[Arthur031221]

Anyone who defaces many images from one Python process (a loop over a BIDS dataset calling deface_image) leaks four open file descriptors per image, which eventually fails with OSError: Too many open files; separately, every run leaves one temporary .mat file behind even with cleanup on.

generate_tmpfiles discards the first value of tempfile.mkstemp, which is an open descriptor, so the four descriptors are never closed. deface_image also never removes warped_mask_mat: it is not part of the return value, so neither forcecleanup=True nor the CLI's cleanup_files(...) call can reach it.

Reproduction on current master (Linux; FSL is not installed here, so FLIRT.run is stubbed to write a blank image and the second check only shows that the file is not removed; TMPDIR pointed at an empty directory):

fds leaked after 10 generate_tmpfiles calls: 40
temp files left after forcecleanup=True: ['tmpq5146b62.mat']

With this change both numbers are zero.

Changes:

• generate_tmpfiles closes each descriptor right after mkstemp; the returned paths and their order are unchanged.
• deface_image removes warped_mask_mat once the second FLIRT call has finished.

Two tests added in tests/test_utils.py. Both fail on master (descriptor count grows; one .mat file remains) and pass here. Local run on Linux, Python 3.12: 7 passed, 1 skipped (the skip is test_deface_image, which needs FSL, so nothing here ran against a real FSL install). ruff check and ruff format --check are clean at v0.15.12, the version pinned in .pre-commit-config.yaml.

Targets master, where #83 and #85 were merged.

[effigies]

What about instead doing:

with tempfile.TemporaryDirectory(prefix='pydeface', delete=forcecleanup) as tmpdir:
    if verbose:
        print(f'Temporary directory: {tmpdir}')
    template_reg_mat = os.path.join(tmpdir, 'reg.mat')
    ...

There's no need to close file descriptors, because they aren't opened. The generate_tmpfiles helper goes away, and the files are cleaned up based on the forcecleanup flag.