Fix microtask drain re-entrancy crashing async-generator yield rejection#465
Merged
Fix microtask drain re-entrancy crashing async-generator yield rejection#465
Conversation
TGocciaMicrotaskQueue.DrainQueue kept already-processed tasks in the queue until a final FQueue.Clear at the outermost drain's exit. When a microtask handler triggered a recursive drain — for example via AwaitValue on a settled promise calling DrainMicrotasksAndFetchCompletions — the inner drain re-iterated from index 0 and re-ran every prior task. Async generators yielding a rejected promise (or yielding via yield* over a non-iterable) hit this on every test262 run: iter.next() #1 awaited the yielded rejection, the outer Promise rejected, the .catch handler fired in the next drain, called iter.next() #2, which in turn drained microtasks... until the call stack overflowed (SIGSEGV). DrainQueue now pops each task with FQueue.Delete(0) before invoking it, so recursive drains only see remaining or newly-enqueued tasks. Also updates scripts/test262_syntax_filter.py to mark generators, async-generator, WeakMap, WeakSet, and symbols-as-weakmap-keys as supported (shipped in #432, #437) and removes the syntax patterns and SKIP_PATH_SEGMENTS that previously hid those tests from the suite. test262 (bytecode mode, ASI + compat-var/function/Function on): Eligible 39,025 / 47,416 — passes 19,470 → 19,595, errors 465 → 0. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
Contributor
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (2)
🚧 Files skipped from review as they are similar to previous changes (1)
📝 WalkthroughWalkthroughEnables generator-related and WeakMap/WeakSet syntax in the test filter, rewrites the microtask queue drain to avoid reprocessing tasks during re-entrancy, and adds regression tests for microtask ordering and async-generator rejection/iteration edge cases. Changes
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes Possibly related PRs
Suggested labels
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. Review rate limit: 4/5 reviews remaining, refill in 12 minutes. Comment |
coderabbitai
Bot
added
bug
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 1
🧹 Nitpick comments (1)
source/units/Goccia.MicrotaskQueue.pas (1)
104-109: AvoidDelete(0)on every microtask.Line 109 shifts the entire
TListon each iteration, so drainingnmicrotasks becomes O(n²). That’s fine for landing the correctness fix, but Promise-heavy workloads can turn this into a hotspot. A head index or queue/deque structure would keep the re-entrancy fix without the per-item memmove.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@source/units/Goccia.MicrotaskQueue.pas` around lines 104 - 109, The loop currently calls FQueue.Delete(0) which shifts the whole TList each iteration (O(n²)); change the queue to use a head index to make pops O(1): add a field like FHead: Integer, push items with FQueue.Add as before, read the next task with Task := FQueue[FHead] and then increment FHead instead of calling FQueue.Delete(0); update the loop condition to while FHead < FQueue.Count do (or use Count = FQueue.Count - FHead), and when FHead = FQueue.Count reset FQueue.Clear and FHead := 0 (or periodically compact by moving remaining items to index 0 when FHead grows past a threshold). Apply these changes around the existing CheckExecutionTimeout / CheckInstructionLimit calls so behavior is unchanged but per-item memmoves are removed.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@tests/language/async-generators/yield-rejected-promise.js`:
- Around line 26-34: The current promise only handles the rejection path of the
first iter.next(), so if iter.next() unexpectedly fulfills the test will hang;
modify the first iter.next() call to handle both fulfillment and rejection: on
fulfillment immediately reject (or throw) the outer promise with a clear error
(e.g. "first iter.next() unexpectedly fulfilled"), and on rejection continue to
capture firstRejection and chain the subsequent iter.next() as before
(references: iter.next(), firstRejection, chainSettled).
---
Nitpick comments:
In `@source/units/Goccia.MicrotaskQueue.pas`:
- Around line 104-109: The loop currently calls FQueue.Delete(0) which shifts
the whole TList each iteration (O(n²)); change the queue to use a head index to
make pops O(1): add a field like FHead: Integer, push items with FQueue.Add as
before, read the next task with Task := FQueue[FHead] and then increment FHead
instead of calling FQueue.Delete(0); update the loop condition to while FHead <
FQueue.Count do (or use Count = FQueue.Count - FHead), and when FHead =
FQueue.Count reset FQueue.Clear and FHead := 0 (or periodically compact by
moving remaining items to index 0 when FHead grows past a threshold). Apply
these changes around the existing CheckExecutionTimeout / CheckInstructionLimit
calls so behavior is unchanged but per-item memmoves are removed.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 99c09c90-7f27-4d66-9f92-a8361ec63553
📒 Files selected for processing (4)
scripts/test262_syntax_filter.pysource/units/Goccia.MicrotaskQueue.pastests/built-ins/Promise/microtask-ordering.jstests/language/async-generators/yield-rejected-promise.js
Contributor
Benchmark Results407 benchmarks Interpreted: 🟢 60 improved · 🔴 38 regressed · 309 unchanged · avg +1.3% arraybuffer.js — Interp: 🟢 3, 11 unch. · avg +3.1% · Bytecode: 🔴 11, 3 unch. · avg -6.6%
arrays.js — Interp: 🟢 4, 🔴 1, 14 unch. · avg +1.5% · Bytecode: 🔴 14, 5 unch. · avg -5.2%
async-await.js — Interp: 🔴 3, 3 unch. · avg -1.6% · Bytecode: 🔴 3, 3 unch. · avg -8.5%
async-generators.js — Interp: 2 unch. · avg -0.2% · Bytecode: 🔴 1, 1 unch. · avg -13.0%
base64.js — Interp: 🟢 3, 🔴 2, 5 unch. · avg +0.6% · Bytecode: 🔴 5, 5 unch. · avg -3.1%
classes.js — Interp: 🟢 1, 🔴 1, 29 unch. · avg -0.3% · Bytecode: 🟢 10, 🔴 2, 19 unch. · avg +3.2%
closures.js — Interp: 🟢 1, 🔴 1, 9 unch. · avg -0.8% · Bytecode: 🔴 6, 5 unch. · avg -3.0%
collections.js — Interp: 🔴 1, 11 unch. · avg +0.0% · Bytecode: 🔴 12 · avg -12.3%
csv.js — Interp: 🟢 1, 🔴 1, 11 unch. · avg -1.1% · Bytecode: 🟢 3, 🔴 1, 9 unch. · avg +0.2%
destructuring.js — Interp: 22 unch. · avg +0.5% · Bytecode: 🟢 1, 🔴 7, 14 unch. · avg -2.1%
fibonacci.js — Interp: 8 unch. · avg +0.5% · Bytecode: 🔴 6, 2 unch. · avg -7.6%
float16array.js — Interp: 🟢 8, 🔴 3, 21 unch. · avg +1.1% · Bytecode: 🟢 4, 🔴 21, 7 unch. · avg -7.2%
for-of.js — Interp: 7 unch. · avg +1.7% · Bytecode: 🔴 6, 1 unch. · avg -9.8%
generators.js — Interp: 🔴 1, 3 unch. · avg -2.5% · Bytecode: 🔴 4 · avg -11.8%
iterators.js — Interp: 🟢 3, 🔴 4, 35 unch. · avg +0.1% · Bytecode: 🟢 15, 🔴 8, 19 unch. · avg +1.6%
json.js — Interp: 🟢 4, 16 unch. · avg +1.9% · Bytecode: 🔴 5, 15 unch. · avg -3.9%
jsx.jsx — Interp: 🟢 2, 🔴 5, 14 unch. · avg -0.9% · Bytecode: 🟢 7, 🔴 3, 11 unch. · avg +1.6%
modules.js — Interp: 9 unch. · avg -1.3% · Bytecode: 🔴 7, 2 unch. · avg -6.7%
numbers.js — Interp: 🔴 1, 10 unch. · avg -1.9% · Bytecode: 🟢 1, 🔴 4, 6 unch. · avg -3.5%
objects.js — Interp: 🔴 3, 4 unch. · avg -2.0% · Bytecode: 🟢 3, 4 unch. · avg +2.3%
promises.js — Interp: 🔴 1, 11 unch. · avg -0.9% · Bytecode: 🔴 1, 11 unch. · avg -0.2%
regexp.js — Interp: 11 unch. · avg -0.1% · Bytecode: 🟢 2, 🔴 2, 7 unch. · avg -0.9%
strings.js — Interp: 🟢 5, 14 unch. · avg +0.9% · Bytecode: 🔴 14, 5 unch. · avg -4.3%
tsv.js — Interp: 🟢 1, 🔴 1, 7 unch. · avg -0.1% · Bytecode: 🟢 7, 2 unch. · avg +5.5%
typed-arrays.js — Interp: 🟢 9, 🔴 5, 8 unch. · avg -0.6% · Bytecode: 🟢 8, 🔴 8, 6 unch. · avg -4.4%
uint8array-encoding.js — Interp: 🟢 6, 🔴 3, 9 unch. · avg +1.3% · Bytecode: 🟢 14, 🔴 1, 3 unch. · avg +49.1%
weak-collections.js — Interp: 🟢 9, 🔴 1, 5 unch. · avg +31.3% · Bytecode: 🟢 1, 🔴 13, 1 unch. · avg -28.4%
Measured on ubuntu-latest x64. Benchmark ranges compare cached main-branch min/max ops/sec with the PR run; overlapping ranges are treated as unchanged noise. Percentage deltas are secondary context. |
Contributor
Suite Timing
Measured on ubuntu-latest x64. |
- yield-rejected-promise.js: the first iter.next() handler only covered the rejection path, so an engine regression that fulfilled instead would hang the test until the runner's timeout. Now both branches are registered: fulfillment fails the outer Promise with a clear error, rejection captures firstRejection and chains the next call as before. - Goccia.MicrotaskQueue.pas: replace TList.Delete(0) (O(n) shift per pop, O(n^2) per drain) with a head index. FHead advances as tasks run and the underlying list is compacted (Clear + reset) once the queue drains. HasPending and ClearQueue follow the head; the re-entrancy guarantee from the previous commit is preserved because nested drains see only tasks past FHead. Co-Authored-By: Claude Opus 4.7 (1M context) <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
1 participant
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.
Summary
DrainQueue— for example, an async generator that yields a rejected promise, whereiter.next()'s rejection handler synchronously callsiter.next()again. The previousTGocciaMicrotaskQueue.DrainQueuekept already-processed tasks in the queue until the outermost drain's finalFQueue.Clear, so any nested drain re-iterated from index 0 and re-ran every prior task. The fix pops each task withFQueue.Delete(0)before invoking it, so recursive drains only see remaining or newly-enqueued tasks.scripts/test262_syntax_filter.pyto markgenerators,async-generator,WeakMap,WeakSet, andsymbols-as-weakmap-keysas supported features (shipped via Add generator and async generator support #432 and Implement WeakMap and WeakSet built-ins #437) and removes the syntax patterns andSKIP_PATH_SEGMENTSentries that previously hid those tests from the suite.yield Promise.reject(…),yield*of an iterable whose[Symbol.iterator]returns null, generator that throws synchronously,iter.next()from a rejection handler).test262 baseline (bytecode, ASI +
--compat-var/--compat-function/--unsafe-function-constructor)All 465 previously-crashing async-generator tests now run to completion. 253 of them pass; the rest fail cleanly with assertion errors (mostly spec-compliance gaps in
AsyncGeneratorYield's rejection handling — separate work).Test plan
./build.pas testrunner loader— both binaries build clean./build/GocciaTestRunner tests --asi --unsafe-ffi --no-progress— 8195 / 8200 pass (5 pre-existing FFI fixture failures, unchanged)async function* () { yield Promise.reject(0); yield 1; }withiter.next().catch(() => iter.next())) no longer crashes the bytecode VM🤖 Generated with Claude Code