fix(ui): keep saved trails inside the project by danusha2345 · Pull Request #1667 · colbymchenry/codegraph

danusha2345 · 2026-09-01T08:14:44Z

Summary

Rebased onto v1.6.2 (6560052). The wall-clock replacement for the busiest-symbol test that this PR used to carry is dropped: #2208 did it on main. Changelog lines live in docs/viewer-launch-changelog.md.

reject not-yet-created paths when their nearest existing ancestor is a symlink outside the project
validate every listed trail through the project-root chokepoint and open it with O_NOFOLLOW | O_NONBLOCK
revalidate the trail directory after creation and use a random, exclusive 0600 temp file for atomic writes
cache saved-trail authors per real project root and reject declared unknown trail format versions
move the test workspace to Vitest 4 test.projects, align the Vite toolchain, and preserve Node >=25 contributor testing
update the dev dependency graph (the one remaining npm audit finding, devalue, is the same version as on main)

Reproductions fixed

.codegraph/ui -> outside with missing trails/ previously created outside/trails/<id>.json; the write is now refused.
.codegraph/ui/trails/leak.json -> outside/secret.json previously exposed a trail-shaped external JSON through the list endpoint; it is now skipped.
two repositories with different local Git user names in one process previously reused the first author; they now resolve independently.
a FIFO with no writer in .codegraph/ui/trails/ (a regression of the O_NOFOLLOW open in this PR, caught in review) blocked the synchronous open, and with it the whole UI server, on every list or read; it is now skipped like any other non-regular entry.

Positive controls keep projects opened through a safe root symlink and ordinary unversioned hand-written v1 trails working.

Verification

npm ci, npm run build, npx tsc --noEmit -p .
focused security / UI / trails suites, ui-package in the ui project, child-process-windows-hide: 1,034 passed, 4 skipped
full suite (Linux, Node 22, Vitest 4): everything passes except the two bundle-launcher Windows-archive listings, which hit Vitest's 5 s default timeout on this machine; the other timeouts seen under a loaded machine pass when re-run
compiled read/write/author reproductions all refuse or separate state as expected

🤖 Generated with Claude Code

danusha2345 · 2026-09-08T14:36:14Z

Merged current main (3adf067, post-#1770) into this branch: head afa27e2. The only conflict was the CHANGELOG entry; no source conflict. tsc clean, the branch's tests pass.

danusha2345 · 2026-09-08T14:40:39Z

Merged current main (43271f3) into this branch: head 0187193. The only conflict was the CHANGELOG entry; no source conflict. tsc clean, the branch's tests pass.

danusha2345 · 2026-09-08T14:55:26Z

Merged current main (85550eb) into this branch: head 7dad42e. The only conflict was the CHANGELOG entry; no source conflict. tsc clean, the branch's tests pass.

danusha2345 · 2026-09-08T15:09:57Z

Merged current main (8733c28) into this branch: head e6c1530. The only conflict was the CHANGELOG entry; no source conflict. tsc clean, the branch's tests pass.

danusha2345 · 2026-09-08T15:53:39Z

Merged current main (8c04734) into this branch: head 0a77a4b. The only conflict was the CHANGELOG entry; no source conflict. tsc clean, the branch's tests pass.

danusha2345 · 2026-09-08T16:31:25Z

Merged current main (ee83636) into this branch: head d0dbae3. The only conflict was the CHANGELOG entry; no source conflict. tsc clean, the branch's tests pass.

danusha2345 · 2026-09-08T16:52:36Z

Merged current main (e720f6c) into this branch: head b4dd834. The only conflict was the CHANGELOG entry; no source conflict. tsc clean, the branch's tests pass.

danusha2345 · 2026-09-08T19:06:34Z

Merged current main (de5adba) into this branch: head 10288d9. The only conflict was the CHANGELOG entry; no source conflict. tsc clean, the branch's tests pass.

danusha2345 · 2026-09-08T19:27:13Z

Merged current main (cece072) into this branch: head f2ebce7. The only conflict was the CHANGELOG entry; no source conflict. tsc clean, the branch's tests pass.

danusha2345 · 2026-09-08T20:13:19Z

Merged current main (040ba38) into this branch: head 50ceafc. The only conflict was the CHANGELOG entry; no source conflict. tsc clean, the branch's tests pass.

danusha2345 · 2026-09-08T20:50:49Z

Merged current main (374b3b4) into this branch: head 704ef5c. The only conflict was the CHANGELOG entry; no source conflict. tsc clean, the branch's tests pass.

danusha2345 · 2026-09-08T21:55:37Z

Merged current main (71d049c) into this branch: head c2f90ab. The only conflict was the CHANGELOG entry; no source conflict. tsc clean, the branch's tests pass.

danusha2345 · 2026-09-08T22:35:22Z

Merged current main (9181dd1) into this branch: head 0988b8f. The only conflict was the CHANGELOG entry; no source conflict. tsc clean, the branch's tests pass.

danusha2345 · 2026-09-08T22:47:26Z

Merged current main (64bd45c) into this branch: head e3e6e3c. The only conflict was the CHANGELOG entry; no source conflict. tsc clean, the branch's tests pass.

danusha2345 · 2026-09-09T00:07:17Z

Merged current main (748311f) into this branch: head 06f19d7. The only conflict was the CHANGELOG entry; no source conflict. tsc clean, the branch's tests pass.

danusha2345 · 2026-09-16T08:00:12Z

Merged main (8f80819) and adapted one test the Vitest 4 bump breaks: __tests__/git-index-currency.test.ts (from #1878) injects failures with vi.spyOn on a module it already vi.mocks. Under Vitest 4 that spy IS the mock's vi.fn, so the captured "real" function recursed into itself (Maximum call stack size exceeded) and, since restoreAllMocks no longer resets a vi.fn, the injection leaked into the six following tests — 7 of 13 red on a clean main checkout run with Vitest 4, 13/13 with Vitest 2. The rewrite routes the injections through the mocks' own vi.fns and reinstates the real functions explicitly; 13/13 under both 2.1.9 and 4.1.11. Full suite otherwise unchanged.

danusha2345 · 2026-09-16T12:37:25Z

One more Vitest 4 adaptation: #1883 passes --liftoff-only to pool workers through poolOptions.forks.execArgv, which Vitest 4 no longer reads, so on this branch the #1779 worker death (Fatal process out of memory: Zone in the turboshaft wasm tier-up) came back in a full run. Vitest 4's option is test.execArgv; a probe confirms the fork's process.execArgv now carries the flag and the 655-test extraction file passes. If #1883 lands first, this branch carries the Vitest 4 form of the same line.

…cbac0a; content already in integration

colbymchenry#1937/colbymchenry#1667 rebased)

danusha2345 · 2026-10-01T15:47:09Z

Merged current main (34ede4d) into this branch. The only conflict was __tests__/ui-server-api.test.ts with #2208. #2208 now times the engine's busiest symbol as the best of five warmed requests inside the regular test, which is what this branch's isolated median gate (npm run test:perf) was for, so I took main's file and dropped the gate together with the test:perf script; the branch no longer touches that test. What remains: the trail-path fixes (trail-store, trails, utils, trail-codec) with their tests, and the Vitest 4 migration they were written against. tsc clean; git-index-currency, ui-trails, ui-server-api, security, ui-package, viewer-gate and cli-ui-command pass under Vitest 4.1.11 (195 passed, 3 skipped: two Windows-only cases and the own-index test, which needs a local .codegraph/).

🤖 Generated with Claude Code

devEngines and the ui package's engines keep their Node 22.12 floor for the Vite 7 toolchain under Vitest 4, but drop the upper bound: it would stop npm itself on a newer Node, where contributors are meant to be able to run the suite (the test config sets CODEGRAPH_ALLOW_UNSAFE_NODE, colbymchenry#478). This commit used to add an isolated median timing gate for the viewer's busiest symbol, and a test:perf script to run it. On this base the regular test already times that request as the best of five (colbymchenry#2208), so both are left out. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

The O_NOFOLLOW open introduced in 746e774 runs before the isFile() check that used to reject non-regular entries. A named pipe with no writer in .codegraph/ui/trails/ therefore parked the synchronous open, and with it the whole UI server, on every list or read. Open with O_NONBLOCK as well: a regular file ignores it, a FIFO returns at once and fstat skips it as before. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

…uiltin Under Vitest 4 `vi.spyOn(cp, 'execFileSync')` on the module this file already mocks returns the same `vi.fn`, so the "real" function a test captured is the spy itself and the injected implementation recurses until the stack overflows; `vi.restoreAllMocks()` also stops resetting a `vi.fn`'s implementation, so the injection leaked into the six tests that follow. Route both injections through the module mocks' own `vi.fn`s, fall through to `vi.importActual`, and reinstate the real functions explicitly. 13/13 under Vitest 2.1.9 and 4.1.11. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

The Vitest 4 bump in this branch drops `poolOptions`, so the `poolOptions.forks.execArgv` fix from colbymchenry#1883 is silently ignored here and a pool worker still compiles tree-sitter grammars on the turboshaft tier — the `Fatal process out of memory: Zone` worker death from colbymchenry#1779 came back in a full run on this branch. Vitest 4 reads `test.execArgv`; the engine project inherits it through `extends`. A probe test confirms the fork's `process.execArgv` now carries `--liftoff-only`, and the 655-test extraction file passes where it died before. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

danusha2345 mentioned this pull request Sep 16, 2026

test(vitest): run pool workers with the WASM runtime flags (#1779) #1883

Closed

danusha2345 pushed a commit to danusha2345/codegraph that referenced this pull request Sep 27, 2026

Merge colbymchenry#1667 (fork/fix/ui-review-findings), rebased onto 1…

d81fd7d

…cbac0a; content already in integration

danusha2345 pushed a commit to danusha2345/codegraph that referenced this pull request Sep 27, 2026

Merge integration/local-2026-09-27 (upstream 1cbac0a, colbymchenry#1875/

affbb89

colbymchenry#1937/colbymchenry#1667 rebased)

danusha2345 and others added 8 commits October 3, 2026 11:20

fix(ui): keep saved trails inside the project

eef3014

test: stabilize UI checks on Vitest 4

7d457c3

fix(ui): harden saved trail reads and metadata

3830b7c

test(ui): bound FIFO regression check with a child timeout

a7a7e78

danusha2345 force-pushed the fix/ui-review-findings branch from e35c791 to 111bfa4 Compare October 3, 2026 09:04

Conversation

danusha2345 commented Sep 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Reproductions fixed

Verification

Uh oh!

danusha2345 commented Sep 8, 2026

Uh oh!

danusha2345 commented Sep 8, 2026

Uh oh!

danusha2345 commented Sep 8, 2026

Uh oh!

danusha2345 commented Sep 8, 2026

Uh oh!

danusha2345 commented Sep 8, 2026

Uh oh!

danusha2345 commented Sep 8, 2026

Uh oh!

danusha2345 commented Sep 8, 2026

Uh oh!

danusha2345 commented Sep 8, 2026

Uh oh!

danusha2345 commented Sep 8, 2026

Uh oh!

danusha2345 commented Sep 8, 2026

Uh oh!

danusha2345 commented Sep 8, 2026

Uh oh!

danusha2345 commented Sep 8, 2026

Uh oh!

danusha2345 commented Sep 8, 2026

Uh oh!

danusha2345 commented Sep 8, 2026

Uh oh!

danusha2345 commented Sep 9, 2026

Uh oh!

danusha2345 commented Sep 16, 2026

Uh oh!

danusha2345 commented Sep 16, 2026

Uh oh!

danusha2345 commented Oct 1, 2026

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

danusha2345 commented Sep 1, 2026 •

edited

Loading