ENH: Drop *.o/*.a between ctest_build and ctest_test

[hjmjohnson]

Adds an automatic post-build, pre-test cleanup step to itk_common.cmake that deletes intermediate *.o and *.a files from CTEST_BINARY_DIRECTORY between ctest_build and ctest_test. Free disk for the test phase on size-constrained CI runners (the GitHub Actions Ubuntu 22.04 runner has ~73 GiB root with ~2 GiB free immediately post-build; tests can hit "No space left on device" before they get a chance to run).

What this does

Just after ctest_build returns and before ctest_test is called, the script removes every *.o and *.a under CTEST_BINARY_DIRECTORY. The linked executables and shared libraries are already complete at this point; the intermediates aren't needed by ctest_test, ctest_memcheck, or ctest_submit.

Two opt-outs:

• dashboard_do_coverage — automatic skip. gcov needs the .gcno note files colocated with the corresponding .o files in the build tree.
• dashboard_keep_objects — new per-client opt-out, e.g. for clients that re-link incrementally between dashboard iterations.

The step is wrapped in a CI fold (ci_section_start / ci_section_end "Free build-tree object files") so it appears next to the build and test sections in the GitHub Actions log.

Motivating CI failure

From PR #6318's Pixi-Cxx (ubuntu-22.04) job:

****** df -h /  -- post build
/dev/root        73G   71G  2.0G  98% /
****** df -h /  -- post cleanup
/dev/root        73G   63G   11G  87% /
...
System.IO.IOException: No space left on device :
  '/home/runner/actions-runner/cached/2.334.0/_diag/Worker_20260521-171155-utc.log'

The GitHub Actions workflow has been working around this by running find build -type f -name "*.o" -delete + find build -type f -name "*.a" -delete in a dedicated "Free disk space after build" step (see .github/workflows/itk.pixi.yml). This PR moves that same logic into the canonical dashboard script so it applies to every consumer of itk_common.cmake (Azure, CircleCI, Jenkins, custom nightly clients) — not just GitHub Actions.

[dzenanz]

LGTM. I have not tried running this. It would be good if someone else took a look.

[greptile-apps]

Greptile Summary

This PR adds an automatic post-build, pre-test step to itk_common.cmake that removes all *.o and *.a files under CTEST_BINARY_DIRECTORY to free disk space on constrained CI runners. Two opt-outs are provided: an automatic skip when dashboard_do_coverage is set (gcov requires .gcno files colocated with .o), and a new dashboard_keep_objects flag for incremental-relinking clients.

• The file(GLOB_RECURSE) + file(REMOVE) pattern is correct CMake idiom; LIST_DIRECTORIES FALSE ensures no directory paths contaminate the list passed to file(REMOVE).
• The cleanup only covers Unix-style intermediates (*.o, *.a); MSVC builds produce .obj and .lib instead, so Windows CI runners facing the same disk-pressure issue would see no benefit.
• ci_section_start / ci_section_end bracket the entire block including the zero-file case, producing an empty log fold on Windows or when the build produced no intermediates.

Confidence Score: 4/5

Safe to merge for Linux/macOS CI; Windows runners won't benefit from the disk cleanup.

The change is well-scoped and the CMake idioms used are correct. The main gap is that MSVC intermediates (.obj/.lib) are not covered, so any Windows CI runner facing the same disk-pressure scenario is left unchanged. The empty CI log fold on zero-file runs is cosmetic. Both are improvement opportunities, not blocking defects.

Only itk_common.cmake changed; the cleanup block at lines 554-568 is the area to focus on.

Important Files Changed

Filename	Overview
itk_common.cmake	Adds a post-build, pre-test cleanup step that removes all *.o and *.a files under CTEST_BINARY_DIRECTORY using file(GLOB_RECURSE) + file(REMOVE). Two opt-outs are provided: automatic skip under dashboard_do_coverage, and a new dashboard_keep_objects flag. Logic and flow are correct for Unix/Linux targets; Windows/MSVC intermediates are not addressed.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[ctest_build] --> B[ci_report_build_diagnostics]
    B --> C{dashboard_do_coverage\nOR dashboard_keep_objects?}
    C -- Yes --> F[ci_section_start: test]
    C -- No --> D[GLOB_RECURSE *.o *.a]
    D --> E2{count > 0?}
    E2 -- Yes --> E3[file REMOVE all matched files]
    E2 -- No --> E4[no-op]
    E3 --> E5[unset variables]
    E4 --> E5
    E5 --> F
    F --> G[ctest_test]
    G --> H{dashboard_do_coverage?}
    H -- Yes --> I[ctest_coverage]
    H -- No --> J{dashboard_do_memcheck?}
    I --> J
    J -- Yes --> K[ctest_memcheck]
    J -- No --> L[ctest_submit]
    K --> L

Loading

_{Reviews (1): Last reviewed commit: "ENH: Drop *.o/*.a between ctest_build an..." | Re-trigger Greptile}

[blowekamp]

Looks reasonable to me. Consider windows files extensions too.

[hjmjohnson]

Pushed 6ba12c3 addressing two greptile P2 findings:

• Added *.obj and *.lib to the cleanup glob so Windows MSVC dashboard runners also free disk.
• Auto-skip when dashboard_continuous is set (incremental relink in the next loop iteration would otherwise be defeated).
• The ci_section_start/ci_section_end pair is now inside the "files found" branch — no empty fold appears when nothing matches (Windows-no-MSVC build or a partially-failed build).

@dzenanz thanks for the LGTM — these two refinements don't change the spirit of the change, just broaden it to the MSVC pipelines and tidy the log output.