fix: Clean up temp dirs when update_features! creates its own tmpdir

[njakobsen]

Problem

update_features! has leaked one temp directory per archive import since 2022 (1ecc01a). Line 25 creates a tracked temp dir:

tmpdir = options.fetch(:tmpdir) { Dir.mktmpdir("ruby_spatial_features") }

but Hash#fetch with a block doesn't write the default back into the hash, so line 28 passed options[:tmpdir] — still nil — down to the importers. Unzip.extract then minted its own untracked temp dir for every KMZ / zipped-shapefile import, and the ensure block only removed the empty tracked dir (a decoy).

Bulk runs amplify this: a nightly update_features! sweep over thousands of records with skip_invalid: true never aborts and leaks thousands of extraction directories per run. This is what got Stolo's nightly update_geometry cron disabled in Nov 2024.

Fix

Pass the local tmpdir to spatial_feature_imports instead of options[:tmpdir], so extraction happens inside the tracked directory the existing ensure already removes. One line.

Tests

The existing cleanup spec only covered the caller-supplied-tmpdir path (where options[:tmpdir] is set), which is why the default path went untested for four years. Added two regression specs for the no-tmpdir path — import success and import failure under skip_invalid: true (unzip happens before the failure, so this is exactly the leak scenario). Both fail against the old code and pass with the fix. Full suite: 248 examples, 0 failures.

Not addressed (known, much smaller drips)

• Download.normalize_file uses blockless Tempfile.new (GC-finalizer cleanup only)
• Shapefile#project_to_4326 leaks the ogr2ogr companion files (.dbf/.shx/.prj/.cpg) — the ensure deletes only the .shp

Both are plain files (not directories), so age-based /tmp sweeps reclaim them.

🤖 Generated with Claude Code