Clearer error when a job hits a full disk (ENOSPC) (#1899)

[arimu1]

Problem

Fixes #1899.

When the user's temporary drive (or output directory) fills up while a job is running, the underlying OSError carries errno 28 (ENOSPC, "No space left on device"). In JobBase._execute this fell through to the catch-all branch:

else:
    _logger.exception("Exception while running job: %s", str(e), ...)
processStatus = "permanentFail"

…which dumps a stack trace and gives the user no hint that the actual problem is a full disk (the symptom reported on the CWL forum thread linked from the issue: a tool silently produces empty outputs).

Fix

Add a dedicated ENOSPC branch to the existing OSError handler in cwltool/job.py, right next to the ENOENT ("command not found") case that's already special-cased there. It logs a concise, actionable message instead of a traceback:

[job foo] No space left on device. The temporary directory (/tmp/…) and/or
output directory (/…/out) may be full; free up space, or point --tmpdir-prefix
and --outdir at a location with more capacity.

The full traceback is still available with --debug (the branch passes exc_info=runtimeContext.debug, matching the sibling cases). I also changed the existing magic-number e.errno == 2 to the named errno.ENOSPC's counterpart errno.ENOENT for readability, now that errno is imported.

Tests

tests/test_examples.py::test_disk_full_error_message injects an OSError(errno.ENOSPC, …) at the job's output-handling stage (monkeypatching cwltool.job.bytes2str_in_dicts) during a normal echo run, and asserts:

• the message contains "No space left on device" and the --tmpdir-prefix guidance,
• no raw Traceback (most recent call last) is printed,
• exit code is 1 (permanentFail).

I verified this test fails against the unpatched code (it hits the old generic-exception/traceback path), proving it guards the behavior. black, flake8, and isort are clean on the changed files.

This addresses the "provide a better error message" ask in #1899. Actually probing free space ahead of time is a larger, platform-specific change and isn't attempted here.

---

This change was produced with the assistance of Claude Code (model: Claude Opus). The diff, root-cause analysis, and test were reviewed by a human before submission.

[mr-c]

@arimu1 thanks for using cwltool and contributing your fixes. Let's get the existing PRs resolved before opening more.

Perhaps you can fix the codecov upload issue. Maybe the version needa bumping? If that doesn't work, I would accept a temporary disabling of CodeCov, but then reviews will take longer as I will have to personally check the code coverage locally.

[codecov]

Codecov Report

❌ Patch coverage is 75.00000% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 85.19%. Comparing base (9b603df) to head (4fa981f).

Files with missing lines	Patch %	Lines
cwltool/job.py	75.00%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2293      +/-   ##
==========================================
- Coverage   85.21%   85.19%   -0.02%     
==========================================
  Files          46       46              
  Lines        8621     8624       +3     
  Branches     2020     2021       +1     
==========================================
+ Hits         7346     7347       +1     
- Misses        808      811       +3     
+ Partials      467      466       -1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[arimu1]

The three failing checks look unrelated to this change:

• zizmor failed with a GitHub advisories API 503 Service Unavailable ("fatal: no audit was performed"), i.e. a transient API outage.
• Tox (3, 11, unit) failures are test_mpi.py::test_singularity / test_udocker complaining about an undefined #MPIRequirement reference — the MPI extension-schema registration flake, unrelated to this diff. This PR only touches the ENOSPC branch in job.py plus one self-contained monkeypatch-based test, and main is currently green.
• codecov/patch cascades from the aborted Tox run.

Could a maintainer re-run the failed jobs when convenient? Happy to rebase if that's easier. Thanks!

[mr-c]

@arimu1 Please enable "Allow edits from maintainers." for your PRs, thank you! https://docs.github.com/en/pull-requests/how-tos/work-with-forks/allowing-changes-to-a-pull-request-branch-created-from-a-fork