@exodus/bytes 1.0.0-rc.5 → 1.0.0-rc.6

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@exodus/bytes",
-  "version": "1.0.0-rc.5",
+  "version": "1.0.0-rc.6",
   "description": "Various operations on Uint8Array data",
   "scripts": {
     "lint": "eslint .",
@@ -39,6 +39,7 @@
   "type": "module",
   "files": [
     "/fallback/_utils.js",
+    "/fallback/latin1.js",
     "/fallback/base32.js",
     "/fallback/base64.js",
     "/fallback/hex.js",
@@ -74,12 +75,12 @@
     "@exodus/crypto": "1.0.0-rc.29",
     "@exodus/eslint-config": "^5.24.0",
     "@exodus/prettier": "^1.0.0",
-    "@exodus/test": "^1.0.0-rc.107",
+    "@exodus/test": "^1.0.0-rc.108",
     "@noble/hashes": "^2.0.1",
     "@scure/base": "^1.2.6",
     "@stablelib/base64": "^2.0.1",
     "@stablelib/hex": "^2.0.1",
-    "@types/node": "^24.0.10",
+    "@types/node": "^22.13.0",
     "base-x": "^5.0.1",
     "base32.js": "^0.1.0",
     "base64-js": "^1.5.1",
@@ -91,9 +92,11 @@
     "eslint": "^8.44.0",
     "fast-base64-decode": "^2.0.0",
     "fast-base64-encode": "^1.0.0",
+    "hextreme": "^1.0.7",
     "hi-base32": "^0.5.1",
     "jsvu": "^3.0.0",
-    "text-encoding": "^0.7.0"
+    "text-encoding": "^0.7.0",
+    "typescript": "^5.9.3"
   },
   "prettier": "@exodus/prettier",
   "packageManager": "pnpm@10.12.1+sha256.889bac470ec93ccc3764488a19d6ba8f9c648ad5e50a9a6e4be3768a5de387a3"

package/utf8.js

@@ -1,6 +1,7 @@
 import { assertUint8 } from './assert.js'
 import { typedView } from './array.js'
 import * as js from './fallback/utf8.js'
+import { asciiPrefix, decodeLatin1 } from './fallback/latin1.js'
 
 const { Buffer, TextEncoder, TextDecoder, decodeURIComponent, escape } = globalThis // Buffer is optional
 const haveNativeBuffer = Buffer && !Buffer.TYPED_ARRAY_SUPPORT
@@ -18,7 +19,7 @@ const { E_STRICT, E_STRICT_UNICODE } = js
 const shouldUseEscapePath = Boolean(globalThis.HermesInternal) // faster only on Hermes, js path beats it on normal engines
 
 function deLoose(str, loose, res) {
-  if (loose) return res
+  if (loose || str.length === res.length) return res // length is equal only for ascii, which is automatically fine
   if (isWellFormed) {
     // We have a fast native method
     if (isWellFormed.call(str)) return res
@@ -51,64 +52,27 @@ function encode(str, loose = false) {
   return js.encode(str, loose)
 }
 
-let escapes
-
-function toEscapesPart(arr, start, end) {
-  let o = ''
-  let i = start
-  const last3 = end - 3
-  // Unrolled loop is faster
-  while (i < last3) {
-    const a = arr[i++]
-    const b = arr[i++]
-    const c = arr[i++]
-    const d = arr[i++]
-    o += escapes[a]
-    o += escapes[b]
-    o += escapes[c]
-    o += escapes[d]
-  }
-
-  while (i < end) o += escapes[arr[i++]]
-  return o
-}
-
 function decode(arr, loose = false) {
   assertUint8(arr)
   if (haveDecoder) return loose ? decoderLoose.decode(arr) : decoderFatal.decode(arr) // Node.js and browsers
   // No reason to use native Buffer: it's not faster than TextDecoder, needs rechecks in non-loose mode, and Node.js has TextDecoder
 
-  // This codepath gives a ~2x perf boost on Hermes
-  if (shouldUseEscapePath && escape && decodeURIComponent) {
-    if (!escapes) escapes = Array.from({ length: 256 }, (_, i) => escape(String.fromCharCode(i)))
-    const length = arr.length
-    let o
-    if (length > 30_000) {
-      // Limit concatenation to avoid excessive GC
-      // TODO: recheck thresholds on Hermes (taken from hex)
-      const concat = []
-      for (let i = 0; i < length; ) {
-        const step = i + 500
-        const end = step > length ? length : step
-        concat.push(toEscapesPart(arr, i, end))
-        i = end
-      }
-
-      o = concat.join('')
-      concat.length = 0
-    } else {
-      o = toEscapesPart(arr, 0, length)
-    }
+  // Fast path for ASCII prefix, this is faster than all alternatives below
+  const prefix = decodeLatin1(arr, 0, asciiPrefix(arr))
+  if (prefix.length === arr.length) return prefix
 
+  // This codepath gives a ~3x perf boost on Hermes
+  if (shouldUseEscapePath && escape && decodeURIComponent) {
+    const o = escape(decodeLatin1(arr, prefix.length, arr.length))
     try {
-      return decodeURIComponent(o) // asci to utf8, escape() is precalucated
+      return prefix + decodeURIComponent(o) // Latin1 to utf8
     } catch {
       if (!loose) throw new TypeError(E_STRICT)
       // Ok, we have to use manual implementation for loose decoder
     }
   }
 
-  return js.decode(arr, loose)
+  return prefix + js.decode(arr, loose, prefix.length)
 }
 
 export const utf8fromString = (str, format = 'uint8') => typedView(encode(str, false), format)