test(#4515): migrate installer-runs batch to named timeout constants (#4575)

* test(#4515): migrate installer-runs batch to named timeout constants

Batch 4 of the ad hoc timeout literal migration (epic #4445). Replaces
every bare numeric timeout/timeoutMs object-literal property across
tests/install.test.cjs, tests/install-minimal-hooks.test.cjs,
tests/fragment-single-edit-propagation.install.test.cjs,
tests/install-regressions.test.cjs, tests/install-runtime-artifacts.test.cjs,
tests/npm-integrity-gate.test.cjs, tests/faulty-deps.test.cjs,
tests/release-tarball-smoke.install.test.cjs,
tests/install-write-confinement.test.cjs, tests/plugin-manifest.test.cjs,
and tests/packaging-shipped-scripts-require-only-shipped.test.cjs with a
named constant, per eslint-rules/no-adhoc-timeout-literal.cjs. Removes
these 11 files from the rule's allowlist.

Adds one new shared constant to tests/helpers/timeouts.cjs for a
fixture-JSON value (in seconds, not ms) mimicking a Claude Code
settings.json hook-entry's own timeout field, shared across two files in
this batch. Every other new constant is file-local, each carrying a
comment explaining why its call site is a distinct operational class from
the existing shared norms even where its digits numerically coincide with
one. No src/bin file touched, no numeric timeout value changed anywhere.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

* fix: combine gsd-worktree-path-guard's git rev-parse calls to cut subprocess count under CI load

Discovered while verifying PR #4575 (a full test (windows-latest) failure in an unrelated test, tests/kilo-upgrades.test.cjs's worktree-path-guard rejection test). Root cause: this hook's up-to-4 sequential git spawns (git-dir, branch, worktree-toplevel, file-toplevel), invoked inside a native-plugin's own 8000ms-capped subprocess wrapper, left zero margin for node/git startup overhead — the guard is deliberately fail-open on any timeout, so CI-load-induced latency in this chain silently downgrades a security block into an allow. Combines the first three git rev-parse calls (git-dir, branch via --abbrev-ref, worktree-toplevel) into ONE spawn instead of three, using git's documented multi-flag rev-parse output (one line per flag, in order) — verified empirically including the detached-HEAD edge case. No timeout value changed; this reduces subprocess COUNT, not any tolerance.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

* docs: add changeset fragment for the worktree-path-guard fix

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

---------

Co-authored-by: sim <sim@local>
Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
This commit is contained in:
Tom Boucher
2026-09-09 11:45:44 -04:00
committed by GitHub
parent 5823d2ec7a
commit 30468f16fe
15 changed files with 221 additions and 97 deletions

View File

@@ -23,6 +23,19 @@ const REPO_ROOT = path.join(__dirname, '..');
const { extractRequires } = require('./helpers/copy-script-fixture.cjs');
/**
* package.json's `prepack`/`prepare` runs a full `npm run build:lib` (tsc)
* before `npm pack` computes the file list — a bounded but non-trivial
* subprocess (~2s in isolation). Under the full parallel 28,000+-test CI
* matrix this has been observed to take 60.6s and trip a 60_000ms bound
* (#2931 gsd-test run d52d2ee4, linux-node24, duration_ms 60637.56),
* aborting this file's `before()` hook and cascading every sibling test to
* "cancelled". 120s keeps the bound finite (never unbounded, per the
* subprocess-timeout convention) while giving real headroom for a
* contended CI box.
*/
const NPM_PACK_TIMEOUT_MS = 120_000;
/**
* Resolve the tarball file list via `npm pack --dry-run --json`.
* This is the ACTUAL set of files that ship — not a hardcoded list — so adding
@@ -34,16 +47,7 @@ function resolveTarballFiles() {
encoding: 'utf-8',
shell: true, // Windows: npm is npm.cmd and needs a shell
stdio: ['pipe', 'pipe', 'pipe'],
// package.json's `prepack`/`prepare` runs a full `npm run build:lib`
// (tsc) before `npm pack` computes the file list — a bounded but
// non-trivial subprocess (~2s in isolation). Under the full parallel
// 28,000+-test CI matrix this has been observed to take 60.6s and trip
// a 60_000ms bound (#2931 gsd-test run d52d2ee4, linux-node24,
// duration_ms 60637.56), aborting this file's `before()` hook and
// cascading every sibling test to "cancelled". 120s keeps the bound
// finite (never unbounded, per the subprocess-timeout convention) while
// giving real headroom for a contended CI box.
timeout: 120_000,
timeout: NPM_PACK_TIMEOUT_MS,
});
const parsed = JSON.parse(raw);
return packListToPathSet(parsed);