Conversation
ThreatCrush Security Scan4586 finding(s) HIGH/CRITICAL: 43 | MEDIUM: 4032 | LOW: 511
…and 4536 more. Full results in the Security tab. Snippets are redacted; ThreatCrush never prints matched credential material. |
timoteo7
force-pushed
the
fix/pre-json-into-side-pnpm-concurrency
branch
from
September 23, 2026 22:04
ce712e5 to
c0dbd01
Compare
timoteo7
changed the title
timoteo7
force-pushed
the
fix/pre-json-into-side-pnpm-concurrency
branch
from
September 23, 2026 22:35
464c5f4 to
b84f2ef
Compare
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
| FNXC:TriagePlanningRecovery 2026-09-19-04:04: | ||
| The sweep announced "Recovering specified triage task <id>" on every poll for cards it never | ||
| recovered, and recorded no outcome when the recovery declined or when its own gates skipped the | ||
| candidate. The recorded engine log held 180 announcements and zero reasons, so an operator could |
| paused: false, | ||
| steps: [{ title: "Implement", status: "pending" }], | ||
| workflowStepResults: [ | ||
| { workflowStepId: "plan-review", workflowStepName: "Plan Review", phase: "pre-merge", status: "passed", verdict: "APPROVE" }, |
timoteo7
force-pushed
the
fix/pre-json-into-side-pnpm-concurrency
branch
3 times, most recently
from
September 24, 2026 00:26
82d59cc to
4af29b0
Compare
timoteo7
pushed a commit
that referenced
this pull request
Sep 24, 2026
…ing-spin + graph identity + session watchdog) FNXC:BoardReliability 2026-09-24-00:46: Consolidates the board-reliability fixes (the engine theme, kept separate from verify-perf in #11): 1. SOURCE-OF-TRUTH stranded count (stranded-commits.ts): fork-point + "0 task-unique = 0 stranded"; a failed fork-point falls back to ZERO, never the raw base. listStrandedCommits() uses the single source (the 2nd call site). 2. Planning-spin: an unchanged-but-valid authoritative PROMPT.md is SUCCESS (was re-planning forever). 3. Graph-failure identity: a parse/compile failure names the failing stage, not node 'unknown'. 4. Session-idle watchdog: aborts a runaway session (budget 250 tools / idle 30min) with RUNAWAY_SESSION (call fn_task_done). 5. Manual retry clears ANY non-user pause. Plus the typecheck step stays READ-ONLY (tsc --build --noEmit). Symptom Verification: - Original symptom: "N stranded commits since <base>" treadmill + planning-spin + node 'unknown' + 4h runaway sessions. - Exact reproduction: a task branch at the main tip vs merge-base; a valid unchanged PROMPT.md; a session that never calls fn_task_done. - Assertion it is gone: countStrandedCommits()==0 for shared lineage; planning accepts a valid unchanged spec; the watchdog aborts runaway sessions; the typecheck emits nothing (read-only).
timoteo7
pushed a commit
that referenced
this pull request
Sep 24, 2026
…ead-only) FNXC:VerifyPerf 2026-09-24-01:00: MOVED from #17 (board-reliability) to #11 (verify-perf theme, per review). The typecheck step keeps the incremental tsbuildinfo cache (warm runs recheck only changed files) while staying READ-ONLY (--noEmit: no dist emit, no partial dist on a killed typecheck). Symptom Verification: warm verify typecheck rechecks only changed files; the step emits nothing (read-only).
timoteo7
changed the title
timoteo7
force-pushed
the
fix/pre-json-into-side-pnpm-concurrency
branch
from
September 25, 2026 04:15
2150a29 to
0dfb730
Compare
Owner
Author
|
Review notes for the audited SHA The TypeScript build-cache change is worth keeping: a focused compiler check confirmed that One description needs correction: Git's built-in |
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
2 participants
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.
Problem
Two root causes parked task runs for hours on the 2026-09-23 nightly fleet (12GB desktop):
.changeset/pre.jsoncorruption on merge/rebase — when a branch rebased across achore(release)commit (e.g.v0.78.0-beta.6) and the conflict was resolved from the older side, the beta anchor broke andcheck-pre-json-anchorfailed deterministically on that branch. Two review tasks sat terminal-failed 8–9h with verification blocked. Same regression class as fix(RUFU-075): safest scheduler queue/mission-store writes for minimal stores Runfusion/Fusion#3472.pnpm -rdefaults to 4 concurrent projects, and pnpm 10.33 silently ignores the~/.npmrc workspace-concurrencykey (it echoes the value; behavior stays at 4). Each task typecheck fanned out 4tsc --noEmitworkers (500MB–1.4GB each) and thrashed the machine: memory PSIfull94%, engine HTTP timeouts, runs stuck in swap for 70+ minutes.Fix
.gitattributes:.changeset/pre.json merge=binary— built-in git merge driver that keeps the copy from the branch being merged INTO (the exact rulescripts/check-pre-json-anchor.mjsprescribes manually). No per-clone configuration needed — the commonmerge=oursrecipe requiresgit config merge.ours.driver truein every clone and otherwise leaves full conflict markers. The check stays as detection; this prevents the corruption at conflict-resolution time.pnpm-workspace.yaml:workspaceConcurrency: 2— caps recursive package commands at 2 projects at a time (the live pnpm 10.x config channel). Scope is onepnpm -rinvocation only. The manifest comment now states why the value stays a constant — pnpm 10 has no live env channel for it, and this is the only backstop for an ad-hocpnpm -rno script wraps — and that it is a blunt guardrail overridable per command with--workspace-concurrency, not a project requirement. Callers this repo owns already pass that flag explicitly (scripts/test-changed.mjsderives it fromFUSION_TEST_WORKSPACE_CONCURRENCY).scripts/verify-fast.mjs:TYPECHECK_PARALLEL_LIMITnow derives from RAM instead of core count — this is the guard that actually covers the typecheck fan-out of root cause 2.verify:fastspawns onepnpm --filter <pkg> typecheckprocess per changed package, and those bypassworkspaceConcurrencyentirely, so the key above cannot bound them. The oldmin(4, cpuCount / 2)gave 4 here;tscis memory-bound (1–1.4GB RSS each), so the ceiling is now 1 worker per 5GB of total RAM, capped at 4 — 2 on this 12GB box, and a 64GB machine is no longer slowed down by it.FUSION_VERIFY_FAST_TYPECHECK_CONCURRENCYoverrides the derived default, alongside the existingFUSION_VERIFY_FAST_SERIAL. Measured 2026-09-23: the 4-wide batch sat in D state at 3–7% CPU per worker for 10+ min with zero progress, drove memory PSIfullto 80%, and froze the desktop twice.STATIC_CHECK_PARALLEL_LIMITis untouched (those steps are node-startup dominated, not memory bound).scripts/dev-with-memory.mjs: the dev heap ceiling wasFUSION_DEV_MEMORY_MB || "8192"— a constant sized after one workstation, handing every node child an 8GB heap ceiling on a 12GB box. It now derives fromos.totalmem()(half of RAM, clamped 1–8GB → 6381MB here) with the env override kept. The wrapper also stopped prepending a second--max-old-space-sizewhenNODE_OPTIONSalready pins one (callers like the local Fusion launcher pin 2048, and precedence between repeated V8 flags is not worth relying on).Tests
pre.jsonrebase conflict leaves<<<<<<<markers with the older side able to win (bug reproduced); withmerge=binarythe file resolves to the INTO side automatically.node --check scripts/verify-fast.mjsclean; importing the module asserts the derived ceiling lands in 1–4 (2 on this 12GB box) and thatbuildTypecheckStep("@fusion/core", { hasTypecheck: true }).parallelLimitequals that constant rather than a literal, so a future edit cannot silently leave the step at 4. Override path checked too:FUSION_VERIFY_FAST_TYPECHECK_CONCURRENCY=3yields 3, while a garbage or0value falls back to the derived default. Exits non-zero if any assertion breaks.node --check scripts/dev-with-memory.mjsclean; formula check asserts the derived value lands in the 1–8GB clamp (6381MB on this 12GB box, no clamping in play here), and the NODE_OPTIONS dedup asserts three cases — an inherited--max-old-space-sizeis left untouched, an empty inheritance injects one, and other inherited flags (--trace-warnings) are preserved. Exits non-zero if any assertion breaks.node scripts/check-fnxc-future-dates.mjs— 0 future-dated stamps (both FNXC notes use adate -ustamp).TYPECHECK_PARALLEL_LIMITacross the tree: no test asserts the old value of 4.pnpm install --frozen-lockfile --ignore-scripts --offline,pnpm check:workspace-package-graph,pnpm check:pre-json— all exit 0.Changeset
None — repository build/merge configuration only; no published package behavior changes.
Implements FUSI-021 / FUSI-022.