fix(cron): convert Windows backslash script path to MSYS form for git-bash

[fhucko]

Summary

Fixes .sh/.bash cron jobs failing on Windows with exit 127 ("No such file or directory"). The scheduler passed a Windows backslash path directly to git-bash, which treats \ as an escape and collapses the path so the script is never found.

Root cause

In cron/scheduler.py, _run_job_script() did argv = [_bash, str(path)]. On Windows str(path) is a backslash path (C:\Users\...\x.sh); git-bash collapses it to C:Users... → exit 127. The script never runs. .py cron scripts are unaffected (run via sys.executable).

Fix

Normalize the path to MSYS form (/c/Users/...) when sys.platform == "win32", guarded so POSIX behavior is unchanged.

Test plan

• Reproduced: nightly .sh cron job failed exit 127 on Windows 10
• After patch: same job runs via hermes cron run → success, pushes to GitHub
• ast.parse clean; no change to POSIX path handling
• Regression tests added in tests/cron/test_cron_no_agent.py: test_run_job_script_windows_argv_uses_msys_path captures the argv passed to subprocess.run and asserts a native C:\... script path is converted to /c/... (no backslashes); test_run_job_script_posix_argv_unchanged asserts the POSIX argv is passed through verbatim.

Refs #62514

🤖 Generated with Claude Code

[alt-glitch]

This was generated by AI during triage.

Duplicate of #23405 (earliest-open PR for the Windows-cron git-bash POSIX/MSYS path mechanism at the same _run_job_script site). Related to #62514 (this PR's spec issue, deduped into #23404) and #60892 (sibling fix). The fuller drive-letter conversion here is a minor extra over the same mechanism.

[teknium1]

Thanks for addressing a real native-Windows cron failure. Current main still constructs shell-script argv as [_bash, str(path)] in cron/scheduler.py:2091, so the fix targets the live defect and covers both no-agent and pre-run script callers.

Problems

• No regression test covers the new Windows-only conversion. The existing shell tests at tests/cron/test_cron_no_agent.py:288-309 execute .sh/.bash scripts but do not verify the path passed to bash.

Suggested changes

• Add a focused Windows regression test that captures bash argv and asserts a native C:\\... script path is converted to /c/...; retain an assertion that POSIX argv is unchanged.

Automated hermes-sweeper review.

[teknium1]

Please add regression coverage for this Windows-only branch. Existing shell-script tests execute .sh/.bash files but do not assert the argv passed to bash; capture subprocess.run and verify a native drive path becomes /c/....

[fhucko]

Added the regression coverage as requested. Both new tests are in tests/cron/test_cron_no_agent.py:

• test_run_job_script_windows_argv_uses_msys_path — captures the argv passed to subprocess.run and asserts a native C:... script path is converted to /c/... (no backslashes), regressing the original exit-127 failure.
• test_run_job_script_posix_argv_unchanged — asserts the POSIX argv is passed through verbatim (no /c/ MSYS transformation).

cron/scheduler.py is otherwise unchanged from the fix. Both tests pass against the real code. PTAL.

[monerostar]

Native Win11 verification (monerostar)

Host: Windows 11 build 26200, Python 3.11.15. Bash on PATH is Hermes-bundled Git/MSYS bash 5.3.9 (…\hermes\git\usr\bin\bash.EXE).

Bug class is real

Passing a native backslash path as argv to bash can collapse separators (POSIX-style escape handling), producing the classic missing-file path:

native: C:\Users\Admin\AppData\Local\Temp\…\nightly.sh
# one invocation path:
err: /bin/bash: C:UsersAdminAppDataLocalTemp…nightly.sh: No such file or directory  (rc=127)

That matches the PR description (exit 127 / mangled C:Users…).

This PR’s conversion

C:\Users\Admin\…\nightly.sh  →  /c/Users/Admin/…/nightly.sh

Focused unit tests on this branch:

pytest tests/cron/test_cron_no_agent.py::test_run_job_script_windows_argv_uses_msys_path \
       tests/cron/test_cron_no_agent.py::test_run_job_script_posix_argv_unchanged \
       -o addopts= -v
# 2 passed

Live with Hermes bash + MSYS argv form outside %TEMP%: script runs (LIVE_OK, rc=0).
C:/… forward-slash form also runs on this bash binary.

Notes for maintainers

• Sibling/overlapping approach: fix(cron): use path.as_posix() for .sh scripts on Windows #44350 (path.as_posix() → C:/… rather than /c/…). Both target the same Windows .sh cron argv problem; worth picking one canonical form (or accepting either if bash on PATH is Git Bash).
• Discovery is still shutil.which("bash") — fine when Git Bash is first on PATH.
• On this host the usertemp mount maps %TEMP% → /tmp; exercising scripts under AppData\Local\Temp can behave differently than under e.g. C:\Users\Admin\… depending on how argv is rewritten. Conversion still looks correct for normal HERMES_HOME/scripts/*.sh locations.

Verdict: LGTM for the Windows branch + tests. Happy to defer if maintainers prefer consolidating with #44350’s as_posix style.