Repository navigation
fix: results index skips campaign task records; batch wait loop keeps the fit's exit status #1749
Description
Overview
These are two defects the independent review of autolens_assistant#174 (hpc-campaign p5a) traced back to PyAutoFit:
- Phantom results:
autofit.results indexcounts the stage task records underoutput/.campaign_tasks/as results. - Clean finish recorded as a failure: the generated Slurm batch script records a clean finish after a pre-timeout checkpoint as exit 138.
Heart RED at the door (2026-10-31). Reasons:
- release validation FAILED (stage integrate)
- PyAutoHeart/workflows/Release Integrate: failure
- Release readiness: release validation FAILED (stage integrate)
- autogalaxy: red
- Release validation: NOT release_ready — v2026.10.31.1.dev85201 profile=release
- validation_report/stages/integrate: fail
- autolens_profiling: red
- Worktree drift: 0 orphan / 0 missing / 1 dirty
Human override for Heart RED (development only) extended live by the human on 2026-10-31 to this task ("yes extend the override, go ahead"). Merging needs green CI and a separate /prm; no release.
Plan
- Skip the hidden
.campaign_tasksdirectory when the results index walks an output tree. - Make the generated batch script keep only the fit's own exit status after a forwarded signal.
- Add a test for each. The wait-loop test drives a real USR1 against a generated script.
- Tier: judge. Merge mode: human /prm.
Detailed plan
autofit/results/index.py: add.campaign_tasksto_SKIP_DIRSas a named constant, with a comment on where those records come from (the assistant template's stage entry; they are the campaign collector's, read byautofit.campaign.collector.scan_status, not results). Confirm the collector still reads them, since it scans status records separately. Test intest_autofit/results/test_index.py: a tree with a real result plus.campaign_tasks/demo/stage/0.status.jsonindexes one result.autofit/campaign/scheduler/slurm.pysbatch_script: replace theuntil waitloop withwhile true; do status=0; wait "$child" || status=$?; kill -0 "$child" 2>/dev/null || break; done. Test intest_autofit/campaign/test_slurm.py: write the generated script with a stand-in command that traps USR1, sleeps briefly and exits 0; run it with bash, send USR1 to it, and assert exit 0. Also assert exit 3 when the command exits 3. Skip the test if bash is missing.- Run the full PyAutoFit suite. It must not run in parallel with another PyAutoFit suite, because
test_aggregate_csvwrites a fixed/tmp/summary.csv.
Branch: feature/campaign-task-records-wait-loop. Worktree root: ~/Code/PyAutoLabs-wt/campaign-task-records-wait-loop/.
Original prompt
Click to expand
See PyAutoMind active/campaign_task_records_and_wait_loop.md.
🤖 Generated with Claude Code
Activity
Jammy2211 commentedon Oct 31, 2026
Library PR created: #1750. Full suite green (4170 passed); independent review CLEAN. No workspace impact. Awaiting human /prm.
Jammy2211 commentedon Oct 31, 2026
Shipped
Merged in #1750 (
06a414851+2caeaab3e); pending release.autofit.results indexno longer walks into.campaign_tasks/.autofit.campaign).