meta: reconcile residual #942/#974 follow-ups after #974 close-out

[zackees]

Context

Follow-up to the #974 close-out. #974 was closed after verifying the explicit round-2 sub-issues (#966/#986, #970, #953, #954, #960, #964), but a broader audit found remaining tracker debt around the original #942/#974 performance/CI thread.

This issue exists so those leftovers are visible and actionable instead of being buried in #974/#990 close-out comments.

Investigation sub-issues

Work these one by one. Each item should end with either the referenced issue closed with concrete evidence, or a narrower follow-up issue filed and linked here.

• perf(build): framework libs (~150s) recompiled per project; core cache never hydrates #952 - Verify the fw-libs/core-cache tracker against fix(build): ESP32 sketch+core compiles hit zccache cross-project (#966) #985/fix(build): route framework-lib compiles through embedded zccache — fw-libs cross-project hit (#986) #987/perf(build): fw-lib (framework library) object compiles still miss zccache cross-project #986. Confirm whether warm fresh-project fw-libs now meets the <=5s target from the meta: profiling-driven build performance burndown (Docker Linux, NightDriverStrip, cold vs hot cache) #942 harness and decide whether the remaining core-cache-hydrate concern is obsolete, still valid, or needs a narrower follow-up before perf(build): framework libs (~150s) recompiled per project; core cache never hydrates #952 is closed.
• perf(build): first rebuild after a cold build recompiles once (signature drift between cold and steady state) #957 - Verify the first-rebuild-after-cold tracker against fix(build): don't treat depfile's own mtime as stale → no more first-rebuild recompile (#957) #971. Confirm the root cause, before/after numbers, and whether the first no-change rebuild after cold is now <=10s; then close perf(build): first rebuild after a cold build recompiles once (signature drift between cold and steady state) #957 or update it with the still-failing case.
• Dylint CI gate is red on main (driver build_library failure) — lints are not enforcing #939 - Verify the old Dylint red-gate tracker against ci(dylint): Dylint job is red on main — alias-loop workaround fails, so NO custom lints are enforced #964/ci(dylint): wire dylint-link for all lint crates so libs get @toolchain names (#964) #989/fix(dylint): clean up violations after dylint-link rollout #991 and current main. Confirm the Dylint workflow is green and actually enforcing custom lints, then close Dylint CI gate is red on main (driver build_library failure) — lints are not enforcing #939 as superseded/resolved or split out the remaining branch-protection/required-status-check work.
• PR fix(dylint): clean up violations after dylint-link rollout #991 Windows check cancellation - Investigate the merged PR's Check (windows-latest) cancellation after ~30m19s: https://github.com/FastLED/fbuild/actions/runs/28844571221/job/85553518607. Determine whether this is the known full-workspace Windows timeout or a regression from fix(dylint): clean up violations after dylint-link rollout #991, then either file/attach a dedicated issue or rerun/fix until the Windows check is green.

Findings

#952 - fw-libs/core-cache tracker is still open

#952 was part of #942 and says warm fresh-project fw-libs should drop from ~150s to <=5s and core-cache-hydrate should hit. Later comments say #966/#986 should deliver the real fw-libs win, and #974's close-out says #966/#986 are resolved, including fw-libs routed through embedded zccache at ~152s -> ~2.2s.

Remaining action: verify whether #952's acceptance criteria are now satisfied by #985/#987/#986 and close it with the measurement evidence, or split any still-real core-cache-hydrate concern into a narrower follow-up before closing #952.

#957 - first-rebuild-after-cold tracker is still open

#942's final report says #957 was fixed by PR #971 and that first-rebuild-after-cold is now a true no-op, but #957 itself is still open.

Remaining action: verify the #971 evidence against #957's acceptance criteria and close #957 if proven, or update it with the missing work if the fix was incomplete.

#939 - older Dylint red-gate issue is still open

#939 reports the Dylint gate red on main and custom lints not enforcing. #964/#989/#991 appear to supersede it: PR #991 shows Dylint passing in 19m12s and #964 is closed. However #939 also asks whether Dylint should be a required status check, which was not explicitly closed out.

Remaining action: verify Dylint is green on current main, close #939 as superseded/resolved by #964/#989/#991 if appropriate, and either document that Dylint is already required or file a separate branch-protection follow-up if it is not.

PR #991 - Windows check still failed/cancelled after merge

PR #991 merged with Dylint green, but Check (windows-latest) on the merged PR check rollup failed/cancelled after ~30m19s: https://github.com/FastLED/fbuild/actions/runs/28844571221/job/85553518607

#990 noted this as a CI note and relied on local Windows/Docker validation for touched crates. That may be acceptable for #964's Dylint acceptance criterion, but it is still unresolved CI debt and should be tracked explicitly.

Remaining action: decide whether this is the known full-workspace Windows timeout or a regression from #991, then either file/attach it to the existing Windows CI timeout tracker or fix/rerun until the Windows check is green.

Proposal

Do a tracker-hygiene pass over the residual #942/#974 items:

• Close issues that are genuinely resolved, with concrete links to PRs, measurements, and check runs.
• Split still-real work into smaller issues instead of leaving broad stale trackers open.
• Track the PR fix(dylint): clean up violations after dylint-link rollout #991 Windows check timeout/failure somewhere explicit.
• Leave meta: fbuild build-performance burndown — round 2 (residuals from #942) #974 closed, but treat this issue as the correction rollup for the broader leftover audit.

Acceptance criteria

• perf(build): framework libs (~150s) recompiled per project; core cache never hydrates #952 is either closed with current fw-libs/core-cache evidence or updated/split so any remaining work is specific and current.
• perf(build): first rebuild after a cold build recompiles once (signature drift between cold and steady state) #957 is either closed with fix(build): don't treat depfile's own mtime as stale → no more first-rebuild recompile (#957) #971 verification evidence or updated with the still-failing first-rebuild-after-cold case.
• Dylint CI gate is red on main (driver build_library failure) — lints are not enforcing #939 is either closed as superseded/resolved by ci(dylint): Dylint job is red on main — alias-loop workaround fails, so NO custom lints are enforced #964/ci(dylint): wire dylint-link for all lint crates so libs get @toolchain names (#964) #989/fix(dylint): clean up violations after dylint-link rollout #991 or narrowed to the remaining branch-protection/status-check work.
• The PR fix(dylint): clean up violations after dylint-link rollout #991 Check (windows-latest) cancellation is tracked in an issue or resolved by a green rerun/fix.
• meta: fbuild build-performance burndown — round 2 (residuals from #942) #974 has a correction comment pointing here.

Related issues and PRs

• Parent trackers: meta: profiling-driven build performance burndown (Docker Linux, NightDriverStrip, cold vs hot cache) #942, meta: fbuild build-performance burndown — round 2 (residuals from #942) #974, Finish #974 tail: dylint violation cleanup (#964) + cold-pipeline overlap (#953) #990
• Residual open issues: perf(build): framework libs (~150s) recompiled per project; core cache never hydrates #952, perf(build): first rebuild after a cold build recompiles once (signature drift between cold and steady state) #957, Dylint CI gate is red on main (driver build_library failure) — lints are not enforcing #939
• Relevant PRs/checks: fix(build): don't treat depfile's own mtime as stale → no more first-rebuild recompile (#957) #971, fix(build): ESP32 sketch+core compiles hit zccache cross-project (#966) #985, fix(build): route framework-lib compiles through embedded zccache — fw-libs cross-project hit (#986) #987, ci(dylint): wire dylint-link for all lint crates so libs get @toolchain names (#964) #989, fix(dylint): clean up violations after dylint-link rollout #991

[zackees]

Closing this reconciliation issue as complete.

Resolved/split items:

• perf(build): framework libs (~150s) recompiled per project; core cache never hydrates #952 is closed. The warm fresh-project fw-libs target is resolved by fix(build): route framework-lib compiles through embedded zccache — fw-libs cross-project hit (#986) #987/perf(build): fw-lib (framework library) object compiles still miss zccache cross-project #986, with the meta: profiling-driven build performance burndown (Docker Linux, NightDriverStrip, cold vs hot cache) #942 harness showing warm fw-libs 152s -> 2.2s. The remaining core-cache-hydrate question is split into perf(build): decide fate of core-cache-hydrate after fw-libs zccache fix #995.
• perf(build): first rebuild after a cold build recompiles once (signature drift between cold and steady state) #957 is closed. The first-rebuild-after-cold false-stale bug is resolved by fix(build): don't treat depfile's own mtime as stale → no more first-rebuild recompile (#957) #971, with depfile-mtime regression coverage. The separate build_unflags asymmetry was tracked in fix(build): rebuild_signature omits build_unflags stripping → persistent recompiles for build_unflags projects #970 and is already closed.
• Dylint CI gate is red on main (driver build_library failure) — lints are not enforcing #939 is closed. The Dylint red-gate repair is resolved by ci(dylint): wire dylint-link for all lint crates so libs get @toolchain names (#964) #989 and the accumulated violations were cleaned up in fix(dylint): clean up violations after dylint-link rollout #991. The remaining branch-protection/status-check policy question is split into ci(dylint): decide and enforce Dylint as a required status check #994.
• The PR fix(dylint): clean up violations after dylint-link rollout #991 Windows cancellation is addressed by merged PR ci: pin latest soldr for Windows check #993. That PR pins Check Windows to zackees/setup-soldr@v0.9.66, explicitly installs soldr 0.8.0, and adds a version assertion so the job fails fast if setup installs an older tool. This keeps the 30-minute job budget rather than extending soldr's wait timeout.

Local validation for #993:

• git diff --check
• uv run --no-project python ci/render_workflows.py --check
• soldr --version -> soldr 0.8.0

Main now includes #993 at merge commit 56990d0. Per the close-out instruction, I did not wait for the new remote workflow run after local validation.