You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
ruff-format.test.sh: 4 failures on Windows Git Bash (probe-gate node loader, finding report shape) #6531
On Windows Git Bash, plugins/ruff-format/hooks/ruff-format.test.sh fails 4 cases on main (3569033). The failures reproduced before any change from #6530.
3 FAILs: probe-gate node loader errors
1 FAIL: finding report shape
These failures are not filesystem leaks. Found while verifying #6530.
Repro: env MSYS=winsymlinks:nativestrict bash plugins/ruff-format/hooks/ruff-format.test.sh on Windows 11, Git Bash.
Thanks for the detailed repro. I understand this as: on Windows Git Bash, plugins/ruff-format/hooks/ruff-format.test.sh fails 4 cases on main at 3569033, independent of #6530 — 3 around probe-gate node loader errors and 1 around the finding report shape, reproduced with env MSYS=winsymlinks:nativestrict bash plugins/ruff-format/hooks/ruff-format.test.sh.
The report is clear and self-contained; no missing info from my side. It looks related in kind (Windows Git Bash test failures) to #6135 and #6527, though those cover different test suites, so it doesn't appear to be an exact duplicate. Flagging as a test-only, Windows-specific failure affecting one plugin's suite rather than runtime behavior, which is why I'd treat it as lower priority relative to user-facing bugs.
Held: the four failures need a re-run on Windows Git Bash. The suite has changed since the report (#6677, #7616), and the lane has no Windows host to confirm whether they remain.
Re-run on Windows Git Bash at origin/main 39916bd575ec3a942d40bf01ddd2c049c7852b2c: still 4 failures, PASS=36 FAIL=4, exit 1. A plain run and an MSYS=winsymlinks:nativestrict run gave identical results.
Environment: GNU bash 5.3.15(1)-release (x86_64-pc-cygwin, Git for Windows), node v24.21.0, ruff 0.17.0, python 3.14.8. Detached worktree of a Windows-side clone, no \\wsl$.
FAIL: probe-gate: ruff_format_enabled=false should print nothing and exit 0 (rc=1 out=node:internal/modules/cjs/loader:1568
FAIL: probe-gate: ruff_format_enabled unset should print the missing-ruff notice (rc=1 out=node:internal/modules/cjs/loader:1568
FAIL: probe-gate: ruff_format_enabled=true should print the missing-ruff notice (rc=1 out=node:internal/modules/cjs/loader:1568
FAIL: spawn budget: a batch with no Python costs only the node launcher: want [spawns=1] got []
PASS=36 FAIL=4
The three probe-gate failures are the same as the original report: node's CJS loader reports Cannot find module '<worktree>\plugins\ruff-format\lib\prerequisites.mjs' (MODULE_NOT_FOUND), although the file exists. That points to a POSIX-versus-Windows path form, or to CJS resolution of an ES-module path. Not yet traced in the test.
The fourth failure has a new cause. The original "finding report shape" failure is gone, and a spawn-budget case fails in its place: hook-census: the fire exited 1; stderr: ...\Git\usr\bin\strace.exe: unknown option -- s. Git Bash's strace is the Cygwin tool, which has no -s, so the census counts no spawns. The census assumes Linux strace and has no skip on Git Bash.
Leaving this open: the failures are real on the current suite.
This was generated by AI during an unattended host run over /fleet:reach (melo-desk-001, Windows 11 Git Bash).
Re-run at origin/main 9965f0c07: still PASS=36 FAIL=4, the same four failures as the 2026-10-22 run. Both causes are traced:
Probe-gate (3 cases). A Windows jq.exe writes CRLF, and read -r keeps the CR. A traced run of the block shows ARG[...\lib\prerequisites.mjs^M], so node looks for a file whose name ends in a CR. The node module loader and path forms are not involved.
Spawn budget (1 case). The case gated on command -v strace, which Cygwin's strace satisfies although it has no -s. ruff-format-end-of-turn.test.sh has the same gate and fails its census case too (8 passed, 1 failed).
The fix is committed on branch fix/ruff-format-test-windows-git-bash. The PR opens once the current pre-push hold lifts. With the fix applied on this host:
Suite
Windows Git Bash
Linux
ruff-format.test.sh
PASS=39 FAIL=0, plain and with MSYS=winsymlinks:nativestrict; census case skipped
PASS=40 FAIL=0; census case run
ruff-format-end-of-turn.test.sh
8 passed, 0 failed; census case skipped
9 passed, 0 failed
Six other formatter suites fail on this host from the same two causes, filed as #9889.
Activity
github-actions commentedon Oct 7, 2026
Thanks for the detailed repro. I understand this as: on Windows Git Bash,
plugins/ruff-format/hooks/ruff-format.test.shfails 4 cases on main at 3569033, independent of #6530 — 3 around probe-gate node loader errors and 1 around the finding report shape, reproduced withenv MSYS=winsymlinks:nativestrict bash plugins/ruff-format/hooks/ruff-format.test.sh.The report is clear and self-contained; no missing info from my side. It looks related in kind (Windows Git Bash test failures) to #6135 and #6527, though those cover different test suites, so it doesn't appear to be an exact duplicate. Flagging as a test-only, Windows-specific failure affecting one plugin's suite rather than runtime behavior, which is why I'd treat it as lower priority relative to user-facing bugs.
Automated triage. A maintainer reviews it.
kyle-sexton commentedon Oct 19, 2026
This was generated by AI during triage.
Held: the four failures need a re-run on Windows Git Bash. The suite has changed since the report (#6677, #7616), and the lane has no Windows host to confirm whether they remain.
kyle-sexton commentedon Oct 22, 2026
This was generated by AI during an attended run.
Re-run on Windows Git Bash at origin/main
39916bd575ec3a942d40bf01ddd2c049c7852b2c: still 4 failures,PASS=36 FAIL=4, exit 1. A plain run and anMSYS=winsymlinks:nativestrictrun gave identical results.Environment: GNU bash 5.3.15(1)-release (x86_64-pc-cygwin, Git for Windows), node v24.21.0, ruff 0.17.0, python 3.14.8. Detached worktree of a Windows-side clone, no
\\wsl$.Cannot find module '<worktree>\plugins\ruff-format\lib\prerequisites.mjs'(MODULE_NOT_FOUND), although the file exists. That points to a POSIX-versus-Windows path form, or to CJS resolution of an ES-module path. Not yet traced in the test.hook-census: the fire exited 1; stderr: ...\Git\usr\bin\strace.exe: unknown option -- s. Git Bash'sstraceis the Cygwin tool, which has no-s, so the census counts no spawns. The census assumes Linuxstraceand has no skip on Git Bash.Leaving this open: the failures are real on the current suite.
kyle-sexton commentedon Oct 24, 2026
This was generated by AI during an unattended host run over
/fleet:reach(melo-desk-001, Windows 11 Git Bash).Re-run at origin/main
9965f0c07: stillPASS=36 FAIL=4, the same four failures as the 2026-10-22 run. Both causes are traced:jq.exewrites CRLF, andread -rkeeps the CR. A traced run of the block showsARG[...\lib\prerequisites.mjs^M], so node looks for a file whose name ends in a CR. The node module loader and path forms are not involved.command -v strace, which Cygwin'sstracesatisfies although it has no-s.ruff-format-end-of-turn.test.shhas the same gate and fails its census case too (8 passed, 1 failed).The fix is committed on branch
fix/ruff-format-test-windows-git-bash. The PR opens once the current pre-push hold lifts. With the fix applied on this host:ruff-format.test.shPASS=39 FAIL=0, plain and withMSYS=winsymlinks:nativestrict; census case skippedPASS=40 FAIL=0; census case runruff-format-end-of-turn.test.shSix other formatter suites fail on this host from the same two causes, filed as #9889.