porffor 0.0.0-bddcdc3 → 0.0.0-ebc0491

package/README.md

@@ -77,6 +77,7 @@ these include some early (stage 1/0) and/or dead (last commit years ago) proposa
 - string concat (`+`) (eg `'a' + 'b'`)
 - truthy/falsy (eg `!'' == true`)
 - string comparison (eg `'a' == 'a'`, `'a' != 'b'`)
+- nullish coalescing operator (`??`)
 
 ### built-ins
 
@@ -100,17 +101,29 @@ these include some early (stage 1/0) and/or dead (last commit years ago) proposa
 - intrinsic functions (see below)
 - inlining wasm via ``asm`...``\` "macro"
 
-## soon todo
+## todo
+no particular order and no guarentees, just what could happen soon™
+
 - arrays
   - member setting (`arr[0] = 2`)
   - more of `Array` prototype
   - arrays/strings inside arrays
+  - destructuring
+  - for .. of
 - strings
   - member setting
+- objects
+  - basic object expressions (eg `{}`, `{ a: 0 }`)
+- wasm
+  - *basic* wasm engine (interpreter) in js
+- regex
+  - *basic* regex engine (in wasm compiled aot or js interpreter?)
 - more math operators (`**`, etc)
 - `do { ... } while (...)`
+- rewrite `console.log` to work with strings/arrays
 - exceptions
-  - `try { } finally {}`
+  - rewrite to use actual strings (optional?)
+  - `try { } finally { }`
   - rethrowing inside catch
 - optimizations
   - rewrite local indexes per func for smallest local header and remove unused idxs
@@ -118,6 +131,11 @@ these include some early (stage 1/0) and/or dead (last commit years ago) proposa
   - remove const ifs (`if (true)`, etc)
   - use data segments for initing arrays
 
+## porfformance
+*for the things it supports*, porffor is blazingly faster compared to most interpreters, and engines running without JIT. for those with JIT, it is not that much slower like a traditional interpreter would be.
+
+![Screenshot of comparison chart](https://github.com/CanadaHonk/porffor/assets/19228318/76c75264-cc68-4be1-8891-c06dc389d97a)
+
 ## test262
 porffor can run test262 via some hacks/transforms which remove unsupported features whilst still doing the same asserts (eg simpler error messages using literals only). it currently passes >10% (see latest commit desc for latest and details). use `node test262` to test, it will also show a difference of overall results between the last commit and current results.
 
@@ -135,6 +153,7 @@ mostly for reducing size. do not really care about compiler perf/time as long as
 - `i64.extend_i32_s`, `i32.wrap_i64` -> ``
 - `f64.convert_i32_u`, `i32.trunc_sat_f64_s` -> ``
 - `return`, `end` -> `end`
+- change const, convert to const of converted valtype (eg `f64.const`, `i32.trunc_sat_f64_s -> `i32.const`)
 - remove some redundant sets/gets
 - remove unneeded single just used vars
 - remove unneeded blocks (no `br`s inside)
@@ -190,11 +209,12 @@ you can also use deno (`deno run -A ...` instead of `node ...`), or bun (`bun ..
 - `-no-run` to not run wasm output, just compile
 - `-opt-log` to log some opts
 - `-code-log` to log some codegen (you probably want `-funcs`)
-- `-funcs` to log funcs (internal representations)
+- `-funcs` to log funcs
 - `-opt-funcs` to log funcs after opt
 - `-sections` to log sections as hex
 - `-opt-no-inline` to not inline any funcs
-- `-tail-call` to enable tail calls (not widely implemented)
+- `-tail-call` to enable tail calls (experimental + not widely implemented)
+- `-compile-hints` to enable V8 compilation hints (experimental + doesn't seem to do much?)
 
 ## vscode extension
 there is a vscode extension in `porffor-for-vscode` which tweaks js syntax highlighting to be nicer with porffor features (eg highlighting wasm inside of inline asm).

package/compiler/codeGen.js

@@ -35,7 +35,14 @@ const debug = str => {
 };
 
 const todo = msg => {
-  throw new Error(`todo: ${msg}`);
+  class TodoError extends Error {
+    constructor(message) {
+      super(message);
+      this.name = 'TodoError';
+    }
+  }
+
+  throw new TodoError(`todo: ${msg}`);
 
   const code = [];
 
@@ -167,7 +174,6 @@ const generate = (scope, decl, global = false, name = undefined) => {
         }
 
         if (asm[0] === 'memory') {
-          scope.memory = true;
           allocPage('asm instrinsic');
           // todo: add to store/load offset insts
           continue;
@@ -281,7 +287,7 @@ const generateReturn = (scope, decl) => {
     ];
   }
 
-  if (!scope.returnType) scope.returnType = getNodeType(scope, decl.argument);
+  scope.returnType = getNodeType(scope, decl.argument);
 
   return [
     ...generate(scope, decl.argument),
@@ -298,11 +304,11 @@ const localTmp = (scope, name, type = valtypeBinary) => {
   return idx;
 };
 
-const performLogicOp = (scope, op, left, right) => {
+const performLogicOp = (scope, op, left, right, leftType, rightType) => {
   const checks = {
-    '||': Opcodes.eqz,
-    '&&': [ Opcodes.i32_to ]
-    // todo: ??
+    '||': falsy,
+    '&&': truthy,
+    '??': nullish
   };
 
   if (!checks[op]) return todo(`logic operator ${op} not implemented yet`);
@@ -313,7 +319,8 @@ const performLogicOp = (scope, op, left, right) => {
   return [
     ...left,
     [ Opcodes.local_tee, localTmp(scope, 'logictmp') ],
-    ...checks[op],
+    ...checks[op](scope, [], leftType),
+    Opcodes.i32_to,
     [ Opcodes.if, valtypeBinary ],
     ...right,
     [ Opcodes.else ],
@@ -328,8 +335,6 @@ const concatStrings = (scope, left, right, global, name, assign) => {
   // todo: optimize by looking up names in arrays and using that if exists?
   // todo: optimize this if using literals/known lengths?
 
-  scope.memory = true;
-
   const rightPointer = localTmp(scope, 'concat_right_pointer', Valtype.i32);
   const rightLength = localTmp(scope, 'concat_right_length', Valtype.i32);
   const leftLength = localTmp(scope, 'concat_left_length', Valtype.i32);
@@ -464,8 +469,6 @@ const compareStrings = (scope, left, right) => {
   // todo: optimize by looking up names in arrays and using that if exists?
   // todo: optimize this if using literals/known lengths?
 
-  scope.memory = true;
-
   const leftPointer = localTmp(scope, 'compare_left_pointer', Valtype.i32);
   const leftLength = localTmp(scope, 'compare_left_length', Valtype.i32);
   const rightPointer = localTmp(scope, 'compare_right_pointer', Valtype.i32);
@@ -563,75 +566,100 @@ const compareStrings = (scope, left, right) => {
   ];
 };
 
-const falsy = (scope, wasm, type) => {
+const truthy = (scope, wasm, type) => {
   // arrays are always truthy
   if (type === TYPES._array) return [
     ...wasm,
     [ Opcodes.drop ],
-    number(0)
+    ...number(1)
   ];
 
   if (type === TYPES.string) {
-    // if "" (length = 0)
+    // if not "" (length = 0)
     return [
       // pointer
       ...wasm,
+      Opcodes.i32_to_u,
 
       // get length
       [ Opcodes.i32_load, Math.log2(ValtypeSize.i32) - 1, 0 ],
 
-      // if length == 0
-      [ Opcodes.i32_eqz ],
+      // if length != 0
+      /* [ Opcodes.i32_eqz ],
+      [ Opcodes.i32_eqz ], */
       Opcodes.i32_from_u
     ]
   }
 
-  // if = 0
+  // if != 0
   return [
     ...wasm,
 
-    ...Opcodes.eqz,
-    Opcodes.i32_from_u
+    /* Opcodes.eqz,
+    [ Opcodes.i32_eqz ],
+    Opcodes.i32_from */
   ];
 };
 
-const truthy = (scope, wasm, type) => {
+const falsy = (scope, wasm, type) => {
   // arrays are always truthy
   if (type === TYPES._array) return [
     ...wasm,
     [ Opcodes.drop ],
-    number(1)
+    ...number(0)
   ];
 
   if (type === TYPES.string) {
-    // if not "" (length = 0)
+    // if "" (length = 0)
     return [
       // pointer
       ...wasm,
+      Opcodes.i32_to_u,
 
       // get length
       [ Opcodes.i32_load, Math.log2(ValtypeSize.i32) - 1, 0 ],
 
-      // if length != 0
-      /* [ Opcodes.i32_eqz ],
-      [ Opcodes.i32_eqz ], */
+      // if length == 0
+      [ Opcodes.i32_eqz ],
       Opcodes.i32_from_u
     ]
   }
 
-  // if != 0
+  // if = 0
   return [
     ...wasm,
 
-    /* Opcodes.eqz,
-    [ Opcodes.i32_eqz ],
-    Opcodes.i32_from */
+    ...Opcodes.eqz,
+    Opcodes.i32_from_u
+  ];
+};
+
+const nullish = (scope, wasm, type) => {
+  // undefined
+  if (type === TYPES.undefined) return [
+    ...wasm,
+    [ Opcodes.drop ],
+    ...number(1)
+  ];
+
+  // null (if object and = "0")
+  if (type === TYPES.object) return [
+    ...wasm,
+    ...Opcodes.eqz,
+    Opcodes.i32_from_u
+  ];
+
+  // not
+  return [
+    ...wasm,
+    [ Opcodes.drop ],
+    ...number(0)
   ];
 };
 
 const performOp = (scope, op, left, right, leftType, rightType, _global = false, _name = '$undeclared', assign = false) => {
   if (op === '||' || op === '&&' || op === '??') {
-    return performLogicOp(scope, op, left, right);
+    return performLogicOp(scope, op, left, right, leftType, rightType);
   }
 
   if (codeLog && (!leftType || !rightType)) log('codegen', 'untracked type in op', op, _name, '\n' + new Error().stack.split('\n').slice(1).join('\n'));
@@ -772,7 +800,7 @@ const includeBuiltin = (scope, builtin) => {
 };
 
 const generateLogicExp = (scope, decl) => {
-  return performLogicOp(scope, decl.operator, generate(scope, decl.left), generate(scope, decl.right));
+  return performLogicOp(scope, decl.operator, generate(scope, decl.left), generate(scope, decl.right), getNodeType(scope, decl.left), getNodeType(scope, decl.right));
 };
 
 const TYPES = {
@@ -936,7 +964,8 @@ const generateLiteral = (scope, decl, global, name) => {
 const countLeftover = wasm => {
   let count = 0, depth = 0;
 
-  for (const inst of wasm) {
+  for (let i = 0; i < wasm.length; i++) {
+    const inst = wasm[i];
     if (depth === 0 && (inst[0] === Opcodes.if || inst[0] === Opcodes.block || inst[0] === Opcodes.loop)) {
       if (inst[0] === Opcodes.if) count--;
       if (inst[1] !== Blocktype.void) count++;
@@ -1040,7 +1069,7 @@ const generateCall = (scope, decl, _global, _name) => {
   }
 
   let out = [];
-  let protoFunc, protoName, baseType, baseName = '$undeclared';
+  let protoFunc, protoName, baseType, baseName;
   // ident.func()
   if (name && name.startsWith('__')) {
     const spl = name.slice(2).split('_');
@@ -1063,11 +1092,11 @@ const generateCall = (scope, decl, _global, _name) => {
 
     out = generate(scope, decl.callee.object);
     out.push([ Opcodes.drop ]);
+
+    baseName = [...arrays.keys()].pop();
   }
 
   if (protoFunc) {
-    scope.memory = true;
-
     let pointer = arrays.get(baseName);
 
     if (pointer == null) {
@@ -1075,7 +1104,7 @@ const generateCall = (scope, decl, _global, _name) => {
       if (codeLog) log('codegen', 'cloning unknown dynamic pointer');
 
       // register array
-      const [ , pointer ] = makeArray(scope, {
+      0, [ , pointer ] = makeArray(scope, {
         rawElements: new Array(0)
       }, _global, baseName, true, baseType === TYPES.string ? 'i16' : valtype);
 
@@ -1173,7 +1202,6 @@ const generateCall = (scope, decl, _global, _name) => {
     args = args.slice(0, func.params.length);
   }
 
-  if (func && func.memory) scope.memory = true;
   if (func && func.throws) scope.throws = true;
 
   for (const arg of args) {
@@ -1189,7 +1217,7 @@ const generateNew = (scope, decl, _global, _name) => {
   // hack: basically treat this as a normal call for builtins for now
   const name = mapName(decl.callee.name);
   if (internalConstrs[name]) return internalConstrs[name].generate(scope, decl, _global, _name);
-  if (!builtinFuncs[name]) return todo(`new statement is not supported yet (new ${unhackName(name)})`);
+  if (!builtinFuncs[name]) return todo(`new statement is not supported yet`); // return todo(`new statement is not supported yet (new ${unhackName(name)})`);
 
   return generateCall(scope, decl, _global, _name);
 };
@@ -1297,8 +1325,6 @@ const generateAssign = (scope, decl) => {
     const name = decl.left.object.name;
     const pointer = arrays.get(name);
 
-    scope.memory = true;
-
     const aotPointer = pointer != null;
 
     const newValueTmp = localTmp(scope, '__length_setter_tmp');
@@ -1349,8 +1375,27 @@ const generateAssign = (scope, decl) => {
     ];
   }
 
+  const op = decl.operator.slice(0, -1);
+  if (op === '||' || op === '&&' || op === '??') {
+    // todo: is this needed?
+    // for logical assignment ops, it is not left @= right ~= left = left @ right
+    // instead, left @ (left = right)
+    // eg, x &&= y ~= x && (x = y)
+
+    return [
+      ...performOp(scope, op, [
+        [ isGlobal ? Opcodes.global_get : Opcodes.local_get, local.idx ]
+      ], [
+        ...generate(scope, decl.right),
+        [ isGlobal ? Opcodes.global_set : Opcodes.local_set, local.idx ],
+        [ isGlobal ? Opcodes.global_get : Opcodes.local_get, local.idx ]
+      ], getType(scope, name), getNodeType(scope, decl.right), isGlobal, name, true),
+      [ isGlobal ? Opcodes.global_get : Opcodes.local_get, local.idx ]
+    ];
+  }
+
   return [
-    ...performOp(scope, decl.operator.slice(0, -1), [ [ isGlobal ? Opcodes.global_get : Opcodes.local_get, local.idx ] ], generate(scope, decl.right), getType(scope, name), getNodeType(scope, decl.right), isGlobal, name, true),
+    ...performOp(scope, op, [ [ isGlobal ? Opcodes.global_get : Opcodes.local_get, local.idx ] ], generate(scope, decl.right), getType(scope, name), getNodeType(scope, decl.right), isGlobal, name, true),
     [ isGlobal ? Opcodes.global_set : Opcodes.local_set, local.idx ],
     [ isGlobal ? Opcodes.global_get : Opcodes.local_get, local.idx ]
   ];
@@ -1377,13 +1422,14 @@ const generateUnary = (scope, decl) => {
 
     case '!':
       // !=
-      return falsy(scope, generate(scope, decl.argument));
+      return falsy(scope, generate(scope, decl.argument), getNodeType(scope, decl.argument));
 
     case '~':
+      // todo: does not handle Infinity properly (should convert to 0) (but opt const converting saves us sometimes)
       return [
         ...generate(scope, decl.argument),
         Opcodes.i32_to,
-        [ Opcodes.i32_const, signedLEB128(-1) ],
+        [ Opcodes.i32_const, ...signedLEB128(-1) ],
         [ Opcodes.i32_xor ],
         Opcodes.i32_from
       ];
@@ -1665,7 +1711,16 @@ const allocPage = reason => {
   let ind = pages.size;
   pages.set(reason, ind);
 
-  if (codeLog) log('codegen', `allocated new page of memory (${ind}) | ${reason}`);
+  if (allocLog) log('alloc', `allocated new page of memory (${ind}) | ${reason}`);
+
+  return ind;
+};
+
+const freePage = reason => {
+  let ind = pages.get(reason);
+  pages.delete(reason);
+
+  if (allocLog) log('alloc', `freed page of memory (${ind}) | ${reason}`);
 
   return ind;
 };
@@ -1727,8 +1782,6 @@ const makeArray = (scope, decl, global = false, name = '$undeclared', initEmpty
   // local value as pointer
   out.push(...number(pointer));
 
-  scope.memory = true;
-
   return [ out, pointer ];
 };
 
@@ -1747,8 +1800,6 @@ export const generateMember = (scope, decl, _global, _name) => {
     const name = decl.object.name;
     const pointer = arrays.get(name);
 
-    scope.memory = true;
-
     const aotPointer = pointer != null;
 
     return [
@@ -1768,8 +1819,6 @@ export const generateMember = (scope, decl, _global, _name) => {
   const name = decl.object.name;
   const pointer = arrays.get(name);
 
-  scope.memory = true;
-
   const aotPointer = pointer != null;
 
   if (type === TYPES._array) {
@@ -1879,6 +1928,7 @@ const generateFunc = (scope, decl) => {
     locals: {},
     localInd: 0,
     returns: [ valtypeBinary ],
+    returnType: null,
     memory: false,
     throws: false,
     name
@@ -1905,7 +1955,6 @@ const generateFunc = (scope, decl) => {
     returns: innerScope.returns,
     returnType: innerScope.returnType,
     locals: innerScope.locals,
-    memory: innerScope.memory,
     throws: innerScope.throws,
     index: currentFuncIndex++
   };
@@ -1920,6 +1969,8 @@ const generateFunc = (scope, decl) => {
 
   if (name !== 'main' && func.returns.length !== 0 && wasm[wasm.length - 1]?.[0] !== Opcodes.return && countLeftover(wasm) === 0) {
     wasm.push(...number(0), [ Opcodes.return ]);
+
+    if (func.returnType === null) func.returnType = TYPES.undefined;
   }
 
   // change v128 params into many <type> (i32x4 -> i32/etc) instead as unsupported param valtype
@@ -1930,9 +1981,7 @@ const generateFunc = (scope, decl) => {
     if (local.type === Valtype.v128) {
       vecParams++;
 
-      /* func.memory = true; // mark func as using memory
-
-      wasm.unshift( // add v128 load for param
+      /* wasm.unshift( // add v128 load for param
         [ Opcodes.i32_const, 0 ],
         [ ...Opcodes.v128_load, 0, i * 16 ],
         [ Opcodes.local_set, local.idx ]

package/compiler/decompile.js

@@ -42,8 +42,8 @@ export default (wasm, name = '', ind = 0, locals = {}, params = [], returns = []
       out += ` ${read_ieee754_binary64(inst.slice(1))}`;
     } else if (inst[0] === Opcodes.i32_const || inst[0] === Opcodes.i64_const) {
       out += ` ${read_signedLEB128(inst.slice(1))}`;
-    } else if (inst[0] === Opcodes.i32_load || inst[0] === Opcodes.i64_load || inst[0] === Opcodes.f64_load || inst[0] === Opcodes.i32_store || inst[0] === Opcodes.i64_store || inst[0] === Opcodes.f64_store) {
-      out += ` ${inst[1]} ${read_unsignedLEB128(inst.slice(2))}`
+    } else if (inst[0] === Opcodes.i32_load || inst[0] === Opcodes.i64_load || inst[0] === Opcodes.f64_load || inst[0] === Opcodes.i32_store || inst[0] === Opcodes.i64_store || inst[0] === Opcodes.f64_store || inst[0] === Opcodes.i32_store16 || inst[0] === Opcodes.i32_load16_u) {
+      out += ` ${inst[1]} ${read_unsignedLEB128(inst.slice(2))}`;
     } else for (const operand of inst.slice(1)) {
       if (inst[0] === Opcodes.if || inst[0] === Opcodes.loop || inst[0] === Opcodes.block) {
         if (operand === Blocktype.void) continue;

package/compiler/encoding.js

@@ -22,15 +22,15 @@ export const encodeLocal = (count, type) => [
   type
 ];
 
+// todo: this only works with integers within 32 bit range
 export const signedLEB128 = n => {
-  // todo: this only works with integers within 32 bit range
+  n |= 0;
 
   // just input for small numbers (for perf as common)
   if (n >= 0 && n <= 63) return [ n ];
   if (n >= -64 && n <= 0) return [ 128 + n ];
 
   const buffer = [];
-  n |= 0;
 
   while (true) {
     let byte = n & 0x7f;
@@ -50,6 +50,8 @@ export const signedLEB128 = n => {
 };
 
 export const unsignedLEB128 = n => {
+  n |= 0;
+
   // just input for small numbers (for perf as common)
   if (n >= 0 && n <= 127) return [ n ];
 

package/compiler/index.js

@@ -14,7 +14,8 @@ const bold = x => `\u001b[1m${x}\u001b[0m`;
 const areaColors = {
   codegen: [ 20, 80, 250 ],
   opt: [ 250, 20, 80 ],
-  sections: [ 20, 250, 80 ]
+  sections: [ 20, 250, 80 ],
+  alloc: [ 250, 250, 20 ]
 };
 
 globalThis.log = (area, ...args) => console.log(`\u001b[90m[\u001b[0m${rgb(...areaColors[area], area)}\u001b[90m]\u001b[0m`, ...args);
@@ -38,6 +39,7 @@ const logFuncs = (funcs, globals, exceptions) => {
 export default (code, flags) => {
   globalThis.optLog = process.argv.includes('-opt-log');
   globalThis.codeLog = process.argv.includes('-code-log');
+  globalThis.allocLog = process.argv.includes('-alloc-log');
 
   for (const x in BuiltinPreludes) {
     if (code.indexOf(x + '(') !== -1) code = BuiltinPreludes[x] + code;
@@ -63,5 +65,12 @@ export default (code, flags) => {
   const sections = produceSections(funcs, globals, tags, pages, flags);
   if (flags.includes('info')) console.log(`4. produced sections in ${(performance.now() - t3).toFixed(2)}ms`);
 
+  if (allocLog) {
+    const wasmPages = Math.ceil((pages.size * pageSize) / 65536);
+    const bytes = wasmPages * 65536;
+    log('alloc', `\x1B[1mallocated ${bytes / 1024}KiB\x1B[0m for ${pages.size} things using ${wasmPages} Wasm page${wasmPages === 1 ? '' : 's'}`);
+    // console.log([...pages.keys()].map(x => `\x1B[36m - ${x}\x1B[0m`).join('\n'));
+  }
+
   return { wasm: sections, funcs, globals, tags, exceptions, pages };
 };

package/compiler/opt.js

@@ -1,5 +1,6 @@
 import { Opcodes, Valtype } from "./wasmSpec.js";
 import { number } from "./embedding.js";
+import { read_signedLEB128, read_ieee754_binary64 } from "./encoding.js";
 
 // deno compat
 if (typeof process === 'undefined' && typeof Deno !== 'undefined') {
@@ -20,7 +21,7 @@ export default (funcs, globals) => {
   if (optLevel === 0) return;
 
   const tailCall = process.argv.includes('-tail-call');
-  if (tailCall) log('opt', 'tail call proposal is not widely implemented! (you used -tail-call)');
+  if (tailCall) log('opt', 'warning: tail call proposal is not widely implemented! (you used -tail-call)');
 
   if (optLevel >= 2 && !process.argv.includes('-opt-no-inline')) {
     // inline pass (very WIP)
@@ -95,7 +96,6 @@ export default (funcs, globals) => {
         }
 
         if (t.index > c.index) t.index--; // adjust index if after removed func
-        if (c.memory) t.memory = true;
       }
 
       funcs.splice(funcs.indexOf(c), 1); // remove func from funcs
@@ -213,9 +213,9 @@ export default (funcs, globals) => {
         // i32.const 0
         // drop
         // -->
-        // <nothing>>
+        // <nothing>
 
-        wasm.splice(i - 1, 2); // remove this inst
+        wasm.splice(i - 1, 2); // remove these inst
         i -= 2;
         continue;
       }
@@ -259,6 +259,36 @@ export default (funcs, globals) => {
         continue;
       }
 
+      if (lastInst[0] === Opcodes.const && (inst === Opcodes.i32_to || inst === Opcodes.i32_to_u)) {
+        // change const and immediate i32 convert to i32 const
+        // f64.const 0
+        // i32.trunc_sat_f64_s || i32.trunc_sat_f64_u
+        // -->
+        // i32.const 0
+
+        wasm[i - 1] = number((valtype === 'f64' ? read_ieee754_binary64 : read_signedLEB128)(lastInst.slice(1)), Valtype.i32)[0]; // f64.const -> i32.const
+
+        wasm.splice(i, 1); // remove this inst
+        i--;
+        if (optLog) log('opt', `converted const -> i32 convert into i32 const`);
+        continue;
+      }
+
+      if (lastInst[0] === Opcodes.i32_const && (inst === Opcodes.i32_from || inst === Opcodes.i32_from_u)) {
+        // change i32 const and immediate convert to const (opposite way of previous)
+        // i32.const 0
+        // f64.convert_i32_s || f64.convert_i32_u
+        // -->
+        // f64.const 0
+
+        wasm[i - 1] = number(read_signedLEB128(lastInst.slice(1)))[0]; // i32.const -> f64.const
+
+        wasm.splice(i, 1); // remove this inst
+        i--;
+        if (optLog) log('opt', `converted i32 const -> convert into const`);
+        continue;
+      }
+
       if (tailCall && lastInst[0] === Opcodes.call && inst[0] === Opcodes.return) {
         // replace call, return with tail calls (return_call)
         // call X

package/compiler/prototype.js

@@ -25,7 +25,6 @@ export const PrototypeFuncs = function() {
 
   this[TYPES._array] = {
     // lX = local accessor of X ({ get, set }), iX = local index of X, wX = wasm ops of X
-    // todo: out of bounds (>) properly
     at: (pointer, length, wIndex, iTmp) => [
       ...wIndex,
       Opcodes.i32_to,
@@ -147,7 +146,6 @@ export const PrototypeFuncs = function() {
   this[TYPES._array].push.noArgRetLength = true;
 
   this[TYPES.string] = {
-    // todo: out of bounds properly
     at: (pointer, length, wIndex, iTmp, arrayShell) => {
       const [ newOut, newPointer ] = arrayShell(1, 'i16');
 
@@ -157,9 +155,9 @@ export const PrototypeFuncs = function() {
         [ Opcodes.drop ],
 
         ...number(0, Valtype.i32), // base 0 for store later
-        Opcodes.i32_to_u,
 
         ...wIndex,
+        Opcodes.i32_to_u,
         [ Opcodes.local_tee, iTmp ],
 
         // if index < 0: access index + array length
@@ -265,7 +263,7 @@ export const PrototypeFuncs = function() {
     },
   };
 
-  this[TYPES.string].at.local = valtypeBinary;
+  this[TYPES.string].at.local = Valtype.i32;
   this[TYPES.string].at.returnType = TYPES.string;
   this[TYPES.string].charAt.returnType = TYPES.string;
   this[TYPES.string].charCodeAt.local = Valtype.i32;

package/compiler/rhemyn/compile.js

@@ -0,0 +1,173 @@
+import { Blocktype, Opcodes, Valtype, PageSize, ValtypeSize } from '../wasmSpec.js';
+import { signedLEB128, unsignedLEB128 } from '../encoding.js';
+import parse from './parse.js';
+
+// local indexes
+const BasePointer = 0; // base string pointer
+const IterPointer = 1; // this iteration base pointer
+const Counter = 2; // what char we are running on
+const Pointer = 3; // next char BYTE pointer
+const Length = 4;
+const Tmp = 5;
+
+const generate = (node, get = true) => {
+  let out = [];
+  switch (node.type) {
+    case 'Expression':
+      out = [
+        // set length local
+        [ Opcodes.local_get, BasePointer ],
+        [ Opcodes.i32_load, Math.log2(ValtypeSize.i32) - 1, ...unsignedLEB128(0) ],
+        [ Opcodes.local_set, Length ],
+
+        // set iter pointer local as base + sizeof i32 initially
+        [ Opcodes.local_get, BasePointer ],
+        ...number(ValtypeSize.i32, Valtype.i32),
+        [ Opcodes.i32_add ],
+        [ Opcodes.local_set, IterPointer ],
+
+        [ Opcodes.loop, Blocktype.void ],
+
+        // reset pointer as iter pointer
+        [ Opcodes.local_get, IterPointer ],
+        [ Opcodes.local_set, Pointer ],
+
+        [ Opcodes.block, Blocktype.void ],
+        // generate checks
+        ...generate(x),
+
+        // reached end without branching out, successful match
+        ...number(1, Valtype.i32),
+        [ Opcodes.return ],
+
+        [ Opcodes.end ],
+
+        // increment iter pointer by sizeof i16
+        [ Opcodes.local_get, IterPointer ],
+        ...number(ValtypeSize.i16, Valtype.i32),
+        [ Opcodes.i32_add ],
+        [ Opcodes.local_set, IterPointer ],
+
+        // increment counter by 1, check if eq length, if not loop
+        [ Opcodes.local_get, Counter ],
+        ...number(1, Valtype.i32),
+        [ Opcodes.i32_add ],
+        [ Opcodes.local_tee, Counter ],
+
+        [ Opcodes.local_get, Length ],
+        [ Opcodes.i32_ne ],
+        [ Opcodes.br_if, 1 ],
+
+        [ Opcodes.end ],
+
+        // no match, return 0
+        ...number(0, Valtype.i32)
+      ];
+
+      break;
+
+    case 'Character':
+      out = generateChar(node, get);
+      break;
+
+    case 'Set':
+      out = generateSet(node, get);
+      break;
+
+    case 'Group':
+      out = generateGroup(node, get);
+      break;
+
+    case 'Range':
+      out = generateRange(node, get);
+      break;
+  }
+
+  return out;
+};
+
+const getNextChar = () => [
+  // get char from pointer
+  [ Opcodes.local_get, Pointer ],
+  [ Opcodes.i32_load16_u, Math.log2(ValtypeSize.i16) - 1, ...unsignedLEB128(0) ],
+
+  // pointer += sizeof i16
+  [ Opcodes.local_get, Pointer ],
+  ...number(ValtypeSize.i16, Valtype.i32),
+  [ Opcodes.i32_add ],
+  [ Opcodes.local_set, Pointer ]
+];
+
+const checkFailure = () => [
+  // surely we do not need to do this for every single mismatch, right?
+  /* [ Opcodes.if, Blocktype.void ],
+  ...number(0, Valtype.i32),
+  [ Opcodes.return ],
+  [ Opcodes.end ], */
+
+  [ Opcodes.br, 1 ]
+];
+
+const generateChar = (node, get) => {
+  return [
+    ...(get ? getNextChar() : []),
+    ...number(String.fromCharCode(node.char), Valtype.i32),
+    [ Opcodes.i32_ne ],
+    ...checkFailure()
+  ];
+};
+
+const generateSet = (node, get) => {
+  const out = [
+    ...(get ? getNextChar() : []),
+    [ Opcodes.local_set, Tmp ],
+  ];
+
+  for (const x of node.body) {
+    out = [
+      ...out,
+      [ Opcodes.local_get, Tmp ],
+      ...generate(x, false)
+    ];
+  }
+
+  out = out.concat(new Array(node.body.length - 1).fill([ Opcodes.i32_or ]));
+
+  return [
+    ...out,
+    ...checkFailure()
+  ];
+};
+
+const generateGroup = (node, get) => {
+
+};
+
+const generateRange = (node, get) => {
+
+};
+
+
+export const match = regex => {
+
+};
+
+export const test = regex => {
+  const code = generate(parse(regex));
+};
+
+const wasmify = code => {
+  const funcs = [{
+    name: 'exp',
+    export: true,
+    params: [ Valtype.i32 ],
+    returns: [ Valtype.i32 ],
+    locals: {
+      iterPointer: { idx: 1, type: Valtype.i32 },
+      counter: { idx: 2, type: Valtype.i32 },
+      pointer: { idx: 3, type: Valtype.i32 },
+      length: { idx: 4, type: Valtype.i32 },
+      tmp: { idx: 5, type: Valtype.i32 },
+    }
+  }]
+};

package/compiler/rhemyn/parse.js

@@ -0,0 +1,125 @@
+const State = {
+  none: 0,
+  insideSet: 1
+};
+
+const Quantifiers = {
+  '*': [ 0 ], // 0 -
+  '+': [ 1 ], // 1 -
+  '?': [ 0, 1 ], // 0 - 1
+};
+const QuantifierKeys = Object.keys(Quantifiers);
+
+export default str => {
+  const out = {
+    type: 'Expression',
+    body: []
+  };
+  let node = out, parents = [];
+
+  let state = State.none, escape = false;
+  for (let i = 0; i < str.length; i++) {
+    const c = str[i];
+
+    const addChar = () => {
+      node.body.push({
+        type: 'Character',
+        char: c
+      });
+    };
+
+    const seek = () => {
+      const cNext = str[++i];
+
+      if (cNext === '\\') return str[++i];
+      return cNext;
+    };
+
+    if (escape) {
+      addChar();
+
+      escape = false;
+      continue;
+    }
+
+    if (c === '\\') {
+      escape = true;
+      continue;
+    }
+
+    switch (state) {
+      case State.none:
+        if (c === '[') {
+          parents.push(node);
+          node = {
+            type: 'Set',
+            body: []
+          };
+
+          parents.at(-1).body.push(node);
+
+          state = State.insideSet;
+          continue;
+        }
+
+        if (c === '(') {
+          parents.push(node);
+          node = {
+            type: 'Group',
+            body: []
+          };
+
+          parents.at(-1).body.push(node);
+          continue;
+        }
+
+        if (c === ')') {
+          if (node.type !== 'Group') throw new SyntaxError('Unmatched closing parenthesis');
+
+          node = parents.pop();
+          continue;
+        }
+
+        if (QuantifierKeys.includes(c)) {
+          const amount = Quantifiers[c];
+          node.body.at(-1).quantifier = amount;
+          continue;
+        }
+
+        addChar();
+        break;
+
+      case State.insideSet:
+        if (c === ']') {
+          state = State.none;
+          node = parents.pop();
+
+          continue;
+        }
+
+        // range
+        if (c === '-') {
+          const from = node.body.pop().char;
+          const to = seek();
+
+          node.body.push({
+            type: 'Range',
+            from,
+            to
+          });
+          continue;
+        }
+
+        addChar();
+        break;
+    }
+  }
+
+  // still in a group by the end
+  if (node.type !== 'Expression') throw new SyntaxError('Unmatched opening parenthesis');
+
+  // still in a set by the end
+  if (state === State.insideSet) throw new SyntaxError('Unmatched opening square bracket');
+
+  return out;
+};

package/compiler/rhemyn/test/parse.js

@@ -0,0 +1,20 @@
+import util from 'node:util';
+
+import parse from '../parse.js';
+
+const tests = {
+  'a': {},
+  'a(b)': {},
+  'a(b(c))': {},
+  'ab': {},
+  '[ab]': {},
+  '[a-z]': {},
+  'a*': {},
+  'a+': {},
+  'a?': {},
+  'a(b)+': {}
+}
+
+for (const str in tests) {
+  console.log(str, util.inspect(parse(str), false, null, true));
+}

package/compiler/sections.js

@@ -8,11 +8,26 @@ const createSection = (type, data) => [
   ...encodeVector(data)
 ];
 
+const customSection = (name, data) => [
+  Section.custom,
+  ...encodeVector([...encodeString(name), ...data])
+];
+
+const chHint = (topTier, baselineTier, strategy) => {
+  // 1 byte of 4 2 bit components: spare, top tier, baseline tier, compilation strategy
+  // tiers: 0x00 = default, 0x01 = baseline (liftoff), 0x02 = optimized (turbofan)
+  // strategy: 0x00 = default, 0x01 = lazy, 0x02 = eager, 0x03 = lazy baseline, eager top tier
+  return (strategy | (baselineTier << 2) | (topTier << 4));
+};
+
 export default (funcs, globals, tags, pages, flags) => {
   const types = [], typeCache = {};
 
   const optLevel = parseInt(process.argv.find(x => x.startsWith('-O'))?.[2] ?? 1);
 
+  const compileHints = process.argv.includes('-compile-hints');
+  if (compileHints) log('sections', 'warning: compile hints is V8 only w/ experimental arg! (you used -compile-hints)');
+
   const getType = (params, returns) => {
     const hash = `${params.join(',')}_${returns.join(',')}`;
     if (optLog) log('sections', `getType(${JSON.stringify(params)}, ${JSON.stringify(returns)}) -> ${hash} | cache: ${typeCache[hash]}`);
@@ -36,7 +51,7 @@ export default (funcs, globals, tags, pages, flags) => {
     // tree shake imports
     for (const f of funcs) {
       for (const inst of f.wasm) {
-        if (inst[0] === Opcodes.call && inst[1] < importedFuncs.length) {
+        if ((inst[0] === Opcodes.call || inst[0] === Opcodes.return_call) && inst[1] < importedFuncs.length) {
           const idx = inst[1];
           const func = importedFuncs[idx];
 
@@ -51,10 +66,11 @@ export default (funcs, globals, tags, pages, flags) => {
     // fix call indexes for non-imports
     const delta = importedFuncs.length - importFuncs.length;
     for (const f of funcs) {
+      f.originalIndex = f.index;
       f.index -= delta;
 
       for (const inst of f.wasm) {
-        if (inst[0] === Opcodes.call && inst[1] >= importedFuncs.length) {
+        if ((inst[0] === Opcodes.call || inst[0] === Opcodes.return_call) && inst[1] >= importedFuncs.length) {
           inst[1] -= delta;
         }
       }
@@ -73,6 +89,14 @@ export default (funcs, globals, tags, pages, flags) => {
     encodeVector(funcs.map(x => getType(x.params, x.returns))) // type indexes
   );
 
+  // compilation hints section - unspec v8 only
+  // https://github.com/WebAssembly/design/issues/1473#issuecomment-1431274746
+  const chSection = !compileHints ? [] : customSection(
+    'compilationHints',
+    // for now just do everything as optimise eager
+    encodeVector(funcs.map(_ => chHint(0x02, 0x02, 0x02)))
+  );
+
   const globalSection = Object.keys(globals).length === 0 ? [] : createSection(
     Section.global,
     encodeVector(Object.keys(globals).map(x => [ globals[x].type, 0x01, ...number(globals[x].init ?? 0, globals[x].type).flat(), Opcodes.end ]))
@@ -145,6 +169,7 @@ export default (funcs, globals, tags, pages, flags) => {
     ...typeSection,
     ...importSection,
     ...funcSection,
+    ...chSection,
     ...memorySection,
     ...tagSection,
     ...globalSection,

package/compiler/wrap.js

@@ -90,6 +90,15 @@ export default async (source, flags = [ 'module' ], customImports = {}, print =
             return Array.from(new Uint16Array(memory.buffer, pointer + 4, length)).map(x => String.fromCharCode(x)).join('');
           }
 
+          case 'function': {
+            // wasm func index, including all imports
+            const func = funcs.find(x => (x.originalIndex ?? x.index) === ret);
+            if (!func) return ret;
+
+            // make fake empty func for repl/etc
+            return {[func.name]() {}}[func.name];
+          }
+
           default: return ret;
         }
       } catch (e) {

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "porffor",
   "description": "a basic experimental wip aot optimizing js -> wasm engine/compiler/runtime in js",
-  "version": "0.0.0-bddcdc3",
+  "version": "0.0.0-ebc0491",
   "author": "CanadaHonk",
   "license": "MIT",
   "dependencies": {

package/r.js

@@ -0,0 +1 @@
+/a(b)/.test('hi');

package/runner/index.js

@@ -3,8 +3,26 @@
 import compile from '../compiler/wrap.js';
 import fs from 'node:fs';
 
+if (process.argv.includes('-compile-hints')) {
+  const v8 = await import('node:v8');
+  v8.setFlagsFromString(`--experimental-wasm-compilation-hints`);
+
+  // see also these flags:
+  // --experimental-wasm-branch-hinting
+  // --experimental-wasm-extended-const
+  // --experimental-wasm-inlining (?)
+  // --experimental-wasm-js-inlining (?)
+  // --experimental-wasm-return-call (on by default)
+}
+
 const file = process.argv.slice(2).find(x => x[0] !== '-');
 if (!file) {
+  if (process.argv.includes('-v')) {
+    // just print version
+    console.log((await import('./version.js')).default);
+    process.exit(0);
+  }
+
   // run repl if no file given
   await import('./repl.js');
 

package/runner/repl.js

@@ -1,14 +1,7 @@
 import compile from '../compiler/wrap.js';
+import rev from './version.js';
 
 import repl from 'node:repl';
-import fs from 'node:fs';
-
-let rev = 'unknown';
-try {
-  rev = fs.readFileSync(new URL('../.git/refs/heads/main', import.meta.url), 'utf8').trim().slice(0, 7);
-} catch {
-  rev = JSON.parse(fs.readFileSync(new URL('../package.json', import.meta.url), 'utf8')).version.split('-')[1];
-}
 
 // process.argv.push('-O0'); // disable opts
 
@@ -48,7 +41,7 @@ const memoryToString = mem => {
   return out;
 };
 
-const alwaysPrev = process.argv.includes('-always-prev');
+const alwaysPrev = process.argv.includes('-prev');
 
 let prev = '';
 const run = async (source, _context, _filename, callback, run = true) => {
@@ -56,7 +49,7 @@ const run = async (source, _context, _filename, callback, run = true) => {
   if (alwaysPrev) prev = toRun + ';\n';
 
   const { exports, wasm, pages } = await compile(toRun, []);
-  fs.writeFileSync('out.wasm', Buffer.from(wasm));
+  // fs.writeFileSync('out.wasm', Buffer.from(wasm));
 
   if (run && exports.$) {
     lastMemory = exports.$;

package/runner/transform.js

@@ -8,7 +8,8 @@ const source = fs.readFileSync(file, 'utf8');
 const { wasm } = await compile(source);
 
 // const out = `(async () => { const print = str => process.stdout.write(str); (await WebAssembly.instantiate(Uint8Array.from([${wasm.toString()}]), {'': { p: i => print(i.toString()), c: i => print(String.fromCharCode(i))}})).instance.exports.m()})()`;
-const out = `new WebAssembly.Instance(new WebAssembly.Module(new Uint8Array([${wasm.toString()}])),{'':{p:i=>process.stdout.write(i.toString())}}).exports.m()`;
+// const out = `new WebAssembly.Instance(new WebAssembly.Module(new Uint8Array([${wasm.toString()}])),{'':{p:i=>process.stdout.write(i.toString())}}).exports.m()`;
+const out = `const a=new WebAssembly.Instance(new WebAssembly.Module(new Uint8Array([${wasm.toString()}])));const b=a.exports.m();console.log(Array.from(new Uint16Array(a.exports.$.buffer,b+4,new Int32Array(a.exports.$.buffer,b,1))).map(x=>String.fromCharCode(x)).join(''))`;
 
 console.log(out);
 eval(out);

package/runner/version.js

@@ -0,0 +1,10 @@
+import fs from 'node:fs';
+
+let rev = 'unknown';
+try {
+  rev = fs.readFileSync(new URL('../.git/refs/heads/main', import.meta.url), 'utf8').trim().slice(0, 7);
+} catch {
+  rev = JSON.parse(fs.readFileSync(new URL('../package.json', import.meta.url), 'utf8')).version.split('-')[1].slice(0, 7);
+}
+
+export default rev;