Repository navigation
Conversation
When Emscripten's WASM memory grows (due to -sALLOW_MEMORY_GROWTH), all existing TypedArray views become detached because the underlying ArrayBuffer is replaced. This caused bugs where reading from views after FFI calls returned undefined values if memory had grown. The fix introduces RefreshableTypedArray, a wrapper that lazily recreates the TypedArray view when HEAPU8.buffer changes. This is a simple reference comparison that only triggers view recreation when actually needed. Affected call sites: - runtime.ts: executePendingJobs - reads ctxPtrOut after QTS_ExecutePendingJob - context.ts: newPromise - reads resolve/reject handles after QTS_NewPromiseCapability - context.ts: getLength - reads uint32Out after QTS_GetLength - context.ts: getOwnPropertyNames - reads outPtr and uint32Out after QTS_GetOwnPropertyNames Also fixed: getOwnPropertyNames was using HEAP8.buffer instead of HEAPU8.buffer Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
- getArrayBuffer returned a HEAPU8.subarray view that is detached by the next WASM memory growth; return a copy instead. - consumeJSCharPointer / consumeHeapCharPointer leaked the C string when UTF8ToString threw (e.g. strings longer than V8's max string length). - newArrayBuffer leaked the host buffer (and returned an exception-tagged handle) when QuickJS failed to create the ArrayBuffer. It now frees the buffer and throws. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…reelist - ModuleMemory helpers now throw on a null _malloc instead of writing at address 0. - getOwnPropertyNames freed a plain-malloc'd array with js_free (QTS_FreeVoidPointer), unbalancing QuickJS allocation counters; free it with the module's free. - HostRefMap: FIFO freelist used Array#shift (quadratic when draining); use a head index. Deleting an already-deleted id in a non-empty group no longer queues it for reuse twice. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…y arguments - QTS_RuntimeDumpMemoryUsage wrote into a fixed 1 KiB fmemopen buffer, truncating the report mid-line once more than a few object kinds exist. Use open_memstream. - QTS_Call built a variable-length array of argc values on the wasm shadow stack; a few hundred thousand arguments trapped the whole module. Heap allocate instead, and throw out-of-memory if that fails. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…wnPropertyNames - newContext (sync and asyncify) wrapped a NULL context pointer and carried on; throw. - ContextMemory.heapValueHandle wrapped the NULL returned by jsvalue_to_heap when it cannot malloc; throw. - QTS_GetOwnPropertyNames did not check its output malloc, and sized it per JSValue (8 bytes) instead of per JSValue* (4 bytes); check it, free the property atoms and throw out-of-memory on failure. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
QTS_bjson_encode writes with JS_WRITE_OBJ_REFERENCE so shared and cyclic objects encode, but QTS_bjson_decode read with flags 0, so decoding them failed with "object references are not allowed". Pass JS_READ_OBJ_REFERENCE. Bytecode and SharedArrayBuffer remain disallowed. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…cannot abort
Asyncify's buffer holds the saved locals of each suspended wasm frame, which needs up to
~1.9x the native stack depth that JS_SetMaxStackSize limits (measured across release/debug,
quickjs/quickjs-ng and several recursion shapes). With the buffer set to exactly the limit,
suspending near the deepest recursion QuickJS allows overflowed it and aborted the module
("unreachable"). Set it to 3x the limit (at least the build default), capped at 1 GiB.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
This was referenced Oct 16, 2026
Open
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.
Layer: C interface (
c/interface.c,Asyncify.StackSizeset at runtime) + docs. No build-flag changeReview just this PR: jackkav/quickjs-emscripten@fix-bjson-references...fix-asyncify-stack
Stack: #248 → #276 → #277 → (#278 | #279 → #280 → #281 → #282). Issues: #274, #275.
What
unreachable): the asyncify buffer was sized to exactly the QuickJS stack limit but needs 1.0–1.9x more (measured over 6 recursion shapes, release/debug, both engines).setMaxStackSizenow sizes it to 3x the limit (≥ build default, ≤ 1 GiB), so QuickJS throws its catchablestack overflowfirst.setMaxStackSizecall the default stays 81920 and a ~300-deep suspension can still abort. Fixing the default costs 768 KiB (quickjs) to 3 MiB (quickjs-ng) of transient heap per suspended call; that is your call, happy to add it.Test: 3 recursion shapes (plain at two limits): find the deepest recursion QuickJS allows, suspend there, expect completion. All fail on wasm built from
main(the abort also poisons later tests in the module).Array#map/applyshapes only run at the smaller limit to stay under node's own call stack.Verified on: Full suite on freshly built wasm for every variant in the test matrix (sync/async × release/debug × quickjs/quickjs-ng) at the top of the stack (#282): 751 passed, 0 failed. Debug builds run the leak sanitizer check in
afterAll.Behavior change: With an explicit
setMaxStackSize(n), each suspended async call transiently allocates 3n bytes (was n).Overlap: Touches
QTS_RuntimeSetMaxStackSize; no overlap with #207 beyond possible textual conflicts inc/interface.c.🤖 Generated with Claude Code