Teach metadce about WASM_ESM_INTEGRATION

ChangeLog.md

@@ -43,6 +43,10 @@ See docs/process.md for more on how version tagging works.
   in the next release. (#27261)
 - The default value for `GROWABLE_ARRAYBUFFERS` was reverted to `0` since we
   found issues with Web API compatibility. (#27260)
+- Fixed `-sWASM_ESM_INTEGRATION` builds at `-O2` and above, which previously
+  failed in the JS optimizer's metadce (dead code elimination) pass. metadce now
+  understands the native ES import/export wasm boundary that this mode emits.
+  (#27217)
 
 6.0.2 - 07/01/26
 ----------------

test/codesize/test_codesize_hello_esm_integration.json

@@ -0,0 +1,29 @@
+{
+  "a.out.mjs": 671,
+  "a.out.mjs.gz": 401,
+  "a.out.support.mjs": 6397,
+  "a.out.support.mjs.gz": 2473,
+  "a.out.nodebug.wasm": 1688,
+  "a.out.nodebug.wasm.gz": 973,
+  "total": 8756,
+  "total_gz": 3847,
+  "sent": [
+    "a (fd_write)"
+  ],
+  "imports": [
+    "a (fd_write)"
+  ],
+  "exports": [
+    "b (memory)",
+    "c (__wasm_call_ctors)",
+    "d (main)"
+  ],
+  "funcs": [
+    "$__emscripten_stdout_close",
+    "$__emscripten_stdout_seek",
+    "$__stdio_write",
+    "$__towrite",
+    "$__wasm_call_ctors",
+    "$main"
+  ]
+}

test/codesize/test_codesize_minimal_esm_integration.json

@@ -0,0 +1,22 @@
+{
+  "a.out.mjs": 683,
+  "a.out.mjs.gz": 410,
+  "a.out.support.mjs": 3343,
+  "a.out.support.mjs.gz": 1350,
+  "a.out.nodebug.wasm": 75,
+  "a.out.nodebug.wasm.gz": 87,
+  "total": 4101,
+  "total_gz": 1847,
+  "sent": [],
+  "imports": [],
+  "exports": [
+    "a (memory)",
+    "b (__wasm_call_ctors)",
+    "c (add)",
+    "d (global_val)"
+  ],
+  "funcs": [
+    "$__wasm_call_ctors",
+    "$add"
+  ]
+}

test/js_optimizer/applyDCEGraphRemovals-esm-output.js

@@ -0,0 +1,15 @@
+function fd_write_impl() {}
+
+function fd_close_impl() {}
+
+function unused_import_impl() {}
+
+import { memory, main as _main, used_export as _used_export, memory as wasmMemory } from "./a.out.wasm";
+
+export { fd_write_impl as fd_write, fd_close_impl as fd_close };
+
+export { _main };
+
+var HEAP32;
+
+export { HEAP32 };

test/js_optimizer/applyDCEGraphRemovals-esm.js

@@ -0,0 +1,34 @@
+// WASM_ESM_INTEGRATION: unused wasm imports are dropped from the `export {..}`
+// that sends JS functions to the wasm, and unused wasm exports (including
+// internal ones like the indirect function table) from the `import {..}` that
+// receives them, keeping the two ES module interfaces in sync.
+
+function fd_write_impl() {
+}
+function fd_close_impl() {
+}
+function unused_import_impl() {
+}
+
+import {
+  memory,
+  __indirect_function_table,
+  main as _main,
+  used_export as _used_export,
+  unused_export as _unused_export,
+  memory as wasmMemory,
+} from './a.out.wasm';
+
+export {
+  fd_write_impl as fd_write,
+  fd_close_impl as fd_close,
+  unused_import_impl as unused_import,
+};
+
+export { _main };
+
+// MODULARIZE=instance runtime exports (bare form) must be left untouched.
+var HEAP32;
+export { HEAP32 };
+
+// EXTRA_INFO: { "unusedImports": ["unused_import"], "unusedExports": ["unused_export", "__indirect_function_table"] }

test/js_optimizer/applyImportAndExportNameChanges-esm-output.js

@@ -0,0 +1,13 @@
+function fd_write_impl() {}
+
+function fd_close_impl() {}
+
+import { memory, b as _main, c as _malloc, memory as wasmMemory } from "./a.out.wasm";
+
+export { fd_write_impl as d, fd_close_impl as e };
+
+export { _main };
+
+var HEAP32;
+
+export { HEAP32 };

test/js_optimizer/applyImportAndExportNameChanges-esm.js

@@ -0,0 +1,30 @@
+// WASM_ESM_INTEGRATION: minified wasm import/export names are applied to the
+// native ES import/export specifiers. Only the wasm-facing name of each
+// specifier is renamed; the JS-local binding name is left intact (including the
+// unaliased `memory`, whose local side must survive).
+
+function fd_write_impl() {
+}
+function fd_close_impl() {
+}
+
+import {
+  memory,
+  main as _main,
+  malloc as _malloc,
+  memory as wasmMemory,
+} from './a.out.wasm';
+
+export {
+  fd_write_impl as fd_write,
+  fd_close_impl as fd_close,
+};
+
+// Re-export of a wasm export: the local name (_main) is never in the mapping.
+export { _main };
+
+// MODULARIZE=instance runtime export (bare form): never in the mapping.
+var HEAP32;
+export { HEAP32 };
+
+// EXTRA_INFO: { "mapping": { "main": "b", "malloc": "c", "fd_write": "d", "fd_close": "e" } }

test/js_optimizer/emitDCEGraph-esm-output.js

@@ -0,0 +1,78 @@
+[
+ {
+  "name": "emcc$defun$fd_close_impl",
+  "reaches": [
+   "emcc$defun$helper"
+  ]
+ },
+ {
+  "name": "emcc$defun$fd_write_impl",
+  "reaches": []
+ },
+ {
+  "name": "emcc$defun$helper",
+  "reaches": []
+ },
+ {
+  "name": "emcc$defun$unused_import_impl",
+  "reaches": []
+ },
+ {
+  "name": "emcc$export$__indirect_function_table",
+  "export": "__indirect_function_table",
+  "reaches": []
+ },
+ {
+  "name": "emcc$export$_main",
+  "export": "main",
+  "reaches": [],
+  "root": true
+ },
+ {
+  "name": "emcc$export$_unused_export",
+  "export": "unused_export",
+  "reaches": []
+ },
+ {
+  "name": "emcc$export$_used_export",
+  "export": "used_export",
+  "reaches": [],
+  "root": true
+ },
+ {
+  "name": "emcc$export$memory",
+  "export": "memory",
+  "reaches": [],
+  "root": true
+ },
+ {
+  "name": "emcc$import$fd_close_impl",
+  "import": [
+   "env",
+   "fd_close"
+  ],
+  "reaches": [
+   "emcc$defun$fd_close_impl"
+  ]
+ },
+ {
+  "name": "emcc$import$fd_write_impl",
+  "import": [
+   "env",
+   "fd_write"
+  ],
+  "reaches": [
+   "emcc$defun$fd_write_impl"
+  ]
+ },
+ {
+  "name": "emcc$import$unused_import_impl",
+  "import": [
+   "env",
+   "unused_import"
+  ],
+  "reaches": [
+   "emcc$defun$unused_import_impl"
+  ]
+ }
+]

test/js_optimizer/emitDCEGraph-esm.js

@@ -0,0 +1,49 @@
+// WASM_ESM_INTEGRATION: the wasm<->JS boundary is expressed with native ES
+// import/export syntax rather than the `wasmImports` object and
+// `wasmExports['x']` member uses, and emitDCEGraph must build the same graph
+// from it.
+
+// JS functions implementing wasm imports.
+function fd_write_impl() {
+}
+function unused_import_impl() {
+}
+
+// A JS function only reachable from a wasm import edge.
+function helper() {
+}
+function fd_close_impl() {
+  helper();
+}
+
+// wasm exports received as ES imports. `memory` is imported twice (plain and
+// aliased) and must map to a single export node.
+import {
+  memory,
+  __indirect_function_table,
+  main as _main,
+  used_export as _used_export,
+  unused_export as _unused_export,
+  memory as wasmMemory,
+} from './a.out.wasm';
+
+// JS functions exported to the wasm module (the wasm imports).
+export {
+  fd_write_impl as fd_write,
+  fd_close_impl as fd_close,
+  unused_import_impl as unused_import,
+};
+
+// Re-export of a wasm export to the JS entry point: a top-level use that should
+// root the underlying `main` export, not be treated as a wasm import.
+export { _main };
+
+// MODULARIZE=instance runtime exports. These are plain JS bindings, not wasm
+// import edges, and must be left untouched (bare `export {..}` form).
+var HEAP32;
+var baz = () => {};
+export { HEAP32, baz };
+
+// Top-level uses: root the memory export (via the alias) and one wasm export.
+wasmMemory.buffer;
+_used_export();

test/js_optimizer/emitDCEGraph-sourcePhaseImports-output.js

@@ -0,0 +1,38 @@
+[
+ {
+  "name": "emcc$defun$fd_write_impl",
+  "reaches": []
+ },
+ {
+  "name": "emcc$export$_main",
+  "export": "main",
+  "reaches": [],
+  "root": true
+ },
+ {
+  "name": "emcc$export$_unused_export",
+  "export": "unused_export",
+  "reaches": []
+ },
+ {
+  "name": "emcc$export$_used_export",
+  "export": "used_export",
+  "reaches": [],
+  "root": true
+ },
+ {
+  "name": "emcc$export$memory",
+  "export": "memory",
+  "reaches": []
+ },
+ {
+  "name": "emcc$import$fd_write_impl",
+  "import": [
+   "env",
+   "fd_write"
+  ],
+  "reaches": [
+   "emcc$defun$fd_write_impl"
+  ]
+ }
+]

test/js_optimizer/emitDCEGraph-sourcePhaseImports.js

@@ -0,0 +1,29 @@
+// SOURCE_PHASE_IMPORTS combined with WASM_ESM_INTEGRATION: the glue emits a
+// source-phase import for the wasm module alongside the value-phase import that
+// binds the wasm exports. emitDCEGraph must skip the source-phase import
+// (its specifier has no `imported`) and only treat the value-phase `.wasm`
+// import as wasm export bindings.
+
+// Source-phase import of the wasm module itself. This is NOT a wasm export
+// binding and must be ignored by the DCE graph builder.
+import source wasmModule from './a.out.wasm';
+
+function fd_write_impl() {
+}
+
+// wasm exports received as ES imports (value-phase).
+import {
+  memory,
+  main as _main,
+  used_export as _used_export,
+  unused_export as _unused_export,
+} from './a.out.wasm';
+
+export {
+  fd_write_impl as fd_write,
+};
+
+export { _main };
+
+WebAssembly.instantiate(wasmModule);
+_used_export();