Repository navigation
master CI red: cargo fmt, clippy -D warnings, and ratchet budgets fail on 4bb116454 #1907
Description
The quality job is failing on master at 4bb1164 (and later). Four gates fail:
cargo fmt --all -- --check: three files drifted (session_picker.rs, auth_account_commands.rs, remote.rs, plus a couple of formatting-only spots elsewhere).cargo clippy --locked --all-targets --all-features -- -D warnings:jcode-basefails onlarge_enum_variant:PollOutcome::Approved(MuseCredentials)carries a ~240 byte variant that dwarfs the rest of the enum.- test-only
request_bodyhelpers injcode-base/src/jev.rsandjcode-app-core/src/tool/browser_jev.rsare dead code in production builds but only compiled under tests. type_complexityon the testTEST_SLEEP_OVERRIDEstatic insrc/cli/login/device_login.rs.
- Code-size / test-size / panic / swallowed-error ratchet budgets: the 10-19 pushes grew several tracked oversized files and added new ones, so all four budget scripts reject the current tree.
- Build & Test fails provider_matrix on the muse-subscription experimental gate (separate, tracked in CI red on master: provider_matrix muse-subscription profile counted as unconfigured, test panics at the experimental opt-in gate #1902).
The lib test suites themselves are green upstream (2790 passed / 0 failed on ubuntu at this commit), so the breakage is gate-level, not behavior-level.
Fix incoming in #1906: fmt the drifted files, box the MuseCredentials variant, cfg(test) the test-only helpers, and regenerate the four budget baselines with the scripts' own --update mode.
Activity
1jehuang commentedon Oct 20, 2026
Thanks for the concrete gate list. I confirmed that the current code-size, test-size, panic, and swallowed-error budget checks still fail. I am keeping this open for review of #1906: automatically regenerating all four baselines would accept unrelated growth rather than establish that it is safe. Recommended next step: separate the mechanical lint/format fixes from reviewed budget changes or targeted cleanup. The independent Muse matrix failure is being addressed under #1902.
Jcode agent (automated triage), on behalf of @1jehuang
alecuba16 commentedon Oct 21, 2026
Rebased this PR onto master e168c36 and split it per the recommendation above: commit 1 is the mechanical clippy/machete/muse cleanups (no behavior changes), commit 2 is the budget-baseline refresh covering only the newly merged upstream code (kagi #1916, zellij/kitty-classic #1917). Details in the PR description. Everything the ratchets now grow is upstream-merged code, so the mechanical half is independently reviewable and the budget half is a pure 10-21 sync.
alecuba16 commentedon Oct 24, 2026
Re-checked the gates on the current tip be64b54: still red, and the failure set moved since e168c36.
What fails now (all reproduced locally, same list as the fork CI run on that commit):
cargo fmt --check: drift in the compaction floor test in jcode-config-types (from 638bc82).format_todo_completion_confidenceleft by da87e0c.bg action="status"/"output"hint sentence removed by 90ee503, the KV cache notice test expects the pre-2934994d0 render, the remote catalog test misses the sonnet-4-6 entry added by 803d7e4, and the flicker copy badge test expects the pre-1a544fc5f[Z]form.I pushed a third commit on #1906 (b18faf5) that keeps the same split: all mechanical (collapse ifs, hex-escape the nulls, box the closure list, delete the dead fn) plus four test updates aligned with the new formats, and the budget baselines refreshed only to the counts the upstream-merged code produces. fmt, clippy -D warnings, warning budget and all four budget scripts pass locally on that head.
The macOS-only hibernation/ambient-state failures stay in #1897 / #1880, not this PR.
ianalitis commentedon Oct 25, 2026
Re-confirmed on today's
origin/master(60ccebb00, "docs: update weekly stars chart"), independently of the earlier commit in this thread. All four ratchet scripts are red on unmodified master, plus fmt and clippy:cargo fmt --all --checkcargo clippy --all-targets -- -D warningsunused import: HashSetintool/communicate_tests.rs:22)scripts/check_test_size_budget.pyscripts/check_swallowed_error_budget.pyjcode-tui-tool-display/src/lib.rs)scripts/check_code_size_budget.pyscripts/check_panic_budget.pyscripts/check_wildcard_reexport_budget.pyscripts/check_warning_budget.shWorst offenders worth deciding on first:
check_test_size_budget:server/client_lifecycle_tests.rs1939 -> 2305,usage/tests.rs1317 -> 1355,tui/ui_tests/tools.rs1530 -> 1542,tui/ui_tests/prepare.rs1374 -> 1394.check_code_size_budget:server/client_lifecycle.rs3974 -> 4030,tool/todo.rs2853 -> 2884,server/swarm.rs3518 -> 3519.check_panic_budget:jcode-base/src/provider_activity_oauth.rs3 -> 5, plus a new one injcode-provider-openai-runtime/src/openai_stream_runtime.rs.For every flagged path I checked, the file is byte-identical to
origin/master, so none of this is drift from an open PR branch — it accumulated on master. Two caveats on the fix strategy:--updateon the ratchets would launder all of this in one commit. The budgets are the only thing keeping the refactor pressure on, so a bulk rebaseline effectively deletes that gate. Per-file rebaselines are the honest option, and even then only for files whose growth was reviewed.client_lifecycle.rs/client_lifecycle_tests.rs/tool/todo.rs, which is a split refactor, not a formatting pass. That is the part that needs a decision here, not a mechanical--update.If it helps, I can take the fmt + the two genuinely mechanical clippy fixes (the unused import is one) as a small PR that leaves every budget untouched, and leave the ratchets for a maintainer call. Say the word and I'll open it.
— Jcode agent (automated triage), on behalf of @1jehuang
ianalitis commentedon Oct 25, 2026
Follow-up: opened #1963 for the fmt half (three files, formatting only,
cargo fmt --all --checkgoes green). It is a prerequisite for anyone whose PR otherwise fails Format on unrelated drift, including #1962.Leaving clippy
-D warningsand all four ratchets unopened, since those are not mechanical.ianalitis commentedon Oct 25, 2026
Triaged this on a fresh worktree of current master (
4bb16b96c, not4bb116454). The report is still accurate, but the concrete failure set differs from the original list, so here is the current breakdown. Fix PR: #1964.What fails today, and why
cargo fmt --all -- --checkcrates/jcode-config-types/src/lib.rs:1630cargo clippy -- -D warningscollapsible_if(3),needless_borrow(2),is_some()+unwrap()(1), unused imports (2),dead_code(1)format_todo_completion_confidenceispub(super)with no callers, so the macOS warning budget sawfunction ... is never usedcargo test -p jcode-tui --libThe 6 tests are all assertions that lagged deliberate behavior changes, not logic bugs:
remote_events_reload_03-90ee50330replaced thebg action="status"/"output"handoff copy withBACKGROUND_WAIT_GUIDANCE.session_picker_tests-ff31b8bba/59eabc409collapse finished todo cards to status dots, so the fixture's completed todo no longer renders a card.ui_messages- compact cache-miss notices render as a one-line summary.frame_flicker- the copy badge now renders[<alt label>⇧Z]; the assertion should usecopy_badge_alt_label().remote_startup_input_02-claude-sonnet-4-6is in the shipped subscription catalog.Still red, deliberately untouched
The ratchet budgets in the report are still red, and they are pre-existing growth rather than a gate that anybody turned off:
check_code_size_budget.py:ui_viewport.rs1664 -> 1759,cloud_move.rs2839 -> 2840,commands.rs3632 -> 3917check_test_size_budget.py:ui_tests/prepare.rs,ui_tests/tools.rs,commands_tests.rscheck_panic_budget.pyandcheck_swallowed_error_budget.py: a handful of filesRefreshing those baselines is a policy decision, not a mechanical fix: it either blesses the growth or requires splitting
commands.rs. It needs a maintainer call, which is why this is labelledtriage: needs-decision. #1964 does not touch the baselines.Verification of the fix
cargo clippy --locked --all-targets --all-features -- -D warningsexits0,cargo fmt --all -- --checkexits0,scripts/check_warning_budget.shreportscurrent=0 baseline=0, and the 6 TUI tests pass (283 passed, 0 failedin the filtered run).— Jcode agent (automated triage), on behalf of @1jehuang
ianalitis commentedon Oct 25, 2026
Short update on #1964 after its first CI pass, since it changes two things I said above.
Format is green now on the PR head, so the rustfmt half is done there. That also makes #1963 a no-op: its three hunks (
ui_messages.rs,ui_messages/tests.rs,ui_tests/tools.rs) already exist verbatim on master.ui_messages.rs:2863at4bb16b96calready readsfailure.reason.split_whitespace()broken across four lines, andgit show 4bb16b96c:<path> | rustfmt --edition 2024 --checkreports zero diffs for all three files. So #1963 would apply cleanly and change nothing, while its own Format job is red. Suggest closing it in favor of #1964; I have not closed it myself.CI caught a real mistake in my branch, and it is worth recording because the same trap will hit anyone doing lint cleanup here. The clippy findings included an unused
use std::path::Path;insrc/cli/cloud_managed_tests.rs. On macOS that import really is unused: both users sit insidemod shell, which is#[cfg(all(target_os = "linux", target_arch = "x86_64"))]. Removing it broke the Linux build withE0425/E0433: cannot find type Path in this scopeat lines 47 and 387, failing bothQuality Guardrails(check + clippy) andBuild & Test (ubuntu-latest)at the compile step, before any test ran.Fixed in
d738d1707by moving the import inside the gated module. Verified by temporarily widening that module's cfg tounixlocally so its body actually compiles (cargo check --locked --bins --testsclean), then restoring the cfg and re-runningcargo clippy --locked --all-targets --all-features -- -D warningsclean on macOS.The general point: a clippy run on one OS says nothing about code behind
cfg, so an "unused import" finding inside a platform-gated module is only safe to act on if you compile that module. This is also why the earlier master runs hid downstream breakage in both directions: master's ubuntu run failed at the TUI test step, so the cohorts after it never executed, and my first push failed at the compile step, so they still had not. The current head is the first commit where retention readiness, pty, browser liveness, embedding stability, stdin forwarding, provider matrix, e2e, and security preflight will actually run.Still expecting the four ratchets (Quality Guardrails steps 12-15) to stay red until you decide on
--updateversus splittingcommands.rs; nothing else in #1964 touches them.— Jcode agent (automated triage), on behalf of @1jehuang
ianalitis commentedon Oct 25, 2026
Addendum after running the suites locally rather than only reading CI logs. Two more reds exist on master that no CI job reports, and both are pre-existing at
4bb16b96c(verified on a clean worktree of that commit):jcode-provider-gemini-runtime --libbuild_tools_keeps_required_when_the_object_declares_no_propertiesasserts arequiredlist survives an object with noproperties;7e6dec1e6madeprune_dangling_requireddrop it on purpose--lib --bins --no-run), it never runs itjcode-tui --libidle_hibernation_frozen_frames_do_not_repopulate_prepared_cachesasserts the frozen frame saysprivate mem, butprocess_memory::snapshot()is Linux/Windows only, so macOS renderstype to resumerunner.os == 'Linux'onlyBoth are in #1964 now, with the platform and provider reasoning recorded in the code comments. The gemini one is also answered on #1951, including the one-line alternative if the older #655 wording is preferred.
Related, and unchanged: the TUI library suite still cannot run in parallel on master.
cargo test --workspace --lib --binsdeadlocks locally inside thejcode-tuitest binary, many threads parked with no CPU use, until it is killed. That is the ordering/global-state problem the CI workflow comment and #592 already describe, so the serial--test-threads=1command stays the honest signal and the one this branch was verified against.PR #1964 current state: on
16da8b3aa, Quality Guardrails is red only on the four ratchet gates listed above (format, module declarations, clippy, warning budget, dependency boundaries, SDK parity, wildcard reexports and unused deps all pass), and Format, TypeScript SDK, Setup Friction, PowerShell Syntax, Windows Cross-Target and Windows ARM64 are green.— Jcode agent (automated triage), on behalf of @1jehuang
ianalitis commentedon Oct 25, 2026
Result on #1964's current head
c4063c77f(run 40153071531):The ubuntu job reaching the end matters more than it looks. Because master fails at
Run TUI library tests, the cohorts after it had never executed: retention readiness, secret-input pty, browser liveness, embedding numeric stability, stdin forwarding, provider matrix, e2e, warning budget and security preflight. All of those pass on this head, so the four ratchets below are the only thing standing between master and green.Still open for you: whether the ratchet growth is intentional (
--updateand record new baselines) or should be split, chieflysrc/cli/commands.rsat 3632 → 3917 and its test file at 1805 → 2183.— Jcode agent (automated triage), on behalf of @1jehuang
chore(ci): refresh ratchet budgets for post-v0.99.1 growth
1jehuang commentedon Oct 30, 2026
Partial mechanical repairs are published in #2015: account-row collapsible_if and public-base message-separator formatting. Changed-file rustfmt and strict library Clippy passed for the affected package set. No ratchet baseline was regenerated. Six existing todo parameter-description cap failures remain, so this issue stays open for the broader CI/budget decision.
Jcode agent (automated triage), on behalf of @1jehuang
chore(ci): refresh ratchet budgets for post-v0.99.1 growth
fix: collapse account identity dimming lint (refs #1907)
chore(ci): refresh ratchet budgets for post-v0.99.1 growth
chore(ci): refresh ratchet budgets for post-v0.99.1 growth
chore(ci): refresh ratchet budgets for post-v0.99.1 growth