Respect --tmpdir across annotate (diamond) and predict (miniprot) steps

[nextgenusfs]

Closes #52.

Summary

The --tmpdir CLI option was not honored for all intermediate files. Two distinct places hard-coded /tmp, causing FileNotFoundError on nodes where /tmp is not writable (HPC/slurm, containers) even when the user supplied --tmpdir.

This PR threads the user-supplied tmp directory end-to-end through both the annotate diamond searches and the predict miniprot alignment, and makes create_tmpdir robust against accidental deletion of a user-supplied volume.

Root cause

1. annotate → swissprot_blast / merops_blast → diamond_blast

diamond_blast defaulted to tmpdir="/tmp" and its two wrappers did not accept a tmpdir kwarg at all. annotate.py never forwarded args.tmpdir, so diamond intermediates were always written to /tmp/diamond_<uuid> — exactly the traceback in #52.

2. predict → align_proteins

align_proteins defaulted to tmpdir="/tmp" and wrote mini_tmp = os.path.join(tmpdir, f"{uuid.uuid1()}.miniprot.out"). predict.py never passed a tmpdir, so miniprot output hit /tmp regardless of --tmpdir. Same class of bug, different tool.

3. create_tmpdir semantics

When the user passed any --tmpdir other than /tmp, create_tmpdir returned the directory verbatim — meaning the downstream shutil.rmtree(tmp_dir) in predict.py would wipe the user-supplied volume. It also blocked concurrent runs sharing a scratch volume.

Changes

funannotate2/search.py

• diamond_blast(..., tmpdir="/tmp") → diamond_blast(..., tmpdir=None) with fallback to tempfile.gettempdir() (honors $TMPDIR).
• swissprot_blast(..., tmpdir=None) and merops_blast(..., tmpdir=None) — new kwarg, forwarded to diamond_blast.

funannotate2/annotate.py

• swissprot_blast(Proteins, cpus=args.cpus) → swissprot_blast(Proteins, cpus=args.cpus, tmpdir=tmp_dir).
• merops_blast(Proteins, cpus=args.cpus) → merops_blast(Proteins, cpus=args.cpus, tmpdir=tmp_dir).

funannotate2/align.py

• Added import tempfile.
• align_proteins(..., tmpdir="/tmp") → align_proteins(..., tmpdir=None) with fallback to tempfile.gettempdir().
• align_transcripts(..., tmpdir="/tmp") → align_transcripts(..., tmpdir=None) for API parity (kwarg was and remains unused internally — gapmm2 writes directly to output=evidence; docstring updated to explain).

funannotate2/predict.py

• Pass tmpdir=tmp_dir to both align_transcripts(...) and align_proteins(...).

funannotate2/utilities.py

• create_tmpdir now always creates a unique <base>_<uuid> subdirectory inside the supplied volume (previously only for /tmp). Falls back to tempfile.gettempdir() when no volume is supplied (was CWD). This prevents the shutil.rmtree hazard on user-supplied volumes and makes concurrent runs safe on a shared scratch directory.

Behavior change for users

• funannotate2 annotate --tmpdir /scratch/mydir now writes /scratch/mydir/<base>_<uuid>/... — previously it would write directly into /scratch/mydir and could later delete it.
• When --tmpdir is omitted, temporary files now land in tempfile.gettempdir() (which honors $TMPDIR), not a hardcoded /tmp.

Tests

All new tests extend existing test modules — no new test files created.

• tests/unit/test_utilities.py::TestCreateTmpdir — 5 tests covering unique-subdir creation, absolute paths, $TMPDIR fallback, and subdir isolation from the user-supplied volume.
• tests/unit/test_search_comprehensive.py::TestDiamondTmpdirPlumbing — 4 new tests that mock subprocess.Popen and assert diamond's --out argument is under the caller-supplied tmpdir; plus two existing assertions updated for the new tmpdir=None kwarg on swissprot_blast/merops_blast.
• tests/unit/test_predict.py::TestAlignProteinsTmpdir — 2 new tests that mock runSubprocess and assert the stdout path passed to miniprot is inside the caller-supplied tmpdir (explicit case) or inside tempfile.gettempdir() (fallback case).

Results

$ pytest tests/unit/test_utilities.py tests/unit/test_search_comprehensive.py tests/unit/test_predict.py -q
... 41 passed

$ pytest tests/ -q
196 passed, 2 failed, 61 skipped

The 2 failures (test_funannotate_cli::test_version_command, test_help_command) are pre-existing environment-specific failures — subprocess.run("python -m funannotate2 --version") returns 127 in the integration harness on this machine. Verified identical on git stash (before any change in this PR).

Non-goals / out of scope

• abinitio.py's folder="/tmp" kwargs — these wrap work in tempfile.TemporaryDirectory() internally, so $TMPDIR is already honored at runtime. Stylistic cleanup only; left for a separate change.
• evm.py::evm_consensus has a tmpdir="/tmp" default but is not imported or called anywhere else in the package (dead code).
• Changing gapmm2 behavior to actually consume the tmpdir kwarg — gapmm2 writes directly to the caller-supplied output path; a staging tmp is out of scope.

Rollback

All changes are confined to funannotate2/{search,annotate,align,predict,utilities}.py and three existing unit-test modules. All added kwargs default to None, so the public API is backwards compatible.

---

Pull Request opened by Augment Code with guidance from the PR author