[FarisHrvat]

Querying a labels element that is annotated by a table, but whose cropped area contains none of the annotated instances, crashes polygon_query and bounding_box_query with IndexError: list index out of range in _region_as_str_if_list_of_len_one, which reads the first element of an empty region list.

The helper now only unwraps a list of length one. A left join with no matching rows returns an empty table with region == [], as the docstring and the storage format spec describe, and the query callers already drop empty tables, so the query keeps the labels and drops the table. The exclusive and right joins use the same helper, so they're covered too. Thanks to @jan-glx for the suggestion.

It's also the random failure of test_attributes_are_copied[labels2d-*] in the pre-release job on #1265. The labels there come from the shared RNG in conftest.py, so whether the small query box contains an annotated instance depends on which tests ran before on the same worker. Added a test that builds that case on purpose.

Also addresses the left join part of #1226.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.43%. Comparing base (ce40138) to head (fa156d8).
⚠️ Report is 7 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1266   +/-   ##
=======================================
  Coverage   91.43%   91.43%           
=======================================
  Files          53       53           
  Lines        8395     8395           
=======================================
  Hits         7676     7676           
  Misses        719      719           

Files with missing lines	Coverage Δ
src/spatialdata/_core/query/relational_query.py	93.34% <100.00%> (ø)

... and 1 file with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[jan-glx]

(sorry for the rebase)

Thanks for tracking this down, and for the test that builds the case on purpose.

I'd fix it one level down, though. The IndexError comes from _region_as_str_if_list_of_len_one, which returns region[0] for an empty list. With

def _region_as_str_if_list_of_len_one(region: list[str]) -> str | list[str]:
    return region[0] if len(region) == 1 else region

a single region is still stored as a string (the released behaviour #1160 restored), and the left join returns a 0-row table with region == [] instead of None:

• That's what the docstring of join_spatialelement_table describes for 'left' (the table "is filtered to only include matching rows"; None is documented for 'left_exclusive').
• It's what the storage format spec requires: "a table without rows cannot name elements in region (it sets region to [])" (annotation rules). TableModel.validate accepts it.
• The query callers (_filter_table_by_elements, SpatialData._filter_tables) already drop empty tables, so your new test passes unchanged.
• It also covers the two other call sites of the helper (exclusive and right joins).

I tried this together with your test on a branch, and the query and operations tests seemed happy with it. Feel free to take it over into this PR if it makes sense to you, or I'm happy to push it here myself.

This is also the bug in #1226 (left join part), and the cause of the flaky test_attributes_are_copied[labels2d-*].

[FarisHrvat]

Thanks, that makes sense, the helper is the better place for it. I've taken your change over on top and kept the test. Query and operations tests pass locally. I also updated the description and linked #1226.

[jan-glx]

Thanks!

[jan-glx]

@LucaMarconato could you have a look? This also fixes the flaky test_attributes_are_copied[labels2d-*] in the pre-release job, so it is on the release list.

[LucaMarconato]

LGTM