Repository navigation
go: Go format and lint consumer tests pass on all five native cells #5152
Description
Part of #1265. Needs #5137 and #5144.
Problem
The Go format and lint consumer tests only run on Linux x86_64. #1265 requires the Go tools to pass on all five native cells.
Where
- quality/tests/BUILD.bazel:1447-1470 (
go_consumer_test,target_compatible_with = _LINUX_X86_64). - quality/tests/BUILD.bazel:11-14 (
_LINUX_X86_64). - tools/coverage/cells.txt (the five cells).
Fix
- Once the modfmt and lint cases from the sibling issues land, widen
go_consumer_testto every cell and run it on each native runner. - Fix or file any per-host failure (Windows paths, arm64 tool builds).
Done when
bazel test //quality/tests:go_consumer_testpasses on Linux x86_64, Linux arm64, macOS arm64, Windows x86_64 and Windows arm64 at one SHA, with job links posted here.- The
target_compatible_withrestriction is removed.
Activity
TrapsterDK commentedon Nov 3, 2026
Blocked: needs #5137 and #5144 (both still open) before go_consumer_test is widened. Done-when also needs job links from all five native runners, which cannot be produced locally.
TrapsterDK commentedon Nov 4, 2026
Plan (agent): size S
Decision: the nested
go_consumer_testis gone (deleted in 9ccfe7d / #5307 under #5291), so prove the Go tools per cell with the in-repo runner-matrix and real-result cases that already replace it. gofumpt (matrix_go_format_*) and govet/errcheck/gotype/staticcheck (real_result_*, #5143/#5668) are already flagged for all five cells. The only gap is modfmt: both of its matrix cases are Linux x86_64 only, butGO_MODFMT_FORMAT_SUPPORTclaims all five platforms asqualified. This no longer needs #5137 (closed) or #5144.Rejected: bring back a nested
go_consumer_testand widen it (goes against #5291, no nested consumer tests); close as obsolete (leaves the modfmt claim on 4 cells with no evidence); narrow the modfmt claim to linux_x86_64 (dx_modfmt is pure Go, so there is no reason to drop cells).Cause: quality/testdata/runner_matrix_file_family.bzl:1137-1158 (
matrix_go_module_format_pass/_failhave no platform flags) against quality/support.bzl:670-682 (modfmt claims five platforms). quality/support_tests.bzl:323-327 checks only the four analyzers, so nothing caught the gap.Change:
matrix_go_module_format_passandmatrix_go_module_format_fail: addlinux_arm64,macos_arm64,windows_x86_64,windows_arm64= True, the same as the gofumpt cases in runner_matrix_native.bzl:292-340.gofumptandmodfmtto the tool list, so a Go claim with no case per cell fails the test.Tests: tools/dx test //quality:support_unit //quality/testdata:matrix_go_module_format_pass //quality/testdata:matrix_go_module_format_fail //quality/testdata:matrix_go_format_pass //quality/testdata:matrix_go_format_fail //quality/testdata:matrix_go_format_syntax_fail //quality/testdata:real_result_govet_clean //quality/testdata:real_result_govet_dirty //quality/testdata:real_result_errcheck_clean //quality/testdata:real_result_errcheck_dirty //quality/testdata:real_result_gotype_clean //quality/testdata:real_result_gotype_dirty //quality/testdata:real_result_staticcheck_clean //quality/testdata:real_result_staticcheck_dirty //go/tools/modfmt:modfmt_test
Host: windows and macos. Use an evidence.yml run on linux_arm64, macos_arm64, windows_x86_64 and windows_arm64 with the scope above. If a Windows cell fails before the Go targets build because of build: //cli/cli:dx fails on Windows again with '.rmeta is not writeable' in the exec config (#3312 regression) #4850 (rmeta not writeable), merge the code change anyway, link the run here, and keep this issue open until build: //cli/cli:dx fails on Windows again with '.rmeta is not writeable' in the exec config (#3312 regression) #4850 is fixed. Fix any real Go-specific failure (path separators, CRLF) in go/tools/modfmt/main.go and add a modfmt_test case for it.
Upstream: none.
Done when: the Go tool claims check passes, all five cells are green for the scope above at one SHA, and the links are posted here. The
target_compatible_withitem in the body is obsolete because that target no longer exists.Follow-up (not in this PR): go: consumer test runs govet, errcheck and staticcheck on a multi-package module #5144 rests on the same deleted nested test. It should be rescoped to real-result cases on a multi-package fixture with a go_deps module and a generated source.
quality: prove modfmt on all five cells (part of #5152) (#6636)
TrapsterDK commentedon Nov 5, 2026
origin/main already has the code change (b94ad34, #6636: modfmt matrix cases on all five cells, support test covers gofumpt and modfmt). Remaining is native evidence: an evidence.yml run on all five platforms at one SHA, which cannot be produced locally. Leaving open.
Assisted by Claude.
TrapsterDK commentedon Nov 5, 2026
Evidence rounds on origin/main (code change already landed in b94ad34), scope = Plan list, no cell green yet:
The scope pulls unrelated toolchains (ocaml, python, jdk) on Windows/macOS, so narrowing scope to the Go targets with fewer deps would help. No PR: no code diff. Stays blocked.
Assisted by Claude.
TrapsterDK commentedon Nov 5, 2026
Unblocked: #5137 and #5144 closed.
12 remaining items
php: update macOS arm64 PHP 8.5.12 checksum (#9283)
MODULE.bazel: drop gotreesitter from gazelle_go_deps use_repo, it is …
TrapsterDK commentedon Nov 7, 2026
PR #9282 merged (MODULE.bazel use_repo fix); the module now loads. Evidence run 42374110494 (macos_arm64, windows_arm64) still fails, so this stays open. No nested consumer test was added.
Blocked on: the two causes above.
🤖 Generated with Claude Code
https://claude.ai/code/session_01Xut2hzpb4qWNZKpDCrwUz7
github-actions commentedon Nov 7, 2026
This issue has the
blockedlabel but noBlocked on:line in its body. Add one naming the blocking issue (#N), a pull request URL, or the owner action. See CONTRIBUTING.md.llvm: stage0 objects share the runtime features reset (hermetic-llvm#…
TrapsterDK commentedon Nov 7, 2026
macos_arm64 failure in 42374110494 was the php macos pin (older SHA); main's pin 1ece0162 matches the live archive. Re-dispatched narrow Go scope on macos_arm64 + windows_arm64 at 46d4b4a: run 42378289008. Known windows_arm64 causes from 42374110494: aspect implicit deps (selene, deadnix, topiary, weeder) not valid executables on windows_arm64; rules_py cryptography sdist_build lacks python_3_13_aarch64_pc_windows_msvc; matrix_go_format_syntax_fail golden mismatch.
Blocked on: run 42378289008 result (windows_arm64 aspect deps and sdist_build toolchain if it still fails).
🤖 Generated with Claude Code
https://claude.ai/code/session_01Xut2hzpb4qWNZKpDCrwUz7
go: windows_arm64 aspect tool stubs, staticcheck reports U1000, scrat…
quality: scratch normalization matches mixed Windows separators (part…
TrapsterDK commentedon Nov 7, 2026
Blocked on: macos_arm64 (registry cache size, prost -lm link) and windows_arm64 (MSYS2 libyaml loading errors). Run 42381013335, PR #9332.
🤖 Generated with Claude Code
https://claude.ai/code/session_01Xut2hzpb4qWNZKpDCrwUz7
go: Go evidence on windows_arm64 and macos_arm64 (part of #5152) (#9332)
TrapsterDK commentedon Nov 7, 2026
PR #9341 patches rules_py so it can see python_3_13_aarch64_pc_windows_msvc. That was the real windows_arm64 loading error (cryptography sdist_build); the MSYS2 libyaml output was noise. Upstream: aspect-build/rules_py#1676. Evidence run 42382327062 (macos_arm64, windows_arm64, 14 Go targets): https://github.com/devhusdk/rules_dx/actions/runs/42382327062 . macOS -lm: #9339. cryptography source build on windows_arm64: #9340.
Blocked on: run 42382327062 result.
Assisted by Claude.
rules_py: make the windows_arm64 host interpreter visible to sdist bu…
TrapsterDK commentedon Nov 7, 2026
Run 42382327062 at 8f5e2a4 (PR #9341, merged as 74d3708), with the 14 Go targets:
Both jobs still show as failed because the save-registry-cache size check fails (cache is 149 MB on macOS and 213 MB on Windows, limit 50 MB). A separate PR fixes that, and the evidence step itself passed.
Remaining for Done when: all five cells green at one SHA (Linux x86_64, Linux arm64 and Windows x86_64 need to run alongside these two at the same SHA).
Assisted by Claude.
TrapsterDK commentedon Nov 7, 2026
Evidence run 42384304291 on main at 681c1fa (14 Go targets, all five cells): NOT all green, issue stays open.
no such package '@@+php+dx_php_linux_amd64//': Error downloading https://dl.static-php.dev/v3/php-bin/gnu-bulk/php-8.5.12-cli-linux-x86_64.tar.gz(Checksum was 3c932354...). No tests ran.+php+dx_php_linux_arm64php-8.5.12-cli-linux-aarch64.tar.gz checksum mismatch (was ed2ae18b...). No tests ran.invalid registered toolchain '@local_config_winsdk//:all'while resolving toolchains for every target (e.g. //quality/testdata:matrix_go_format_syntax_fail). No tests ran.No Go-specific failure; all four failures are pre-test toolchain/pin issues.
🤖 Generated with Claude Code
https://claude.ai/code/session_01Xut2hzpb4qWNZKpDCrwUz7