Refs: https://github.com/nodejs/node/issues/61041
+// Refs: https://github.com/nodejs/node/issues/40091
+// Refs: https://encoding.spec.whatwg.org/#big5-decoder
+
+require('../common');
+const assert = require('assert');
+
+function codePoints(str) {
+  return [...str].map((c) => c.codePointAt(0));
+}
+
+for (const label of ['big5', 'Big5', 'BIG5', 'big5-hkscs', 'cn-big5', 'csbig5', 'x-x-big5']) {
+  const decoder = new TextDecoder(label);
+  assert.strictEqual(decoder.encoding, 'big5');
+}
+
+// ASCII round-trips as-is.
+{
+  const decoder = new TextDecoder('big5');
+  assert.strictEqual(decoder.decode(Uint8Array.from([0x41, 0x42, 0x43])), 'ABC');
+}
+
+// A valid 2-byte Big5 sequence decodes to the expected code point.
+// 0xA4 0x40 is the first Hanzi in the Big5 table: U+4E00 ("一", "one").
+{
+  const decoder = new TextDecoder('big5');
+  assert.strictEqual(decoder.decode(Uint8Array.from([0xa4, 0x40])), '一');
+}
+
+// Regression test: an unassigned Big5 pointer must decode to U+FFFD, not to
+// whatever extra character ICU's own (non-spec) Big5 table maps it to.
+// https://github.com/nodejs/node/issues/40091
+{
+  const decoder = new TextDecoder('big5');
+  const result = decoder.decode(Uint8Array.from([0x41, 0x42, 0x83, 0x5c, 0x43, 0x44]));
+  assert.deepStrictEqual(codePoints(result), [0x41, 0x42, 0xfffd, 0x5c, 0x43, 0x44]);
+}
+
+// A lead byte with no trailing byte (end of input) is also an error.
+{
+  const decoder = new TextDecoder('big5');
+  assert.deepStrictEqual(codePoints(decoder.decode(Uint8Array.from([0xa4]))), [0xfffd]);
+}
+
+// `fatal: true` must throw instead of substituting U+FFFD, and must
+// actually recognize this sequence as invalid (unlike plain ICU, which
+// treats it as valid and never throws even in fatal mode).
+{
+  const decoder = new TextDecoder('big5', { fatal: true });
+  assert.throws(() => {
+    decoder.decode(Uint8Array.from([0x83, 0x5c]));
+  }, { name: 'TypeError', code: 'ERR_ENCODING_INVALID_ENCODED_DATA' });
+}
+
+// The four Big5 pointers that map to two combining code points instead of
+// one, per the spec's special-cased steps ahead of the index table lookup.
+{
+  const decoder = new TextDecoder('big5');
+  assert.deepStrictEqual(codePoints(decoder.decode(Uint8Array.from([0x88, 0x62]))), [0xca, 0x0304]);
+  assert.deepStrictEqual(codePoints(decoder.decode(Uint8Array.from([0x88, 0x64]))), [0xca, 0x030c]);
+  assert.deepStrictEqual(codePoints(decoder.decode(Uint8Array.from([0x88, 0xa3]))), [0xea, 0x0304]);
+  assert.deepStrictEqual(codePoints(decoder.decode(Uint8Array.from([0x88, 0xa5]))), [0xea, 0x030c]);
+}
+
+// Streaming: a valid 2-byte sequence split across chunk boundaries must
+// still decode correctly, and must not be flushed early.
+{
+  const decoder = new TextDecoder('big5');
+  const r1 = decoder.decode(Uint8Array.from([0xa4]), { stream: true });
+  assert.strictEqual(r1, '');
+  const r2 = decoder.decode(Uint8Array.from([0x40]));
+  assert.strictEqual(r2, '一');
+}
+
+// A supplementary-plane code point (outside the BMP) must be encoded as a
+// surrogate pair in the resulting JS string.
+{
+  // Find a Big5 pointer whose code point is astral, by scanning the same
+  // byte ranges the decoder itself accepts.
+  const decoder = new TextDecoder('big5');
+  let found = false;
+  outer:
+  for (let lead = 0x81; lead <= 0xfe && !found; lead++) {
+    for (let byte = 0x40; byte <= 0xfe; byte++) {
+      if (!((byte >= 0x40 && byte <= 0x7e) || (byte >= 0xa1 && byte <= 0xfe))) continue;
+      const result = decoder.decode(Uint8Array.from([lead, byte]));
+      if (result.length === 2 && codePoints(result).length === 1) {
+        assert.ok(codePoints(result)[0] > 0xffff);
+        found = true;
+        break outer;
+      }
+    }
+  }
+  assert.ok(found, 'expected to find at least one astral Big5 mapping');
+}