Finish converting JS compiler to ES6 modules. NFC (#21542)

The compiler itself is now purely JS modules.

As part of this change I create a new global context object in which
all library code is processed.  This allows is to have a clean
separation between the compiler code itself and code being processed.
Only symbols explicitly added to the macro context are available to JS
library code.

Author: sbc100

ChangeLog.md

@@ -20,6 +20,10 @@ See docs/process.md for more on how version tagging works.
 
 3.1.57 (in development)
 -----------------------
+- JS library code is now executed in its own context/scope, which limits how
+  much of the compiler internals are accessible. If there are build time JS
+  symbols that you are depending on, but that were not added to this scope,
+  please file a bug and we can add more to this scope. (#21542)
 
 3.1.56 - 03/14/24
 -----------------

src/compiler.mjs

@@ -9,21 +9,9 @@
 
 import * as fs from 'fs';
 import * as path from 'path';
-import * as vm from 'vm';
 import * as url from 'url';
-import assert from 'assert';
 
-globalThis.vm = vm;
-globalThis.assert = assert;
-globalThis.nodePath = path;
-
-globalThis.print = (x) => {
-  process.stdout.write(x + '\n');
-};
-
-globalThis.printErr = (x) => {
-  process.stderr.write(x + '\n');
-};
+import {Benchmarker, applySettings, assert, loadSettingsFile, printErr, read} from './utility.mjs';
 
 function find(filename) {
   assert(filename);
@@ -38,22 +26,9 @@ function find(filename) {
   return filename;
 }
 
-globalThis.read = (filename) => {
-  assert(filename);
-  const absolute = find(filename);
-  return fs.readFileSync(absolute).toString();
-};
-
-function load(f) {
-  vm.runInThisContext(read(f), {filename: find(f)});
-}
-
-// Basic utilities
-load('utility.js');
-
 // Load default settings
-load('./settings.js');
-load('./settings_internal.js');
+loadSettingsFile(find('settings.js'));
+loadSettingsFile(find('settings_internal.js'));
 
 const argv = process.argv.slice(2);
 const symbolsOnlyArg = argv.indexOf('--symbols-only');
@@ -64,11 +39,10 @@ if (symbolsOnlyArg != -1) {
 // Load settings from JSON passed on the command line
 const settingsFile = argv[0];
 assert(settingsFile);
+const user_settings = JSON.parse(read(settingsFile));
+applySettings(user_settings);
 
-const settings = JSON.parse(read(settingsFile));
-Object.assign(global, settings);
-
-globalThis.symbolsOnly = symbolsOnlyArg != -1;
+export const symbolsOnly = symbolsOnlyArg != -1;
 
 // In case compiler.js is run directly (as in gen_sig_info)
 // ALL_INCOMING_MODULE_JS_API might not be populated yet.
@@ -88,18 +62,21 @@ if (symbolsOnly) {
 
 // Side modules are pure wasm and have no JS
 assert(
-  !SIDE_MODULE || (ASYNCIFY && globalThis.symbolsOnly),
+  !SIDE_MODULE || (ASYNCIFY && symbolsOnly),
   'JS compiler should only run on side modules if asyncify is used.',
 );
 
 // Load compiler code
 
-load('modules.js');
-load('parseTools.js');
-load('jsifier.js');
+// We can't use static import statements here because several of these
+// file depend on having the settings defined in the global scope (which
+// we do dynamically above.
+await import('./modules.mjs');
+await import('./parseTools.mjs');
 if (!STRICT) {
-  load('parseTools_legacy.js');
+  await import('./parseTools_legacy.mjs');
 }
+const jsifier = await import('./jsifier.mjs');
 
 // ===============================
 // Main
@@ -108,7 +85,7 @@ if (!STRICT) {
 const B = new Benchmarker();
 
 try {
-  runJSify();
+  jsifier.runJSify(symbolsOnly);
 
   B.print('glue');
 } catch (err) {

src/jsifier.js renamed to src/jsifier.mjs

@@ -4,20 +4,46 @@
  * SPDX-License-Identifier: MIT
  */
 
-// "use strict";
-
 // Convert analyzed data to javascript. Everything has already been calculated
 // before this stage, which just does the final conversion to JavaScript.
 
-globalThis.addedLibraryItems = {};
-
-globalThis.extraLibraryFuncs = [];
+import {
+  ATEXITS,
+  ATINITS,
+  ATMAINS,
+  defineI64Param,
+  indentify,
+  makeReturn64,
+  modifyJSFunction,
+  preprocess,
+  processMacros,
+  receiveI64ParamAsI53,
+} from './parseTools.mjs';
+import {
+  addToCompileTimeContext,
+  assert,
+  error,
+  errorOccured,
+  isDecorator,
+  isJsOnlySymbol,
+  compileTimeContext,
+  print,
+  printErr,
+  read,
+  warn,
+  warningOccured,
+} from './utility.mjs';
+import {LibraryManager, librarySymbols} from './modules.mjs';
+
+const addedLibraryItems = {};
+
+const extraLibraryFuncs = [];
 
 // Some JS-implemented library functions are proxied to be called on the main
 // browser thread, if the Emscripten runtime is executing in a Web Worker.
 // Each such proxied function is identified via an ordinal number (this is not
 // the same namespace as function pointers in general).
-globalThis.proxiedFunctionTable = [];
+const proxiedFunctionTable = [];
 
 // Mangles the given C/JS side function name to assembly level function name (adds an underscore)
 function mangleCSymbolName(f) {
@@ -121,7 +147,7 @@ function preJS() {
   return result;
 }
 
-function runJSify() {
+export function runJSify(symbolsOnly) {
   const libraryItems = [];
   const symbolDeps = {};
   const asyncFuncs = [];
@@ -162,7 +188,6 @@ function runJSify() {
         .split(',')
         .map((name) => name.trim());
       const newArgs = [];
-      let innerArgs = [];
       let argConversions = '';
       if (sig.length > argNames.length + 1) {
         error(`handleI64Signatures: signature too long for ${symbol}`);
@@ -188,7 +213,6 @@ function runJSify() {
         }
       }
 
-      var origArgs = args;
       if (!WASM_BIGINT) {
         args = newArgs.join(',');
       }
@@ -266,7 +290,7 @@ function(${args}) {
       ((i53abi && sig.includes('j')) || ((MEMORY64 || CAN_ADDRESS_2GB) && sig.includes('p')))
     ) {
       snippet = handleI64Signatures(symbol, snippet, sig, i53abi);
-      i53ConversionDeps.forEach((d) => deps.push(d));
+      compileTimeContext.i53ConversionDeps.forEach((d) => deps.push(d));
     }
 
     const proxyingMode = LibraryManager.library[symbol + '__proxy'];
@@ -348,7 +372,7 @@ function(${args}) {
       if (!(symbol in LibraryManager.library)) {
         // Create a new __cxa_find_matching_catch variant on demand.
         const num = +symbol.split('_').slice(-1)[0];
-        addCxaCatch(num);
+        compileTimeContext.addCxaCatch(num);
       }
       // Continue, with the code below emitting the proper JavaScript based on
       // what we just added to the library.
@@ -376,7 +400,7 @@ function(${args}) {
         LibraryManager.library[symbol + '__deps'] = [];
       }
 
-      deps = LibraryManager.library[symbol + '__deps'];
+      const deps = LibraryManager.library[symbol + '__deps'];
       let sig = LibraryManager.library[symbol + '__sig'];
       if (!WASM_BIGINT && sig && sig[0] == 'j') {
         // Without WASM_BIGINT functions that return i64 depend on setTempRet0
@@ -466,12 +490,11 @@ function(${args}) {
           // Create a stub for this symbol which can later be replaced by the
           // dynamic linker.  If this stub is called before the symbol is
           // resolved assert in debug builds or trap in release builds.
+          let target = `wasmImports['${symbol}']`;
           if (ASYNCIFY) {
             // See the definition of asyncifyStubs in preamble.js for why this
             // is needed.
             target = `asyncifyStubs['${symbol}']`;
-          } else {
-            target = `wasmImports['${symbol}']`;
           }
           let assertion = '';
           if (ASSERTIONS) {
@@ -536,8 +559,6 @@ function(${args}) {
       if (VERBOSE) {
         printErr(`adding ${symbol} (referenced by ${dependent})`);
       }
-      const deps_list = deps.join("','");
-      const identDependents = symbol + `__deps: ['${deps_list}']`;
       function addDependency(dep) {
         // dependencies can be JS functions, which we just run
         if (typeof dep == 'function') {
@@ -695,7 +716,7 @@ var proxiedFunctionTable = [
 `);
     }
 
-    if (abortExecution) {
+    if (errorOccured()) {
       throw Error('Aborting compilation due to previous errors');
     }
 
@@ -716,7 +737,7 @@ var proxiedFunctionTable = [
       '//FORWARDED_DATA:' +
         JSON.stringify({
           librarySymbols,
-          warnings,
+          warnings: warningOccured(),
           asyncFuncs,
           ATINITS: ATINITS.join('\n'),
           ATMAINS: STRICT ? '' : ATMAINS.join('\n'),
@@ -741,7 +762,13 @@ var proxiedFunctionTable = [
     finalCombiner();
   }
 
-  if (abortExecution) {
+  if (errorOccured()) {
     throw Error('Aborting compilation due to previous errors');
   }
 }
+
+addToCompileTimeContext({
+  extraLibraryFuncs,
+  addedLibraryItems,
+  preJS,
+});