fix: validate Xenium bundle after UNTAR (drop test-profile skip-validation guard)

[an-altosian]

Motivation

Previously validateXeniumBundle ran inside PIPELINE_INITIALISATION and was gated by:

```groovy
if (!workflow.profile.contains('test')) {
validateXeniumBundle(ch_samplesheet)
}
```

The skip-in-test guard existed because:

• The test samplesheet entry is a .tar.gz URL, not a directory. The validation iterates required files like cell_boundaries.csv.gz over that path and always fails in test mode.
• Cloud-storage inputs (s3://, gs://, az://) also had an early-return because file().exists() is unreliable before Nextflow stages the input.

Both reasons describe the same root cause: validation was running BEFORE staging, so it had to handle multiple un-staged input shapes (directory, tarball, remote URL).

Change

Move the file-presence validation to AFTER the existing UNTAR staging step, where the bundle is unconditionally a resolved local directory regardless of input form.

File	Change
subworkflows/local/utils_nfcore_spatialxe_pipeline/main.nf	Delete the validateXeniumBundle(ch_samplesheet) call + its !workflow.profile.contains('test') guard. Delete the def validateXeniumBundle function (it was tightly coupled to the channel + un-staged-path shape).
workflows/spatialxe.nf	Inline file-presence checks into the existing ch_bundle_path .map block — required-file list errors, optional-file list warns. Same logic, runs once per sample after UNTAR (test mode) or directly against the user-provided directory (production).

The s3/gs/az special-case is no longer needed: by the time ch_bundle_path runs its map, Nextflow has staged the input as a local path, so file().exists() works.

Side effect: nf-core/untar stub fix

Running the validation post-staging exposed a latent bug in nf-core/untar's stub.

Real script (when the archive has a single top-level dir): tar -xavf --strip-components 1 — places extracted files inside ${prefix}/.

Stub script (same condition): iterates tar -tf output and touch \${i} — but \${i} includes the top-level dir prefix (e.g. xenium_bundle/cells.csv.gz), so the files land at ./xenium_bundle/... outside ${prefix}/. The captured Nextflow output is only ${prefix}/, so most of the iterated touches are lost.

The previous local patch in this repo papered over this with 4 hardcoded touch ${prefix}/morphology.ome.tif etc. lines — enough for the test fixture's first-stage consumers, but the test_run/ directory was still missing 12 of the 16 files defined by 10x's bundle spec, which is what made the skip-validation guard necessary.

This PR replaces the band-aid with a proper fix: in the single-top-level-dir branch, strip the leading directory from each tar entry so the stubbed file layout matches the real script. The 4 hardcoded touches are removed since the loop now creates them naturally.

```bash

new stub logic for the single-top-level-dir branch

for i in `tar -tf ${archive}`; do
stripped=${i#*/}
[ -z "$stripped" ] && continue
if [[ ! "${i}" =~ /$ ]]; then
mkdir -p ${prefix}/$(dirname "$stripped")
touch ${prefix}/$stripped
else
mkdir -p ${prefix}/$stripped
fi
done
```

modules/nf-core/untar/untar.diff regenerated to reflect the cleaner change (still a local patch on the nf-core module; upstreaming a separate PR to nf-core/modules would let us drop this patch entirely, but that's a separate effort).

Snapshot updates

Five mode-test snapshots regenerated to reflect the now-complete extracted-bundle file listing:

• tests/coordinate_mode.nf.test.snap
• tests/default.nf.test.snap
• tests/image_mode.nf.test.snap
• tests/preview_mode.nf.test.snap
• tests/segfree_mode.nf.test.snap

Each gained ~15 new entries under coordinate/untar/test_run/, image/untar/test_run/, etc. — the files that the UNTAR stub now correctly creates inside the prefix (cell_boundaries., cell_feature_matrix., cells., metrics_summary.csv, morphology_focus/, nucleus_boundaries., transcripts.zarr.zip, analysis.tar.gz, analysis.zarr.zip, analysis_summary.html, aux_outputs.tar.gz, .end-of-run).

These weren't appearing in the previous snapshots because the broken stub never placed them inside the output directory.

Test plan

• nf-test test tests/{default,coordinate_mode,image_mode,preview_mode,segfree_mode}.nf.test --profile=+docker --ci → 5/5 PASS in 448s after snapshot regen.
• nextflow config -profile docker . parses clean.
• nextflow inspect main.nf -profile test,docker — process graph intact.

Related

This is the architectural follow-up @awgymer and I discussed on #158 — moving validation to post-staging so the skip-in-test escape hatch can go away. PR #158 is independent (params-in-modules cleanup) and not blocking this.

[heylf]

I like this. Becomes now cleaner.