You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 72bd975
Browse filesBrowse the repository at this point in the historyBrowse files
test: run the suite with Sparkplug off, which is what kills the #405 segfault (#477)
#405's crash has an upstream name — nodejs/node#62393, V8 CL 0b94a9fd23ba: the
BaselineOutOfLinePrologue builtin left a random value on the stack where the GC
found it. `ClearStaleLeftTrimmedPointerVisitor` reads it as a heap pointer and
faults at 0xe, the address in all five local reports. Sparkplug generates that
prologue, so `--no-sparkplug` removes the path.
The two decisions #405 was holding both answer no. Node cannot be bumped into the
fix: the backport (nodejs/node#65753) is open, not landed, so no released 24.x has
it, and one upstream report has the crash live on v26.7.0. CI needs no dead-child
retry: every occurrence here and upstream is macOS arm64, and every workflow runs
ubuntu-latest.
Cost is nothing measurable — 55.8 s mean either way over two runs each locally,
~3% on Linux x64 over the two heaviest jsdom files, against a 22-minute review
step. The flag is on the test script only, on purpose: `npm start` and `npm run
dev` keep the Sparkplug path, because one dev server dying is loud where a dead
test child reads as a clean run with a short pass count.
A test pins the flag, and after round 1 it pins its *position*: `node --test
"glob" --no-sparkplug` exits 0, warns nothing, and the child's `execArgv` does not
carry the flag, so a presence-only check could go green through the exact
regression it exists to catch. Drop the flag and that test together when
nodejs/node#65753 ships in a 24.x release.
Two rounds, both approved; round 1's three notes fixed, round 2 clean.
Refs #405
Co-Authored-By: Claude Opus 5 <[email protected]>
Copy file name to clipboardExpand all lines: CONTRIBUTING.md
+17-6Lines changed: 17 additions & 6 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -61,12 +61,23 @@ npm test # the unit suite (node --test; run it through np
61
61
./test/e2e.sh # full API lifecycle against mock GitHub + mock model (needs jq)
62
62
```
63
63
64
-
Run the unit suite through `npm`, not as a bare `node --test`. The `npm test` script
65
-
registers a second reporter (`test/spec-with-signals.mjs`) that prints `signal` and
66
-
`exitCode` when a test file's *process* dies. Node's default reporter shows that as `✖
67
-
some.test.ts` and `'test failed'` — identical to a failed assertion, with nothing on
68
-
stderr — and the tests after the death simply never run, so the pass count reads clean
69
-
while being short. See #405.
64
+
Run the unit suite through `npm`, not as a bare `node --test`. The script carries two
65
+
things a bare run does not:
66
+
67
+
-**A second reporter** (`test/spec-with-signals.mjs`) that prints `signal` and `exitCode`
68
+
when a test file's *process* dies. Node's default reporter shows that as `✖ some.test.ts`
69
+
and `'test failed'` — identical to a failed assertion, with nothing on stderr — and the
70
+
tests after the death simply never run, so the pass count reads clean while being short.
71
+
-**`--no-sparkplug`**, which avoids a V8 bug that segfaults test children inside the
72
+
garbage collector roughly once in ten full runs on macOS arm64. The crash needs code
73
+
Sparkplug generates, so turning that tier off removes the path; it cost nothing
74
+
measurable here (two runs each, 55.8 s either way). It is on the test script only: `npm start`
75
+
and `npm run dev` keep the Sparkplug path on purpose, because one dev server dying is loud,
76
+
where a dead test child reads as a clean run with a short count. **Drop the flag when
77
+
[nodejs/node#65753](https://github.com/nodejs/node/pull/65753) ships in a 24.x release** —
78
+
it is the backport of the V8 fix, and no released 24.x has it yet.
79
+
80
+
Both exist because of #405, which has the crash stack and the evidence.
70
81
71
82
The demo page must stay accessible — it's audited with the project's own axe-core lint and
0 commit comments