Update elm/bytes _Bytes_read_string to use array instead of string as accumulator (#110)

Relates to #107 and elm/bytes#19.

Updated `_Bytes_read_string` to use array instead of string as
accumulator.

When compiling `rtfeldman/elm-css`, before changes:

```
elm-css % rm -rf guida-stuff; time ../guida-lang/compiler/bin/index.js make
Dependencies ready!
Success! Compiled 25 modules.
../guida-lang/compiler/bin/index.js make  21.36s user 2.76s system 101% cpu 23.686 total
```

After changes:

```
elm-css % rm -rf guida-stuff; time ../guida-lang/compiler/bin/index.js make
Dependencies ready!
Success! Compiled 25 modules.
../guida-lang/compiler/bin/index.js make  13.95s user 2.20s system 107% cpu 14.984 total
```

Author: decioferreira

eslint.config.mjs

@@ -21,6 +21,7 @@ export default defineConfig([
   { files: ["lib/browser.js"], languageOptions: { globals: globals.browser } },
   { files: ["lib/node.js"], languageOptions: { globals: globals.node } },
   { files: ["try/**/*.{js,mjs,cjs}"], languageOptions: { globals: { ...globals.browser, ...globals.node } } },
+  { files: ["scripts/*.js"], languageOptions: { globals: globals.node } },
   {
     files: ["**/*.{js,mjs,cjs}"],
     plugins: { js },

scripts/build.sh

@@ -27,6 +27,7 @@ js="$filepath.js"
 min="$filepath.min.js"
 
 guida make --optimize --output=$js $elm_entry
+node scripts/replace-bytes-read-string.js $js
 
 uglifyjs $js --compress "pure_funcs=[F2,F3,F4,F5,F6,F7,F8,F9,A2,A3,A4,A5,A6,A7,A8,A9],pure_getters,keep_fargs=false,unsafe_comps,unsafe" | uglifyjs --mangle --output $min
 

scripts/replace-bytes-read-string.js

@@ -0,0 +1,82 @@
+#!/usr/bin/env node
+
+/* This change prevents V8 from retaining large "concatenated string" chains, which can cause OOMs.
+   Tested against `rtfeldman/elm-css` compilation.
+
+   See the related discussion for context: https://discourse.elm-lang.org/t/guida-compiler-was-there-are-3-elm-compilers-written-in-elm/10329/34
+   and issue: https://github.com/guida-lang/compiler/issues/107
+ */
+
+const fs = require('node:fs');
+
+const argv = process.argv.slice(2);
+const path = argv[0];
+
+const data = fs
+    .readFileSync(path, { encoding: 'utf8', flag: 'r' })
+    .replace(`var _Bytes_read_string = F3(function(len, bytes, offset)
+{
+	var string = '';
+	var end = offset + len;
+	for (; offset < end;)
+	{
+		var byte = bytes.getUint8(offset++);
+		string +=
+			(byte < 128)
+				? String.fromCharCode(byte)
+				:
+			((byte & 0xE0 /* 0b11100000 */) === 0xC0 /* 0b11000000 */)
+				? String.fromCharCode((byte & 0x1F /* 0b00011111 */) << 6 | bytes.getUint8(offset++) & 0x3F /* 0b00111111 */)
+				:
+			((byte & 0xF0 /* 0b11110000 */) === 0xE0 /* 0b11100000 */)
+				? String.fromCharCode(
+					(byte & 0xF /* 0b00001111 */) << 12
+					| (bytes.getUint8(offset++) & 0x3F /* 0b00111111 */) << 6
+					| bytes.getUint8(offset++) & 0x3F /* 0b00111111 */
+				)
+				:
+				(byte =
+					((byte & 0x7 /* 0b00000111 */) << 18
+						| (bytes.getUint8(offset++) & 0x3F /* 0b00111111 */) << 12
+						| (bytes.getUint8(offset++) & 0x3F /* 0b00111111 */) << 6
+						| bytes.getUint8(offset++) & 0x3F /* 0b00111111 */
+					) - 0x10000
+				, String.fromCharCode(Math.floor(byte / 0x400) + 0xD800, byte % 0x400 + 0xDC00)
+				);
+	}
+	return _Utils_Tuple2(offset, string);
+});`, `var _Bytes_read_string = F3(function(len, bytes, offset)
+{
+	var string = [];
+	var end = offset + len;
+	for (; offset < end;)
+	{
+		var byte = bytes.getUint8(offset++);
+		string.push(
+			(byte < 128)
+				? String.fromCharCode(byte)
+				:
+			((byte & 0xE0 /* 0b11100000 */) === 0xC0 /* 0b11000000 */)
+				? String.fromCharCode((byte & 0x1F /* 0b00011111 */) << 6 | bytes.getUint8(offset++) & 0x3F /* 0b00111111 */)
+				:
+			((byte & 0xF0 /* 0b11110000 */) === 0xE0 /* 0b11100000 */)
+				? String.fromCharCode(
+					(byte & 0xF /* 0b00001111 */) << 12
+					| (bytes.getUint8(offset++) & 0x3F /* 0b00111111 */) << 6
+					| bytes.getUint8(offset++) & 0x3F /* 0b00111111 */
+				)
+				:
+				(byte =
+					((byte & 0x7 /* 0b00000111 */) << 18
+						| (bytes.getUint8(offset++) & 0x3F /* 0b00111111 */) << 12
+						| (bytes.getUint8(offset++) & 0x3F /* 0b00111111 */) << 6
+						| bytes.getUint8(offset++) & 0x3F /* 0b00111111 */
+					) - 0x10000
+				, String.fromCharCode(Math.floor(byte / 0x400) + 0xD800, byte % 0x400 + 0xDC00)
+				)
+		);
+	}
+	return _Utils_Tuple2(offset, string.join(''));
+});`);
+
+fs.writeFileSync(path, data, { encoding: 'utf8', flag: 'w' });

src/Compiler/Parse/Primitives.elm

@@ -483,24 +483,17 @@ snippetEncoder (Snippet { fptr, offset, length, offRow, offCol }) =
 
 snippetDecoder : BD.Decoder Snippet
 snippetDecoder =
-    -- This `String.toList` -> `String.fromList` round-trip prevents V8 from retaining
-    -- large "concatenated string" chains, which can cause OOMs.
-    -- Tested against `rtfeldman/elm-css` compilation.
-    -- The split call avoids elm-review flags for this pattern.
-    --
-    -- See the related discussion for context:
-    -- https://discourse.elm-lang.org/t/guida-compiler-was-there-are-3-elm-compilers-written-in-elm/10329/25
     BD.map5
         (\fptr offset length offRow offCol ->
             Snippet
-                { fptr = String.fromList fptr
+                { fptr = fptr
                 , offset = offset
                 , length = length
                 , offRow = offRow
                 , offCol = offCol
                 }
         )
-        (BD.map String.toList BD.string)
+        BD.string
         BD.int
         BD.int
         BD.int