fix: work-around workerd broken utf16-le implementation

Refs: cloudflare/workerd#6028

Author: ChALkeR

fallback/_utils.js

@@ -29,7 +29,6 @@ const getNativeLatin1 = () => {
 }
 
 export const nativeDecoderLatin1 = /* @__PURE__ */ getNativeLatin1()
-export const canDecoders = !!nativeDecoderLatin1
 
 // Block Firefox < 146 specifically from using native hex/base64, as it's very slow there
 // Refs: https://bugzilla.mozilla.org/show_bug.cgi?id=1994067 (and linked issues), fixed in 146

single-byte.js

@@ -1,5 +1,5 @@
 import { assertUint8 } from './assert.js'
-import { canDecoders, nativeEncoder, skipWeb, E_STRING } from './fallback/_utils.js'
+import { nativeDecoderLatin1, nativeEncoder, skipWeb, E_STRING } from './fallback/_utils.js'
 import { encodeAscii, encodeAsciiPrefix, encodeLatin1 } from './fallback/latin1.js'
 import { assertEncoding, encodingDecoder, encodeMap, E_STRICT } from './fallback/single-byte.js'
 
@@ -39,7 +39,7 @@ export function createSinglebyteDecoder(encoding, loose = false) {
   if (typeof loose !== 'boolean') throw new TypeError('loose option should be boolean')
   assertEncoding(encoding)
 
-  if (canDecoders && shouldUseNative(encoding)) {
+  if (nativeDecoderLatin1 && shouldUseNative(encoding)) {
     // In try, as not all encodings might be implemented in all engines which have native TextDecoder
     try {
       const decoder = new TextDecoder(encoding, { fatal: !loose })

utf16.js

@@ -1,7 +1,22 @@
 import * as js from './fallback/utf16.js'
-import { canDecoders, isLE, E_STRING } from './fallback/_utils.js'
+import { nativeDecoder, isLE, E_STRING } from './fallback/_utils.js'
 
-const { TextDecoder } = globalThis // Buffer is optional
+const { TextDecoder } = globalThis
+
+function checkDecoders() {
+  // Not all barebone engines with TextDecoder support something except utf-8
+  // Also workerd specifically has a broken utf-16le implementation
+  if (!nativeDecoder) return false
+  try {
+    const a = new TextDecoder('utf-16le').decode(Uint8Array.of(1, 2, 3, 0xd8))
+    const b = new TextDecoder('utf-16be').decode(Uint8Array.of(2, 1, 0xd8, 3))
+    return a === b && a === '\u0201\uFFFD'
+  } catch {}
+
+  return false
+}
+
+const canDecoders = checkDecoders()
 const ignoreBOM = true
 const decoderFatalLE = canDecoders ? new TextDecoder('utf-16le', { ignoreBOM, fatal: true }) : null
 const decoderLooseLE = canDecoders ? new TextDecoder('utf-16le', { ignoreBOM }) : null