Repository navigation
webviews: bun run check runs every piece and fails at the end - #24438
Merged
Merged
Conversation
A red text-focus guard (AddDeviceView.swift:164, 2026-10-22) stopped the `&&` chain, so fmt, lint and the type check never ran in a local `bun run check`, the web-bundles dispatch or iroh-v2. The chain moves to `check:pieces`; `check` runs scripts/cmux-next/run-check-pieces.py, which runs each piece, prints every red one and exits 1. ci-web and the pre-push lint already ran the pieces one by one and now read them from `check:pieces`. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?harness=claude&session=01a1603a-6462-72f0-a4b0-6937eb58e4c3&host=cosmic-gray
Contributor
|
cmux status (one comment, edited in place; each line is updated by the job that owns it) Review App of this head (app.zip, kept 1 day)
Unclear: CI failed on
Not re-run automatically: Written by red: cmux-next checks (god files, concurrency, crash safety, l10n); no finished base run on e7ab659 to compare; do not merge until green; see lane-rules Gallery before/after for
States
Pick other states with a |
|
Found 7 test failures on Blacksmith runners: Failures
|
![Fix with [code]smith](https://pr-comments-assets.blacksmith.sh/codesmith/fix-with-codesmith-light-3x.png)
Collaborator
Author
|
|
Collaborator
Author
|
|
Collaborator
Author
|
Bead |
Review finding (codex xhigh): a change to run-check-pieces.py on feat-cmux-next after a PR's fast-gates base would not mark that result stale. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?harness=claude&session=01a1603a-6462-72f0-a4b0-6937eb58e4c3&host=cosmic-gray
Collaborator
Author
|
Review (codex xhigh): SHIP at a583efc Round 1 (codex xhigh, at c717727) said BLOCK: scripts/cmux-next/run-check-pieces.py wasn't in scripts/ci/check-paths.txt, so a later runner change would not mark a PR's fast-gates result stale. That's fixed at a583efc, and round 2 found nothing else. |
teamleaderleo
added
the
ready-to-land
Collaborator
Author
|
Orphaned: no activity for 70 h, and no lane holds a lease on it. The orphan sweep could not merge |
Collaborator
Author
|
Overlap coordination for #24438: #25548 also touches |
Collaborator
Author
|
Review (codex gpt-6-astra xhigh, next-review): SHIP at e029ede No blocking findings. |
…eck-runs-every-piece # Conflicts: # webviews/package.json
teamleaderleo
temporarily deployed
to
ci-routing-agent
October 29, 2026 22:24 — with 
GitHub Actions
Inactive
GitHub Actions
Inactive
teamleaderleo
temporarily deployed
to
ci-routing-agent
October 29, 2026 22:24 — with
GitHub Actions
Inactive
GitHub Actions
Inactive
Collaborator
Author
|
Owned by an agent session · session |
Collaborator
Author
|
Review (codex gpt-6-astra high, fast, review-router; drain-other): SHIP at 9511e33 No blocking findings. |
teamleaderleo
added a commit
that referenced
this pull request
Oct 29, 2026
… NAME:pieces Since #24438 webviews' check is run-check-pieces.py over check:pieces. check_pieces() already reads SCRIPT:pieces; light_leaves() now does the same for `bun run NAME`, so a nested runner is never run whole as one (light-looking) python3 command. Tests: a heavy=0 push with the runner-shaped check runs only the allowlisted pieces, and a nested runner resolves to its pieces. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?id=aid-3aa8a14a36831eaec813
Contributor
|
Shipped in:
|
Contributor
|
Shipped in:
|
teamleaderleo
added a commit
that referenced
this pull request
Oct 30, 2026
… CI (#27177) * web-lint: hosts that run no heavy checks leave the type-aware lint to CI The pre-push web lint runs webviews' whole `check` script, whose `vp lint` and `vp check` start tsgolint (about 9 GB and 5 GB, both at once). On shared agent hosts that run no builds, that filled swap. A host that sets CMUX_LOCAL_HEAVY_CHECKS=0 now runs every other piece locally and leaves the heavy ones to CI's web lint job and the pull request's fast gates, saying so and naming cmux-ci test for a pre-push run. Wherever a heavy piece does run outside CI, it runs under nice with GOMAXPROCS, GOMEMLIMIT and Node's heap capped, each only when the caller set none. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?id=aid-7a592ef92c66f080e3cc * web-lint: no install or build on a host that runs no heavy checks Review of 6160990: with CMUX_LOCAL_HEAVY_CHECKS=0 the tui bindings were still installed and built (npm ci, npm run build: tsc) before the check pieces were filtered. Now such a host installs and builds nothing: light pieces run against the node_modules a package already has, and a package with none is left to CI with a named note. Regression test: a cmux-tui/frontends/web change with CMUX_LOCAL_HEAVY_CHECKS=0 spawns no npm, bun install, build, tsc or vp; it fails on 6160990 (npm ci and npm run build ran). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?id=aid-7a592ef92c66f080e3cc * web-lint: cap the bindings build and the --new-only rerun; no base install without heavy checks Review of 5c38839: - the tui bindings build (npm ci, tsc) runs niced and capped outside CI, like the type check; - --new-only reruns a failed piece on the base through the same niced, capped job; - with CMUX_LOCAL_HEAVY_CHECKS=0, --new-only installs nothing in the base tree (no node_modules there: the red stays the change's), and a lint tree shares the checkout's node_modules for packages with local builds too, since nothing is built. Four regression tests; each fails on 5c38839. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?id=aid-7a592ef92c66f080e3cc * web-lint: an allowlist on hosts that run no heavy checks; one capped spawn for everything Five review rounds each found another path that still ran something heavy with CMUX_LOCAL_HEAVY_CHECKS=0, so the denylist is gone. Such a host now runs only LIGHT_COMMANDS: `vp fmt --check`, python3 source checks, and `bun scripts/<name>.ts` source checks. `bun run NAME` is resolved to NAME's pieces and only the allowlisted ones run, so check:effect-cleanup's `vp lint` is skipped. ESLint runs on changed web/ files (syntax only). It does no install, no build and no --new-only base comparison; every other piece, and a package with no node_modules, is named and left to CI or cmux-ci. Every tool this script starts (installs, builds, checks, base reruns) goes through spawn(), which outside CI runs it under nice with GOMAXPROCS, GOMEMLIMIT and Node's heap capped. Tests: with heavy=0, a subprocess spy sees only allowlisted commands across --head, a lint tree (another commit), --new-only, --all and --fix. On a build host every spawn is niced and capped. The allowlist is checked exactly. Each new test fails on f9399db. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?id=aid-7a592ef92c66f080e3cc * web-lint: allowlisted pieces run as argv with no shell; shell syntax is never light Review of 5a4516a: an allowlisted prefix could carry another command (`python3 noop.py ; vp lint` ran the lint with CMUX_LOCAL_HEAVY_CHECKS=0), because accepted strings ran with shell=True. Now a piece holding any shell syntax (; | & $ backticks redirections globs quotes escapes newlines) is never light. An allowlisted piece is parsed with shlex.split, checked as an argv, and run as that argv with shell=False. Tests cover ;, &&, ||, |, &, $(...), backticks, newlines and the reviewer's repro through the full main(), and assert no light piece spawns through a shell; each fails on 5a4516a. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?id=aid-7a592ef92c66f080e3cc * web-lint: nice the shell itself, not the first word of a shell string Review of 5eb2e77: `nice -n 10 CHECK_MODE=1 python3 check.py` ran CHECK_MODE=1 as the program (exit 127), and `nice -n 10 true; python3 x.py` niced only `true`. A shell string now runs as `nice -n 10 /bin/sh -c STRING`, so assignments and every command of a compound string keep their meaning and their priority. Regression test fails on 5eb2e77. The bindings-build test now clears CI and GITHUB_ACTIONS itself, since it asserts outside-CI behaviour (it failed on the CI runner). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?id=aid-7a592ef92c66f080e3cc * web-lint: python3 and bun are light only for a guarded source-check script Review of 3811e05: any shell-free `python3 ...` counted as light, so `python3 -m pip install ..`, `python3 -m compileall` or a bare `python3` ran with CMUX_LOCAL_HEAVY_CHECKS=0. Now the allowlist is `vp fmt --check` exactly, plus `python3 [-I] PATH.py ARGS` or `bun PATH.ts ARGS`, where PATH resolves (relative to the package, no escape) into a scripts/ folder under one of the package's `executed` paths, which the hook already refuses to lint when a push changes them. No -m, -c, other flags, absolute paths or bare interpreter. webviews' real checks all still qualify. On 3811e05, is_light("python3 -m pip install requests") is True; the new test fails there. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?id=aid-7a592ef92c66f080e3cc * web-lint: --dry-run lists exactly the commands a run spawns Review of 91644bb: with CMUX_LOCAL_HEAVY_CHECKS=0 the dry run skipped light_plan(), so it listed packages and ESLint areas a run drops for having no node_modules. A run and a dry run now build their commands with one function, check_jobs(), over the same light_plan(): the dry run checks the node_modules the run would see (in a lint tree only where shares_deps() holds). It prints one `<area>: <command>` line per command. A new test runs both in light mode, on the checked-out head and on another commit, and requires the dry run's list to equal the spawned list. It fails on 91644bb. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?id=aid-7a592ef92c66f080e3cc * web-lint: one plan_steps() for a run and its dry run, setup and fix passes included Review of 7d129c5: `--dry-run --fix` left out the check:fix and ESLint --fix passes a run makes first, and in full mode it summarised builds and installs as one line. plan_steps() now returns everything a run does, in order: the plan (light_plan()'s with heavy checks off), the setup steps (setup_steps(): the builds that are not current and each area's install, from the same deps_command() and build_commands() ensure_deps and ensure_build run), the fix jobs and the check jobs. run_plan() executes it and --dry-run prints it one command per line, so the two can't drift. ensure_build decides "current" with the same build_current() setup_steps() uses. Test: table-driven over {light, full} x {fix, no fix} x {checked-out head, lint tree} (6 cells, --fix never runs in a lint tree). In every cell the dry run's lines equal the commands the run spawns: builds and fix passes match line for line, installs and checks (which run in parallel) match as sets, and the phases come in the run's order. On 7d129c5 the --fix cell fails because its fix lines are missing. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?id=aid-7a592ef92c66f080e3cc * web-lint: a light script is one the hook refuses to lint when a push changes it Review of bead8b5: light_script() accepted a scripts/ folder under the package's src/, test/ or tests/, which executed_changes() treats as source a push may change and still be linted, so `bun src/scripts/evil.ts` in a check script would have run a pushed change with CMUX_LOCAL_HEAVY_CHECKS=0. light_script() now asks executed_changes() itself: only a path the hook refuses to lint when a push changes it can be light. Tests: the src/, test/ and tests/ forms are refused (they fail on bead8b5), and every light script of both packages is one make_plan() marks untrusted. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?id=aid-7a592ef92c66f080e3cc * web-lint: a dry run plans the tree the run happens in A dry run of a commit that runs in a fresh lint tree read the checkout's build stamps and node_modules, so it listed only the check while the run built the bindings and installed the frontend first. setup_steps() and plan_steps() now take a TreeState: the checkout as it is, or the lint tree tree_of() makes (no build output, node_modules only where it links the checkout's). Test: TuiBindingsTests.test_a_dry_run_for_a_lint_tree_lists_the_builds_and_installs_that_tree_needs fails on c146f28 (dry run lists only `sh check.sh`) and passes here; tests/test_web_lint_prepush.py: 48 tests OK. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?id=aid-a7306ef145c005ad24ab * web lint prepush: a light host resolves a nested check runner through NAME:pieces Since #24438 webviews' check is run-check-pieces.py over check:pieces. check_pieces() already reads SCRIPT:pieces; light_leaves() now does the same for `bun run NAME`, so a nested runner is never run whole as one (light-looking) python3 command. Tests: a heavy=0 push with the runner-shaped check runs only the allowlisted pieces, and a nested runner resolves to its pieces. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?id=aid-3aa8a14a36831eaec813 * web lint prepush: CI always runs every piece; accept a bun on a relative PATH entry Review of 014ded9: CMUX_LOCAL_HEAVY_CHECKS=0 inherited by a CI runner made CI run only the light allowlist. CI (CI or GITHUB_ACTIONS=true) now wins over the host setting. ensure_deps() refused any bun that shutil.which() found through a relative PATH entry (node_modules/.bin/bun) as "bun is not on PATH"; it now refuses only when which() finds none. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?id=aid-3aa8a14a36831eaec813 * web lint prepush: a stale build with no build tool always stops the run Review of 9fee512: run_plan() ran prepare_builds() only when setup_steps() listed a spawn. With neither npm nor bun, a stale bindings build spawns nothing, and a frontend with node_modules and no bun.lock installs nothing, so the frontend check ran against the old bindings. With heavy checks on, run_plan() now always goes through prepare_builds() (ensure_build's failure surfaces), and setup_steps()/plan_steps() list that refusal and no checks after it, so --dry-run matches the run. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?id=aid-3aa8a14a36831eaec813 * tests: light-host web lint tests clear CI and GITHUB_ACTIONS CI now always runs every piece (CI or GITHUB_ACTIONS=true wins over CMUX_LOCAL_HEAVY_CHECKS=0), so the bindings tests that mean "a host with no heavy checks, outside CI" inherited the runner's GITHUB_ACTIONS=true and ran the full install and build on CI. They now pass LIGHT_HOST, which clears both. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Session: https://cmux.com/deeplink/agent?id=aid-3aa8a14a36831eaec813 --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
This branch was previously deployed
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
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.
bun run checkin webviews was one&&chain. Today a red text-focus guard (AddDeviceView.swift:164, from 1e6efa8) stops it, so fmt, lint and the type check never run. That affects a localbun run check(which CLAUDE.md tells every agent to run before pushing), the web-bundles dispatch and iroh-v2. The AddDeviceView fix itself is a separate fix-forward owned by land-and-ci; this PR makes sure no single guard can hide the rest again.check:pieces.checkrunsscripts/cmux-next/run-check-pieces.py, which runs every piece withnode_modules/.binon PATH, prints each one's output under a==> pieceheader, lists every red piece at the end and exits 1 if any failed.webviews-checkstep (the non-warn branch) and the pre-push lint (web_lint_prepush.check_pieces) already ran the pieces one by one. They now read them fromcheck:pieces, falling back tocheckfor packages without one (cmux-tui/frontends/web).checknow readcheck:piecesthe same way:test_every_part_of_bun_run_check_runs_in_the_warn_branchandtest_format_never_hides_lint_or_type_errors.Changelog
none
Testing
CheckPiecesTestsin tests/test_web_lint_prepush.py.checkruns the runner, andcheck:piecesstill has text-focus, fmt, lint and the type check.check:pieces, withcheckas the fallback.Observed / Not run
tests/test_web_lint_prepush.py32 OK.test_cmux_next_fast_gatesOK.test_cmux_next_checks_workflow.py: the check-pieces test passes; 14 PushPreflight and SwiftCanary tests fail the same way on base.actionlint ci-web.ymlclean.prepush-gate.py: no new violations.bun run checkin webviews on this Mac; CI's react-apps-check runs it.Risk: normal (CI and lint tooling; no app code).
Agent session: cosmic-gray
🤖 Generated with Claude Code
Need help on this PR? Tag
@codesmith-botwith what you need.Summary by cubic
bun run checkin webviews no longer stops at the first red guard; it runs every piece and fails at the end, listing all red pieces.Refactors
&&chain moves tocheck:pieces;checkrunsrun-check-pieces.py.webviews-checkstep and the pre-push lint read their piece lists fromcheck:pieces, falling back tocheck.Written for commit 9511e33. Summary will update on new commits.