@exodus/bytes 1.4.0 → 1.5.0

package/fallback/encoding.js

@@ -47,6 +47,10 @@ export function normalizeEncoding(label) {
 
 const define = (obj, key, value) => Object.defineProperty(obj, key, { value, writable: false })
 
+// TODO: make this more strict against Symbol.toStringTag
+// Is not very significant though, anything faking Symbol.toStringTag could as well override
+// prototypes, which is not something we protect against
+
 function isAnyArrayBuffer(x) {
   if (x instanceof ArrayBuffer) return true
   if (globalThis.SharedArrayBuffer && x instanceof SharedArrayBuffer) return true
@@ -55,6 +59,12 @@ function isAnyArrayBuffer(x) {
   return s === '[object ArrayBuffer]' || s === '[object SharedArrayBuffer]'
 }
 
+function isAnyUint8Array(x) {
+  if (x instanceof Uint8Array) return true
+  if (!x || !ArrayBuffer.isView(x) || x.BYTES_PER_ELEMENT !== 1) return false
+  return Object.prototype.toString.call(x) === '[object Uint8Array]'
+}
+
 const fromSource = (x) => {
   if (x instanceof Uint8Array) return x
   if (ArrayBuffer.isView(x)) return new Uint8Array(x.buffer, x.byteOffset, x.byteLength)
@@ -217,7 +227,7 @@ export class TextEncoder {
 
   encodeInto(str, target) {
     if (typeof str !== 'string') str = `${str}`
-    if (!(target instanceof Uint8Array)) throw new TypeError('Target must be an Uint8Array')
+    if (!isAnyUint8Array(target)) throw new TypeError('Target must be an Uint8Array')
     if (target.buffer.detached) return { read: 0, written: 0 } // Until https://github.com/whatwg/encoding/issues/324 is resolved
 
     const tlen = target.length

package/fallback/multi-byte.js

@@ -8,37 +8,57 @@ export const E_STRICT = 'Input is not well-formed for this encoding'
 // If the decoder is not cleared properly, state can be preserved between non-streaming calls!
 // See comment about fatal stream
 
-// All except iso-2022-jp are ASCII supersets
-// When adding something that is not an ASCII superset, ajust the ASCII fast path
-const REP = 0xff_fd
-const mappers = {
-  // https://encoding.spec.whatwg.org/#euc-kr-decoder
-  'euc-kr': (err) => {
-    const euc = getTable('euc-kr')
-    let lead = 0
+// Common between euc-kr and big5
+function bigDecoder(err, pair) {
+  let lead = 0
 
-    const pushback = []
-    const bytes = (b) => {
-      if (lead) {
-        const cp = b >= 0x41 && b <= 0xfe ? euc[(lead - 0x81) * 190 + b - 0x41] : undefined
-        lead = 0
-        if (cp !== undefined && cp !== REP) return cp
-        if (b < 128) pushback.push(b)
-        return err()
-      }
+  const decodeLead = (b) => {
+    const str = pair(lead, b)
+    lead = 0
+    if (str) return str
+    return b < 128 ? String.fromCharCode(err(), b) : String.fromCharCode(err())
+  }
+
+  const decode = (arr, start, end, stream) => {
+    let res = ''
+    let i = start
 
-      if (b < 128) return b
-      if (b < 0x81 || b === 0xff) return err()
-      lead = b
+    if (lead && i < end) res += decodeLead(arr[i++])
+    while (i < end) {
+      const b = arr[i++]
+      if (b < 128) {
+        res += String.fromCharCode(b)
+      } else if (b === 0x80 || b === 0xff) {
+        res += String.fromCharCode(err())
+      } else {
+        lead = b
+        if (i < end) res += decodeLead(arr[i++])
+      }
     }
 
-    const eof = () => {
-      if (!lead) return null
+    if (lead && !stream) {
       lead = 0
-      return err()
+      res += String.fromCharCode(err())
     }
 
-    return { bytes, eof, pushback }
+    return res
+  }
+
+  return { decode, isAscii: () => lead === 0 }
+}
+
+// All except iso-2022-jp are ASCII supersets
+// When adding something that is not an ASCII superset, ajust the ASCII fast path
+const REP = 0xff_fd
+const mappers = {
+  // https://encoding.spec.whatwg.org/#euc-kr-decoder
+  'euc-kr': (err) => {
+    const euc = getTable('euc-kr')
+    return bigDecoder(err, (l, b) => {
+      if (b < 0x41 || b > 0xfe) return
+      const cp = euc[(l - 0x81) * 190 + b - 0x41]
+      return cp !== undefined && cp !== REP ? String.fromCharCode(cp) : undefined
+    })
   },
   // https://encoding.spec.whatwg.org/#euc-jp-decoder
   'euc-jp': (err) => {
@@ -47,64 +67,71 @@ const mappers = {
     let j12 = false
     let lead = 0
 
-    const pushback = []
-    const bytes = (b) => {
+    const decodeLead = (b) => {
       if (lead === 0x8e && b >= 0xa1 && b <= 0xdf) {
         lead = 0
-        return 0xfe_c0 + b
+        return String.fromCharCode(0xfe_c0 + b)
       }
 
       if (lead === 0x8f && b >= 0xa1 && b <= 0xfe) {
         j12 = true
         lead = b
-        return
+        return ''
       }
 
-      if (lead) {
-        let cp
-        if (lead >= 0xa1 && lead <= 0xfe && b >= 0xa1 && b <= 0xfe) {
-          cp = (j12 ? jis0212 : jis0208)[(lead - 0xa1) * 94 + b - 0xa1]
+      let cp
+      if (lead >= 0xa1 && lead <= 0xfe && b >= 0xa1 && b <= 0xfe) {
+        cp = (j12 ? jis0212 : jis0208)[(lead - 0xa1) * 94 + b - 0xa1]
+      }
+
+      lead = 0
+      j12 = false
+      if (cp !== undefined && cp !== REP) return String.fromCharCode(cp)
+      return b < 128 ? String.fromCharCode(err(), b) : String.fromCharCode(err())
+    }
+
+    const decode = (arr, start, end, stream) => {
+      let res = ''
+      let i = start
+
+      if (lead && i < end) res += decodeLead(arr[i++])
+      if (lead && i < end) res += decodeLead(arr[i++]) // could be two leads, but no more
+      while (i < end) {
+        const b = arr[i++]
+        if (b < 128) {
+          res += String.fromCharCode(b)
+        } else if ((b < 0xa1 && b !== 0x8e && b !== 0x8f) || b === 0xff) {
+          res += String.fromCharCode(err())
+        } else {
+          lead = b
+          if (i < end) res += decodeLead(arr[i++])
+          if (lead && i < end) res += decodeLead(arr[i++]) // could be two leads
         }
+      }
 
+      if (lead && !stream) {
         lead = 0
-        j12 = false
-        if (cp !== undefined && cp !== REP) return cp
-        if (b < 128) pushback.push(b)
-        return err()
+        j12 = false // can be true only when lead is non-zero
+        res += String.fromCharCode(err())
       }
 
-      if (b < 128) return b
-      if ((b < 0xa1 && b !== 0x8e && b !== 0x8f) || b === 0xff) return err()
-      lead = b
+      return res
     }
 
-    // eslint-disable-next-line sonarjs/no-identical-functions
-    const eof = () => {
-      if (!lead) return null
-      lead = 0
-      return err()
-    }
-
-    return { bytes, eof, pushback }
+    return { decode, isAscii: () => lead === 0 } // j12 can be true only when lead is non-zero
   },
   // https://encoding.spec.whatwg.org/#iso-2022-jp-decoder
-  // Per-letter of the spec, don't shortcut on state changes on EOF. Some code is regrouped but preserving the logic
   'iso-2022-jp': (err) => {
     const jis0208 = getTable('jis0208')
-    const EOF = -1
     let dState = 1
     let oState = 1
-    let lead = 0
+    let lead = 0 // 0 or 0x21-0x7e
     let out = false
 
-    const pushback = []
-    const bytes = (b) => {
-      if (dState < 5) {
-        if (b === EOF) return null
-        if (b === 0x1b) {
-          dState = 6 // escape start
-          return
-        }
+    const bytes = (pushback, b) => {
+      if (dState < 5 && b === 0x1b) {
+        dState = 6 // escape start
+        return
       }
 
       switch (dState) {
@@ -156,7 +183,7 @@ const mappers = {
 
           out = false
           dState = oState
-          if (b !== EOF) pushback.push(b)
+          pushback.push(b)
           return err()
         case 7: {
           // Escape
@@ -185,52 +212,117 @@ const mappers = {
 
           out = false
           dState = oState
-          if (b !== EOF) pushback.push(b)
-          pushback.push(l)
+          pushback.push(b, l)
+          return err()
+        }
+      }
+    }
+
+    const eof = (pushback) => {
+      if (dState < 5) return null
+      out = false
+      switch (dState) {
+        case 5:
+          dState = 4
+          return err()
+        case 6:
+          dState = oState
+          return err()
+        case 7: {
+          dState = oState
+          pushback.push(lead)
+          lead = 0
           return err()
         }
       }
     }
 
-    const eof = () => bytes(EOF)
+    const decode = (arr, start, end, stream) => {
+      let res = ''
+      let i = start
+      const pushback = [] // local and auto-cleared
+
+      // First, dump everything until EOF
+      // Same as the full loop, but without EOF handling
+      while (i < end || pushback.length > 0) {
+        const c = bytes(pushback, pushback.length > 0 ? pushback.pop() : arr[i++])
+        if (c !== undefined) res += String.fromCodePoint(c)
+      }
+
+      // Then, dump EOF. This needs the same loop as the characters can be pushed back
+      if (!stream) {
+        while (i <= end || pushback.length > 0) {
+          if (i < end || pushback.length > 0) {
+            const c = bytes(pushback, pushback.length > 0 ? pushback.pop() : arr[i++])
+            if (c !== undefined) res += String.fromCodePoint(c)
+          } else {
+            const c = eof(pushback)
+            if (c === null) break // clean exit
+            res += String.fromCodePoint(c)
+          }
+        }
+      }
+
+      // Chrome and WebKit fail on this, we don't: completely destroy the old decoder state when finished streaming
+      // > If this’s do not flush is false, then set this’s decoder to a new instance of this’s encoding’s decoder,
+      // > Set this’s do not flush to options["stream"]
+      if (!stream) {
+        dState = oState = 1
+        lead = 0
+        out = false
+      }
+
+      return res
+    }
 
-    return { bytes, eof, pushback }
+    return { decode, isAscii: () => false }
   },
   // https://encoding.spec.whatwg.org/#shift_jis-decoder
   shift_jis: (err) => {
     const jis0208 = getTable('jis0208')
     let lead = 0
 
-    const pushback = []
-    const bytes = (b) => {
-      if (lead) {
-        const l = lead
-        lead = 0
-        if (b >= 0x40 && b <= 0xfc && b !== 0x7f) {
-          const p = (l - (l < 0xa0 ? 0x81 : 0xc1)) * 188 + b - (b < 0x7f ? 0x40 : 0x41)
-          if (p >= 8836 && p <= 10_715) return 0xe0_00 - 8836 + p // 16-bit
-          const cp = jis0208[p]
-          if (cp !== undefined && cp !== REP) return cp
-        }
-
-        if (b < 128) pushback.push(b)
-        return err()
+    const decodeLead = (b) => {
+      const l = lead
+      lead = 0
+      if (b >= 0x40 && b <= 0xfc && b !== 0x7f) {
+        const p = (l - (l < 0xa0 ? 0x81 : 0xc1)) * 188 + b - (b < 0x7f ? 0x40 : 0x41)
+        if (p >= 8836 && p <= 10_715) return String.fromCharCode(0xe0_00 - 8836 + p)
+        const cp = jis0208[p]
+        if (cp !== undefined && cp !== REP) return String.fromCharCode(cp)
       }
 
-      if (b <= 0x80) return b // 0x80 is allowed
-      if (b >= 0xa1 && b <= 0xdf) return 0xff_61 - 0xa1 + b
-      if (b < 0x81 || (b > 0x9f && b < 0xe0) || b > 0xfc) return err()
-      lead = b
+      return b < 128 ? String.fromCharCode(err(), b) : String.fromCharCode(err())
     }
 
-    // eslint-disable-next-line sonarjs/no-identical-functions
-    const eof = () => {
-      if (!lead) return null
-      lead = 0 // this clears state completely on EOF
-      return err()
+    const decode = (arr, start, end, stream) => {
+      let res = ''
+      let i = start
+
+      if (lead && i < end) res += decodeLead(arr[i++])
+      while (i < end) {
+        const b = arr[i++]
+        if (b <= 0x80) {
+          res += String.fromCharCode(b) // 0x80 is allowed
+        } else if (b >= 0xa1 && b <= 0xdf) {
+          res += String.fromCharCode(0xfe_c0 + b)
+        } else if (b === 0xa0 || b > 0xfc) {
+          res += String.fromCharCode(err())
+        } else {
+          lead = b
+          if (i < end) res += decodeLead(arr[i++])
+        }
+      }
+
+      if (lead && !stream) {
+        lead = 0
+        res += String.fromCharCode(err())
+      }
+
+      return res
     }
 
-    return { bytes, eof, pushback }
+    return { decode, isAscii: () => lead === 0 }
   },
   // https://encoding.spec.whatwg.org/#gbk-decoder
   gbk: (err) => mappers.gb18030(err), // 10.1.1. GBK’s decoder is gb18030’s decoder
@@ -252,179 +344,120 @@ const mappers = {
       return b + p - a
     }
 
-    const pushback = []
-    const bytes = (b) => {
-      if (g3) {
-        if (b < 0x30 || b > 0x39) {
-          pushback.push(b, g3, g2)
-          g1 = g2 = g3 = 0
-          return err()
-        }
-
-        const cp = index((g1 - 0x81) * 12_600 + (g2 - 0x30) * 1260 + (g3 - 0x81) * 10 + b - 0x30)
-        g1 = g2 = g3 = 0
-        if (cp !== undefined) return cp // Can validly return replacement
-        return err()
-      }
+    // g1 is 0 or 0x81-0xfe
+    // g2 is 0 or 0x30-0x39
+    // g3 is 0 or 0x81-0xfe
+
+    const decode = (arr, start, end, stream) => {
+      let res = ''
+      let i = start
+      const pushback = [] // local and auto-cleared
+
+      // First, dump everything until EOF
+      // Same as the full loop, but without EOF handling
+      while (i < end || pushback.length > 0) {
+        const b = pushback.length > 0 ? pushback.pop() : arr[i++]
+        if (g3) {
+          if (b < 0x30 || b > 0x39) {
+            pushback.push(b, g3, g2)
+            g1 = g2 = g3 = 0
+            res += String.fromCharCode(err())
+          } else {
+            const p = index((g1 - 0x81) * 12_600 + (g2 - 0x30) * 1260 + (g3 - 0x81) * 10 + b - 0x30)
+            g1 = g2 = g3 = 0
+            if (p === undefined) {
+              res += String.fromCharCode(err())
+            } else {
+              res += String.fromCodePoint(p) // Can validly return replacement
+            }
+          }
+        } else if (g2) {
+          if (b >= 0x81 && b <= 0xfe) {
+            g3 = b
+          } else {
+            pushback.push(b, g2)
+            g1 = g2 = 0
+            res += String.fromCharCode(err())
+          }
+        } else if (g1) {
+          if (b >= 0x30 && b <= 0x39) {
+            g2 = b
+          } else {
+            let cp
+            if (b >= 0x40 && b <= 0xfe && b !== 0x7f) {
+              cp = gb18030[(g1 - 0x81) * 190 + b - (b < 0x7f ? 0x40 : 0x41)]
+            }
 
-      if (g2) {
-        if (b >= 0x81 && b <= 0xfe) {
-          g3 = b
-          return
+            g1 = 0
+            if (cp !== undefined && cp !== REP) {
+              res += String.fromCodePoint(cp)
+            } else {
+              res += String.fromCharCode(err())
+              if (b < 128) res += String.fromCharCode(b) // can be processed immediately
+            }
+          }
+        } else if (b < 128) {
+          res += String.fromCharCode(b)
+        } else if (b === 0x80) {
+          res += '\u20AC'
+        } else if (b === 0xff) {
+          res += String.fromCharCode(err())
+        } else {
+          g1 = b
         }
-
-        pushback.push(b, g2)
-        g1 = g2 = 0
-        return err()
       }
 
-      if (g1) {
-        if (b >= 0x30 && b <= 0x39) {
-          g2 = b
-          return
-        }
-
-        let cp
-        if (b >= 0x40 && b <= 0xfe && b !== 0x7f) {
-          cp = gb18030[(g1 - 0x81) * 190 + b - (b < 0x7f ? 0x40 : 0x41)]
-        }
-
-        g1 = 0
-        if (cp !== undefined && cp !== REP) return cp
-        if (b < 128) pushback.push(b)
-        return err()
+      // if g1 = 0 then g2 = g3 = 0
+      if (g1 && !stream) {
+        g1 = g2 = g3 = 0
+        res += String.fromCharCode(err())
       }
 
-      if (b < 128) return b
-      if (b === 0x80) return 0x20_ac
-      if (b === 0xff) return err()
-      g1 = b
+      return res
     }
 
-    const eof = () => {
-      if (!g1 && !g2 && !g3) return null
-      g1 = g2 = g3 = 0
-      return err()
-    }
-
-    return { bytes, eof, pushback }
+    return { decode, isAscii: () => g1 === 0 } // if g1 = 0 then g2 = g3 = 0
+  },
+  // https://encoding.spec.whatwg.org/#big5
+  big5: (err) => {
+    // The only decoder which returns multiple codepoints per byte, also has non-charcode codepoints
+    // We store that as strings
+    const big5 = getTable('big5')
+    return bigDecoder(err, (l, b) => {
+      if (b < 0x40 || (b > 0x7e && b < 0xa1) || b === 0xff) return
+      return big5[(l - 0x81) * 157 + b - (b < 0x7f ? 0x40 : 0x62)] // strings
+    })
   },
 }
 
 export const isAsciiSuperset = (enc) => enc !== 'iso-2022-jp' // all others are ASCII supersets and can use fast path
 
 export function multibyteDecoder(enc, loose = false) {
-  if (enc === 'big5') return big5decoder(loose)
   if (!Object.hasOwn(mappers, enc)) throw new RangeError('Unsupported encoding')
 
   // Input is assumed to be typechecked already
   let mapper
   const asciiSuperset = isAsciiSuperset(enc)
-  return (arr, stream = false) => {
-    const onErr = loose
-      ? () => REP
-      : () => {
-          mapper.pushback.length = 0 // the queue is cleared on returning an error
-          // The correct way per spec seems to be not destoying the decoder state in stream mode, even when fatal
-          // Decoders big5, euc-jp, euc-kr, shift_jis, gb18030 / gbk - all clear state before throwing unless EOF, so not affected
-          // iso-2022-jp is the only tricky one one where this !stream check matters in non-stream mode
-          if (!stream) mapper = null // destroy state, effectively the same as 'do not flush' = false, but early
-          throw new TypeError(E_STRICT)
-        }
-
-    let res = ''
-    const length = arr.length
-    if (asciiSuperset && !mapper) {
-      res = decodeLatin1(arr, 0, asciiPrefix(arr))
-      if (res.length === arr.length) return res // ascii
-    }
-
-    if (!mapper) mapper = mappers[enc](onErr)
-    const { bytes, eof, pushback } = mapper
-    let i = res.length
-
-    // First, dump everything until EOF
-    // Same as the full loop, but without EOF handling
-    while (i < length || pushback.length > 0) {
-      const c = bytes(pushback.length > 0 ? pushback.pop() : arr[i++])
-      if (c === undefined) continue // consuming
-      res += String.fromCodePoint(c) // gb18030 returns codepoints above 0xFFFF from ranges
-    }
-
-    // Then, dump EOF. This needs the same loop as the characters can be pushed back
-    // TODO: only some encodings need this, most can be optimized
-    if (!stream) {
-      while (i <= length || pushback.length > 0) {
-        const isEOF = i === length && pushback.length === 0
-        const c = isEOF ? eof() : bytes(pushback.length > 0 ? pushback.pop() : arr[i++])
-        if (isEOF && c === null) break // clean exit
-        if (c === undefined) continue // consuming
-        res += String.fromCodePoint(c) // gb18030 returns codepoints above 0xFFFF from ranges
+  let streaming // because onErr is cached in mapper
+  const onErr = loose
+    ? () => REP
+    : () => {
+        // The correct way per spec seems to be not destoying the decoder state in stream mode, even when fatal
+        // Decoders big5, euc-jp, euc-kr, shift_jis, gb18030 / gbk - all clear state before throwing unless EOF, so not affected
+        // iso-2022-jp is the only tricky one one where this !stream check matters in non-stream mode
+        if (!streaming) mapper = null // destroy state, effectively the same as 'do not flush' = false, but early
+        throw new TypeError(E_STRICT)
       }
-    }
-
-    // Chrome and WebKit fail on this, we don't: completely destroy the old decoder instance when finished streaming
-    // > If this’s do not flush is false, then set this’s decoder to a new instance of this’s encoding’s decoder,
-    // > Set this’s do not flush to options["stream"]
-    if (!stream) mapper = null
 
-    return res
-  }
-}
-
-// The only decoder which returns multiple codepoints per byte, also has non-charcode codepoints
-// We store that as strings
-function big5decoder(loose) {
-  // Input is assumed to be typechecked already
-  let lead = 0
-  let big5
   return (arr, stream = false) => {
-    const onErr = loose
-      ? () => '\uFFFD'
-      : () => {
-          // Lead is always already cleared before throwing
-          throw new TypeError(E_STRICT)
-        }
-
     let res = ''
-    const length = arr.length
-    if (!lead) {
+    if (asciiSuperset && (!mapper || mapper.isAscii?.())) {
       res = decodeLatin1(arr, 0, asciiPrefix(arr))
       if (res.length === arr.length) return res // ascii
     }
 
-    if (!big5) big5 = getTable('big5')
-    for (let i = res.length; i < length; i++) {
-      const b = arr[i]
-      if (lead) {
-        let cp
-        if ((b >= 0x40 && b <= 0x7e) || (b >= 0xa1 && b !== 0xff)) {
-          cp = big5[(lead - 0x81) * 157 + b - (b < 0x7f ? 0x40 : 0x62)]
-        }
-
-        lead = 0
-        if (cp) {
-          res += cp // strings
-        } else {
-          res += onErr()
-          // same as pushing it back: lead is cleared, pushed back can't contain more than 1 byte
-          if (b < 128) res += String.fromCharCode(b)
-        }
-      } else if (b < 128) {
-        res += String.fromCharCode(b)
-      } else if (b < 0x81 || b === 0xff) {
-        res += onErr()
-      } else {
-        lead = b
-      }
-    }
-
-    if (!stream && lead) {
-      // Destroy decoder state
-      lead = 0
-      res += onErr()
-    }
-
-    return res
+    streaming = stream // affects onErr
+    if (!mapper) mapper = mappers[enc](onErr)
+    return res + mapper.decode(arr, res.length, arr.length, stream)
   }
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@exodus/bytes",
-  "version": "1.4.0",
+  "version": "1.5.0",
   "description": "Various operations on Uint8Array data",
   "scripts": {
     "lint": "eslint .",