Repository navigation
wasm: Revert enable JSPI as already enabled - #60014
Closed
guybedford wants to merge 1 commit into
Closed
guybedford wants to merge 1 commit into
guybedford wants to merge 1 commit into
Conversation
Collaborator
|
Review requested:
|
nodejs-github-bot
added
c++
guybedford
force-pushed
the
disable-jspi
branch
from
September 25, 2025 18:11
ed56d7e to
c273e58
Compare
cjihrig
approved these changes
Sep 25, 2025
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #60014 +/- ##
==========================================
- Coverage 88.47% 88.46% -0.02%
==========================================
Files 703 703
Lines 207546 207545 -1
Branches 40006 40015 +9
==========================================
- Hits 183631 183601 -30
- Misses 15907 15946 +39
+ Partials 8008 7998 -10
🚀 New features to boost your workflow:
|
lpinca
approved these changes
Sep 25, 2025
Collaborator
anonrig
approved these changes
Sep 25, 2025
Collaborator
Collaborator
jasnell
approved these changes
Sep 28, 2025
Collaborator
Collaborator
guybedford
added a commit
that referenced
this pull request
Oct 1, 2025
PR-URL: #60014 Reviewed-By: Colin Ihrig <cjihrig@gmail.com> Reviewed-By: Luigi Pinca <luigipinca@gmail.com> Reviewed-By: Yagiz Nizipli <yagiz@nizipli.com> Reviewed-By: James M Snell <jasnell@gmail.com>
Contributor
Author
|
Landed in 23b8340. |
targos
added
dont-land-on-v20.x
dont-land-on-v22.x
9 tasks done
jkleinsc
pushed a commit
to electron/electron
that referenced
this pull request
Oct 29, 2026
* support_v8_sandboxed_pointers.patch Ref: nodejs/node#62237 * refactor_attach_cppgc_heap_on_v8_isolate_creation.patch Ref: nodejs/node#58070 * test_accomodate_v8_thenable_stack_trace_change_in_snapshot.patch Ref: nodejs/node#59805 * api_remove_deprecated_getisolate.patch Ref: nodejs/node#59805 Ref: nodejs/node#60223 * api_delete_deprecated_fields_on_v8_isolate.patch Ref: nodejs/node#60488 * api_promote_deprecation_of_v8_context_and_v8_object_api_methods.patch Ref: nodejs/node#60602 * src_refactor_module_wrap_cc_to_update_fixedarray_get_params.patch Ref: nodejs/node#61898 * src_refactor_wasmstreaming_finish_to_accept_a_callback.patch Ref: nodejs/node#61898 * src_stop_using_v8_propertycallbackinfo_t_this.patch Ref: nodejs/node#61898 * fix_add_externalpointertypetag_to_v8_external_api_calls.patch Ref: nodejs/node#61898 * src_stop_using_removed_v8_promise_reject_events.patch Ref: nodejs/node#62336 * src_backport_the_perfetto_trace_agent.patch Ref: nodejs/node#64565 Ref: nodejs/node#64721 * test_account_for_libc_rm_error_on_electron_linux.patch Ref: nodejs/node#66065 * fix_do_not_pass_removed_jspi_v8_flag.patch Ref: nodejs/node#60014 * test_drop_removed_wasm_rab_integration_flag.patch The v26 version of nodejs/node#65845 never added the flag. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU
MarshallOfSound
pushed a commit
to electron/electron
that referenced
this pull request
Oct 30, 2026
* chore: bump node in DEPS to v26.12.1 * chore: bump node in DEPS to v26.13.0 * fix(patch): build: add simdutf dir to include path in GN build Keep upstream's simdutf include dir in the node_js2c include_dirs and add "tools" next to it. The patch's FixedArray::Get(context, i) change in node_builtins.cc is no longer needed: upstream made the same change in nodejs/node#61898. Ref: nodejs/node#65382 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * fix(patch): vfs: integrate with CJS and ESM module loaders Upstream moved the native legacyMainResolve call from esm/resolve.js into the loaderMethods table in internal/modules/helpers.js, so pass the asar-aware fileExists callback from there instead. Ref: nodejs/node#63653 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * fix(patch): src: let embedders supply a builtin code cache without a snapshot BuiltinLoader::RefreshCodeCache() now merges with insert_or_assign upstream, so drop that hunk from the patch. Ref: nodejs/node#65352 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * fix(patch): src: move import.meta initializer to native land PathHelpersLazyGetter moved from node_modules.cc to module_wrap.cc and gained a sibling ImportMetaResolveLazyGetter; switch both to Holder(). Also follow the DatabaseSyncLimits -> DatabaseLimits rename in node_sqlite.cc (nodejs/node#65988) and the Null(isolate) cleanup in node_constants.cc (nodejs/node#59805). Ref: nodejs/node#60603 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * fix(patch): src: fix usage of deprecated V8 API Upstream switched to Get/SetContinuationPreservedEmbedderDataV2(), which Chromium's V8 marks deprecated in favor of the unsuffixed names. Use the unsuffixed getter and setter. Ref: nodejs/node#60174 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * fix(patch): src: rename legacy trace event headers The TRACING_CATEGORY_NODE* macros moved from node_internals.h to tracing/trace_event.h, so define NODE_TRACE_CATEGORY_ENABLED there. Agent::CreateDefault() picks the legacy agent again when V8_USE_PERFETTO is unset (nodejs/node#64053), so only the Perfetto branch is gated on NODE_USE_PERFETTO_SDK. node.gyp now lists the Perfetto sources in node_tracing_perfetto_sources. Add them, with V8's tracing target and config, to the GN libnode target. The removed perfetto backport patch used to carry that wiring. Ref: nodejs/node#64565 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * fix(patch): src: let embedders exempt linked bindings from the addon permission Upstream added EnvironmentFlags::kNoAddonPermissionForLinkedBindings around the same check. Keep the electron_* exemption alongside it and note in the patch description that the patch can be dropped once Electron sets the new flag. Ref: nodejs/node#66067 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * chore: remove upstreamed patches * support_v8_sandboxed_pointers.patch Ref: nodejs/node#62237 * refactor_attach_cppgc_heap_on_v8_isolate_creation.patch Ref: nodejs/node#58070 * test_accomodate_v8_thenable_stack_trace_change_in_snapshot.patch Ref: nodejs/node#59805 * api_remove_deprecated_getisolate.patch Ref: nodejs/node#59805 Ref: nodejs/node#60223 * api_delete_deprecated_fields_on_v8_isolate.patch Ref: nodejs/node#60488 * api_promote_deprecation_of_v8_context_and_v8_object_api_methods.patch Ref: nodejs/node#60602 * src_refactor_module_wrap_cc_to_update_fixedarray_get_params.patch Ref: nodejs/node#61898 * src_refactor_wasmstreaming_finish_to_accept_a_callback.patch Ref: nodejs/node#61898 * src_stop_using_v8_propertycallbackinfo_t_this.patch Ref: nodejs/node#61898 * fix_add_externalpointertypetag_to_v8_external_api_calls.patch Ref: nodejs/node#61898 * src_stop_using_removed_v8_promise_reject_events.patch Ref: nodejs/node#62336 * src_backport_the_perfetto_trace_agent.patch Ref: nodejs/node#64565 Ref: nodejs/node#64721 * test_account_for_libc_rm_error_on_electron_linux.patch Ref: nodejs/node#66065 * fix_do_not_pass_removed_jspi_v8_flag.patch Ref: nodejs/node#60014 * test_drop_removed_wasm_rab_integration_flag.patch The v26 version of nodejs/node#65845 never added the flag. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * chore: update patches (trivial only) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * chore: bump @types/node to v26 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * chore: update filenames.auto.gni Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * chore: remove deleted specs Upstream deleted parallel/test-promise-swallowed-event and parallel/test-warn-multipleResolves when the multipleResolves event reached end-of-life. Ref: nodejs/node#58707 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * test: clear crash dumps with fs.rmSync instead of recursive rmdir Node.js v25 moved the fs.rmdir recursive option to end-of-life, so fs.rmdirSync(dir, { recursive: true }) now throws and @types/node v26 no longer accepts the option. Ref: nodejs/node#58616 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * test: listen on IncomingMessage unions through EventEmitter With @types/node v26, http.IncomingMessage declares typed `on` overloads that do not unify with Electron.IncomingMessage's, so collectStreamBodyBuffer() no longer type-checks when calling `on` on the union. Ref: Unable to locate reference Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * fix: pass an ArrayBuffer-backed view to subtle.digest in asar wrapper @types/node v26 types webcrypto's BufferSource as views over an ArrayBuffer only, so a Uint8Array<ArrayBufferLike> no longer type-checks. Archive reads never use a SharedArrayBuffer, so narrow the type at the call site. Ref: Unable to locate reference Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * node#64565: src: rename legacy trace event headers Node.js renamed src/tracing/trace_event_common.h to trace_event_legacy.h and moved the legacy trace macros out of trace_event.h into trace_event_legacy_inl.h, so feed both to electron_generate_node_defines instead of the removed header. The generated push/undef set matches the previous one apart from header guards, NODE_TRACE_CATEGORY_ENABLED and INTERNAL_DECLARE_SET_TRACE_VALUE_INT, which Node.js no longer defines. Ref: nodejs/node#64565 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * fix(patch): src: add perfetto trace agent Upstream's trace_event_perfetto.h includes the amalgamated "perfetto.h" SDK header, which Chromium's Perfetto does not provide, so libnode fails to compile. Include the Perfetto public headers Chromium ships instead (as the removed backport patch did), add the tracing/trace_config includes agent_perfetto.h relied on, and skip Node's TrackEvent category definitions when Chromium's built-in categories are already in scope. Ref: nodejs/node#64565 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * fix(patch): src: handle indexed properties in `process.env` Upstream added indexed interceptors for process.env. Migrate the new EnvSetterIndexed and EnvDefinerIndexed callbacks to PropertyCallbackInfo<Boolean> like their named counterparts, since Chromium's V8 only accepts the Boolean-returning indexed setter and definer callbacks. Ref: nodejs/node#60826 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * test: read the whole heap snapshot stream Since Node.js 26, readable.read() returns one buffered chunk at a time instead of concatenating everything buffered, so takeHeapSnapshot() parsed only the first 64 KiB of the snapshot. Read readableLength bytes, as upstream test/common/heap.js now does. Ref: nodejs/node#60441 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * test: use --trace-warnings in the node-options-parsing crash case Node.js 26 throws ERR_MISSING_OPTION('--permission') when a boolean permission flag such as --allow-addons is set without --permission (previously only the fs flags were checked), so the utility process in this fixture exited before reaching options.js. Repeat --trace-warnings instead; the case still checks that disallowed NODE_OPTIONS are stripped from a list with repeated options. Ref: nodejs/node#62072 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * fix(patch): restore user timing trace events The usertiming.js hunk that emits node.perf.usertiming trace events for performance.mark() and performance.measure() lived in the perfetto backport patch, which was removed as upstreamed. Upstream Node.js never had it (Electron added it in #50591), so carry it in fix_use_embedder_perfetto_client_when_available.patch. Ref: #50591 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * test: trigger the V8 sandbox fatal error from V8 directly Since Node.js 26, node::Buffer::New() copies memory allocated outside the V8 sandbox into it instead of handing it to v8::ArrayBuffer::NewBackingStore(), so the external-ab addon no longer crashed the utility process and "emits 'error' when fatal error is triggered from V8" timed out. Call NewBackingStore() from the addon so the same v8_ArrayBuffer_NewBackingStore fatal error is still raised. Ref: nodejs/node#62237 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * test: delete directories with rm instead of recursive rmdir Node.js 26 removed the recursive option of fs.rmdir(), fs.rmdirSync() and fsPromises.rmdir() (DEP0147), so they now throw ERR_INVALID_ARG_VALUE. Use rmSync() and promises.rm(), and put an archive in the directory so the tests delete what their titles say. Ref: nodejs/node#58616 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * node#64894: worker: add support for Web Workers Node.js 26 adds a Web Worker API and its pre-execution deletes globalThis.Worker unless --experimental-web-worker is set, without checking noBrowserGlobals, so renderers and web workers with Node integration lost Blink's Worker. Pass the flag in those processes, as is already done for --experimental-eventsource; Node's own Worker is not installed under kNoBrowserGlobals. Ref: nodejs/node#64894 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * node#61803: module: remove --experimental-transform-types Node.js 26 removed the flag, so passing it now fails with "bad option" (exit code 9). Stop forwarding it to Node.js, drop it from the command-line switch docs, document the removal as a 46.0 breaking change, and remove the spec and fixture that exercised it. The spec that checks TypeScript-only syntax is not transformed stays. Ref: nodejs/node#61803 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * node#65589: src: disable V8 external memory reasonable size check Node.js now passes --external-memory-max-reasonable-size=0 to V8 in every process that initializes it, which changes FlagList::Hash() and so the code-cache key. Add the flag to the build-time js2c code cache flavors that initialize Node.js and to electron_xcache's default flags, so their caches keep being accepted (xcache.spec.ts failed on the flag hash). Ref: nodejs/node#65589 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * fix: route Node.js 26's one-shot fs jobs through the asar wrapper Node.js 26 writes files (fs.writeFile, fs.appendFile and their fsPromises versions) with a native WriteFileJob, and copies directory trees for fs.cp() / fsPromises.cp() with a native CpDirJob, instead of going through binding.open() and the fs.promises walk the asar wrapper already handles. Writes into archives therefore failed with ENOTDIR instead of EACCES (and missed unpacked entries), and fs.cp() of an archive directory failed with ENOTDIR. Give binding.WriteFileJob the same archive checks as open(), wrap fs.writeFile() for archive paths since lib/fs.js captured the native job at startup, and copy archive trees for CpDirJob with the walk cpSync() already uses, which now also stamps directories with preserveTimestamps like the native walk. Ref: nodejs/node#65489 Ref: nodejs/node#65488 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * fix(patch): do not leak NodePlatform's default TracingController The NodePlatform hunk that owns the v8::TracingController it creates when the embedder passes none lived in the perfetto backport patch, which was removed as upstreamed, but upstream still allocates it with a bare new. With the perfetto agent, ELECTRON_RUN_AS_NODE passes no controller, so LeakSanitizer failed every run-as-node spec on ASan builds. Carry the hunk in fix_use_embedder_perfetto_client_when_available.patch. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * fix: keep fsPromises.stat() and lstat() native outside archives The asar wrapper replaced fsPromises.stat() and fsPromises.lstat() with promisified callback versions for every path, so they ran as callback requests (FSREQCALLBACK) instead of fsPromises' own. Node.js 26's fsPromises.cp() checks its paths with them before handing the copy to a native job, and parallel/test-fs-cp-async-destination-appears-late tells the two apart. Use the wrapper only for paths inside archives. Ref: nodejs/node#65488 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * fix: match Node.js 26 recursive readdir in the asar wrapper Node.js 26 accepts Buffer paths for recursive readdir, keeps parentPath a Buffer for them and for the 'buffer' encoding, and walks the tree breadth-first, following links to directories without stat()ing every entry. The asar wrapper's own walks (which also descend into archives) broke on Buffer paths and differed in order and parentPath, failing the new parallel/test-fs-readdir-recursive-* tests. Replace the three walks with one breadth-first walk shaped like Node's, which still lists and descends into archives as directories, and lets errors from any directory reach the caller. Ref: nodejs/node#65487 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * node#66138: src: add --process-timeout=N Node.js starts the --process-timeout watchdog in NodeMainInstance::Run(), which ELECTRON_RUN_AS_NODE does not use, so the option was accepted but never applied: parallel/test-process-timeout-* failed, and the fixtures that block the main thread hung the Node.js test job until it was cancelled. Start and stop the watchdog around the event loop in NodeMain() the same way. Ref: nodejs/node#66138 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * test: close the FileHandle in the stale asar descriptor spec Node.js 26 made closing a FileHandle on garbage collection an error (DEP0137 end-of-life). The 'handle' variant of "does not capture a descriptor number that was closed behind its back" dropped its FileHandle, so the renderer running the asar specs later died with an uncaught ERR_INVALID_STATE and every remaining asar spec failed with "Render frame was disposed". Close the handle like the other variants close their descriptors. Ref: nodejs/node#58536 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * fix: allow ArrayBuffers above 2 GiB in Node.js-enabled isolates Node.js 26 allocates Buffer.allocUnsafe(), Buffer.allocUnsafeSlow() and the copies made by AdoptIntoBackingStore() through the isolate's ArrayBuffer::Allocator (nodejs/node#61480, nodejs/node#62237). Electron's sandboxed-pointers patch for Node.js 24 sent those through ArrayBuffer::Allocator::NewDefaultAllocator() instead, which reaches Electron's in-sandbox allocator, and that serves requests beyond PartitionAlloc's single-allocation cap (2 GiB - 2 MiB) from sandbox pages. The isolates of the browser process, the utility-process node service and ELECTRON_RUN_AS_NODE use gin's ArrayBufferAllocator, which has no such fallback, so with Node.js 26 a Buffer.allocUnsafe() of more than 2 GiB - 2 MiB in those processes failed with ERR_MEMORY_ALLOCATION_FAILED, where Electron 45 and 46 allowed up to 32 GiB - 1 (parallel/test-fs-write-buffer-large). Give those isolates an allocator that delegates to gin's partition up to the cap and serves larger requests from sandbox pages, sharing the page bookkeeping with the in-sandbox allocator. Buffer.alloc() and new ArrayBuffer() above the cap now work there as well. The page map is only locked while a page allocation is live. Renderers are unaffected: Blink's isolate allocator already allows giga allocations. Ref: nodejs/node#61480 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * fix: include node_watchdog.h only where --process-timeout is wired node_watchdog.h chooses between its POSIX and Windows members with Node's __POSIX__ define, which Node's own targets set but Electron's do not, so including it from node_includes.h broke every electron_lib source on Linux ("unknown type name 'BOOL'"). Include it in node_main.cc alone, with __POSIX__ set around it on POSIX so the class matches Node's own build. Ref: nodejs/node#66138 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * fix: let Node.js 26's virtual file system answer recursive readdir Node.js 26 adds an experimental virtual file system that fs functions consult (vfsState.handlers) before the real one. The asar wrapper walks recursive readdir itself and never asked it, so with a VFS mounted fs.readdir(..., { recursive: true }) and its sync and promise forms listed the real path behind the mount (parallel/test-vfs-reserved-root, parallel/test-vfs-zip-provider-readdir-recursive failed with ENOTDIR). Ask the VFS first, as Node does. Ref: nodejs/node#63537 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * fix: do not log terminated Electron init scripts as errors CompileAndCall() logged an error whenever running one of Electron's bundled init scripts threw, including when execution was terminated because the environment was being torn down. A Worker that is stopped, or whose process exits, while it is still bootstrapping is not an error. On Node.js 26 the unref()'d Worker spawned by module-hooks/test-async-loader-hooks-with-worker-permission-allowed reaches that point while node_init runs, and the log line broke the test's empty-stderr check. Only log real exceptions. Ref: Unable to locate reference Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * fix(patch): expect the Wasm Global setter WPT to pass with Chromium V8 Chromium's V8 passes "Calling setter without argument" in the Wasm jsapi WPT global/value-get-set.any.js, which Node's own V8 still fails and lists as an expected failure, so the run failed on an unexpected pass. Ref: Unable to locate reference Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * fix: honor --no-node-snapshot under ELECTRON_RUN_AS_NODE ELECTRON_RUN_AS_NODE accepted --no-node-snapshot but always created its isolate from the embedded Node startup snapshot, and Node worker threads inherit that snapshot from their parent. Since Node.js 26 a worker deserializes the whole bootstrapped context from it, so parallel/test-worker-nearheaplimit-deadlock, which passes --no-node-snapshot so that its 0 MB-heap worker does not run out of memory during that deserialization, crashed with SIGSEGV. Skip the snapshot when Node's options turn it off, as node::Start() does; the process then boots from the V8 context snapshot like a build without a Node snapshot. The browser process and the node service parse Node's options only after creating their isolate, so they are unaffected. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * build: fetch Perfetto trace_processor_shell for the Node.js tests nodejs/node#66068 (test: enable trace_events tests with perfetto) moved the trace_events tests from test/parallel into a new test/trace_events suite. When Node.js is built with Perfetto (process.config.variables .v8_use_perfetto, which Electron's build sets), those tests write binary .pftrace files and convert them back to JSON with Perfetto's trace_processor_shell, through test/common/trace_events.js. Without the tool, all 17 trace_events/* tests in nn-test fail with "trace_processor_shell is missing". The same PR added tools/perfetto/get_trace_processor, which upstream's Makefile and CI use to fetch it. Run that script in the Node.js test job. It downloads the Perfetto release matching the vendored deps/perfetto/VERSION (v58.2) from Perfetto's GitHub releases to tools/perfetto/trace_processor_shell, which is the default path the test helper resolves, so TRACE_PROCESSOR_SHELL_PATH is not needed. Cache the binary keyed on that VERSION file. Upstream pins only the version, not a checksum. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * test: disable Node.js source phase import and process-timeout report tests Electron always passes --no-js-source-phase-imports. In renderers Blink owns that V8 flag and freezes it before Node.js starts, and the other process types match it, so the flag is consistent across processes and the embedded code cache, which is keyed on V8's flag hash, stays valid. nodejs/node#60364 made Node.js honour that opt-out instead of always enabling the flag. The es-module/test-esm-wasm-*source-phase* tests are already disabled for the same reason. Node.js 26 brings more tests that need source phase imports: - es-module/test-esm-wasm-jitless (nodejs/node#66533) calls import.source() and fails with "SyntaxError: Unexpected identifier 'source'". - The wasm/jsapi WPT suite, test/wpt/test-wasm-jsapi.mjs (nodejs/node#59034), is new to Electron with this upgrade. Its esm-integration/source-phase*.tentative cases use import.source() and fail with the same SyntaxError. Its module/moduleSource.tentative cases expect the %AbstractModuleSource% intrinsic, which V8 only installs when the flag is on. Each case runs as an .any.html and an .any.worker.html variant, so that is 8 entries. nodejs/node#66138 added --process-timeout and report/test-report-process-timeout. nodejs/node#66428 split out report/test-report-process-timeout-worker. Both validate the generated report with test/common/report.js, which expects header.componentVersions to equal process.versions. In Electron, process.versions also has chrome and electron, so the comparison fails. The other report tests that use this helper are already disabled for the same reason. The process timeout itself works, and both tests get as far as the report check. The WPT cases are listed as wpt/<test file>/<case id>, which is how tools/test.py names them and matches --skip-tests. Teach --validateDisabled to check the test file for these entries. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * test: resolve V8's NewBackingStore by Electron's export name on Windows The external-ab addon now calls v8::ArrayBuffer::NewBackingStore() itself, but on Windows it is built with MSVC's standard library while Electron is built with Chromium's libc++, and MSVC mangles return types into symbol names. electron.exe exports the function as returning std::__Cr::unique_ptr, so the addon's import of the std::unique_ptr variant never resolved, the utility process died without a V8 fatal error and "emits 'error' when fatal error is triggered from V8" timed out on windows-x64. Look the function up by Electron's exported name there instead. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * test: only reference std-free V8 exports from the external-ab addon On macOS and Linux the addon is built against a different libc++ ABI namespace than Electron's std::__Cr, so v8::ArrayBuffer::New(isolate, std::shared_ptr<BackingStore>) is an import that cannot bind. It only went unnoticed because NewBackingStore() raises the fatal error first. Stop after NewBackingStore(), whose mangled name has no standard library types, and throw if it ever returns, as on Windows. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * Revert "build: fetch Perfetto trace_processor_shell for the Node.js tests" This reverts commit 8920a7c. Node.js uses trace_processor_shell only to read back the trace file that a process started with --trace-event-categories writes. Electron never writes that file: Chromium owns tracing, and Electron does not start Node.js's file-based tracing agent. Every trace_events/* test fails before it reaches trace_processor_shell, because no trace file exists or, in test-trace-events-api, because getEnabledCategories() is undefined in the child. Those tests are disabled again in the next commit, so no Node.js test needs trace_processor_shell, and downloading it would only add a network fetch to the job. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * test: disable the Node.js trace_events tests in their new directory nodejs/node#66068 (test: enable trace_events tests with perfetto) moved 17 test-trace-events-* tests out of test/parallel into a new test/trace_events suite. That took them out from under Electron's "parallel/test-trace-events" entry, which --skip-tests matches as a substring. Electron has never run these tests. They were disabled one by one when Electron started running the Node.js tests (1d06f67) and folded into the parallel/ prefix in #49467. They start Node.js with --trace-event-categories and read back the trace file it writes. Electron does not start Node.js's file-based tracing agent, because Chromium owns tracing, so no trace file is ever written. In nn-test, all 17 fail on a missing trace file, and test-trace-events-api also finds getEnabledCategories() undefined in its child. Add the matching "trace_events/test-trace-events" prefix. It covers all 17 files in test/trace_events. Keep the parallel/ entry, which still matches the 13 test-trace-events-* tests that stayed in test/parallel. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * fix: copy directory trees out of archives without blocking fs.cp() nodejs/node#65488 (fs: copy directory trees for fs.cp() on the thread pool) made fs.cp() and fsPromises.cp() copy a new destination directory with a native CpDirJob. The job walks the tree on the thread pool and calls ondone once. Electron's override for archive sources ran the synchronous fs.cpSync() walk in a single process.nextTick, so an asynchronous cp of a directory out of an archive blocked the event loop until the whole tree had been copied. Give the job an asynchronous version of the walk. It goes through the fs.promises overrides, so packed files use the existing asynchronous copyFile path, and it keeps the same ordering, force/errorOnExist, symlink and preserveTimestamps handling and errors as the sync walk. The job still reports through ondone exactly once. fs.cpSync() keeps the synchronous walk. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * test: trigger the utility process V8 fatal error from the testing binding The utilityProcess 'error' event from #43774 reports any V8 fatal error. Its spec raised one with the external-ab native addon, which handed V8 memory from outside the sandbox. Node.js 26 makes every Buffer site sandbox compatible, so that no longer happens through Node.js APIs. The earlier commits on this branch made the addon call V8 directly, and that needed ABI workarounds on Windows and for libc++. As Deepak suggested in review, drop the addon and its workspace entries. Add triggerV8FatalError() to the electron_common_testing binding, which is only built when DCHECKs are on. It fails a V8 API check by setting an internal field on a plain object, so it does not depend on the V8 sandbox. The spec keeps #43774's checks on the event type, location and report trigger, also checks the report's event message, and now runs wherever the testing binding exists, including Windows on Arm. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * test: wait for the utility process exit only after its 'error' event events.once(emitter, 'exit') also listens for 'error' and rejects with its first argument. The previous commit started waiting for 'exit' before the 'error' event arrived, so the utility process's 'FatalError' rejected that promise and the spec failed with "Unknown Error: FatalError". Wait for 'exit' after 'error' has been received, as the spec from #43774 did. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * refactor: use one d8-style PartitionAlloc allocator inside the V8 sandbox Following Deepak's review, replace the sandbox-page fallback added for nodejs/node#61480 and the older InSandboxAllocator with a single PAInSandboxAllocator, modelled on d8's. nodejs/node#61480 moved Buffer.allocUnsafe() onto the isolate's ArrayBuffer allocator. gin's partition caps a single allocation at partition_alloc::MaxAllocationSize() (2 GiB - 2 MiB), so Buffers above that size failed. The previous fix served those requests from whole sandbox pages and tracked them in a separate map. PAInSandboxAllocator owns a PartitionAlloc partition in the configurable pool, which gin::V8Initializer::Initialize() places inside the V8 sandbox. The partition is created lazily, after the pool exists. It has allow_giga_allocations set and allocates with kAllowGigaAllocations, along with kNoOverrideHooks and kNoMemoryToolOverride, as gin does. A single ArrayBuffer can therefore reach MaxGigaAllocationSize() (16 GiB - 2 MiB on 64-bit) without the page fallback. It is both the ArrayBuffer allocator for JavascriptEnvironment isolates and V8's in-sandbox allocator, so Node.js worker_threads isolates use it through NewDefaultAllocator(). Builds without the V8 sandbox keep gin's allocator. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU * fix: reject instead of throwing in asar fs.promises.stat/lstat wrappers The new fs.promises.stat/lstat wrappers call splitPath() before deciding whether to use fsPromises' own implementation. For a URL, splitPath() runs getValidatedPath(), which throws for a non-file: URL, so the error escaped synchronously instead of rejecting the returned promise as it did with util.promisify(fs.stat) on main and as Node's fsPromises does. Make the wrappers async so every input error rejects, and add a spec for it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vf2yMuJ7XxvNktpKuiMPzU --------- Co-authored-by: electron-roller[bot] <84116207+electron-roller[bot]@users.noreply.github.com> Co-authored-by: Claude <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This reverts #59941, while keeping the test for good measure.
It turns out JSPI is already enabled on latest V8, I was just not carefully testing this between different versions.
#59941 should be considered fully reverted and backed out with this change.