Upstream Test Inventory

This inventory ranks the vendored GeoPandas contract slices by expected implementation leverage, optional dependency pressure, and collected pytest weight. Use it to choose which upstream groups to promote next.

Intent

Map the highest-value upstream test groups to:

• actual collected node weight when available

• required or optional dependencies

• expected runtime tier for local smoke versus broader coverage

• the implementation-order epic they most directly unlock

Request Signals

• upstream tests

• inventory

• contract slices

• collect-only

• runtime weight

• dependency groups

• promotion order

Open First

• docs/testing/upstream-inventory.md

• tests/upstream/README.md

• pyproject.toml

• tests/upstream/geopandas/conftest.py

Verify

• uv run python scripts/check_docs.py --check

• uv run python scripts/intake.py "inventory upstream tests by weight dependency group and runtime"

• uv run pytest --collect-only -q tests/upstream/geopandas/tests/test_extension_array.py

Risks

• Collected node counts do not measure wall-clock runtime directly.

• Optional-dependency groups can appear artificially light when collection is skipped.

• Promoting high-weight slices too early can hide narrower kernel milestones.

Ranking

Rank	Slice	Collected nodes	Dependency group	Runtime weight	Target epic
1	tests/test_extension_array.py	504	default	heavy	Phase 6a / o17.6.1, o17.6.4, Phase 7 / o17.7.1
2	tests/test_sindex.py + tools/tests/test_sjoin.py	459	default, pyproj for some cases, GEOS>=3.10 for dwithin	heavy	Phase 3 / o17.3.*, Phase 4 / o17.4.3, Phase 7 / o17.7.2
3	tests/test_geodataframe.py	405	default, pyproj for CRS/sjoin variants	heavy	Phase 6a / o17.6.4, Phase 7 / o17.7.3
4	io/tests/test_file.py + io/tests/test_file_geom_types_drivers.py	409	pyogrio default, fiona optional	heavy	Phase 6b / o17.6.11, o17.6.12, o17.6.13, Phase 7 / o17.7.4
5	tests/test_geom_methods.py	335	default, pyproj and newer GEOS for selected methods	heavy	Phase 4 / o17.4.1, o17.4.2; Phase 5 / o17.5.*
6	tests/test_overlay.py + tools/tests/test_clip.py	219	default, pyproj; some file-backed comparisons need I/O stack	medium-heavy	Phase 5 / o17.5.1 to o17.5.5, Phase 7 / o17.7.3
7	tests/test_array.py	114	default, pyproj for CRS transforms	medium	Phase 2 / o17.2.1 to o17.2.4, Phase 6a / o17.6.1, Phase 7 / o17.7.1
8	io/tests/test_arrow.py + io/tests/test_geoarrow.py	skipped without pyarrow; 0 collected here	pyarrow optional, fsspec/requests/aiohttp for remote cases, pyproj for CRS metadata paths	heavy when enabled	Phase 6b / o17.6.2, o17.6.7, o17.6.8, o17.6.9, Phase 7 / o17.7.4
9	io/tests/test_sql.py	42	sqlalchemy + psycopg[binary] optional, live PostGIS fixture expectations	medium	Phase 6b / SQL fallback adapters, Phase 7 / o17.7.4
10	io/tests/test_infer_schema.py + io/tests/test_pickle.py	27	default	light	Phase 6b fallback and compatibility bridges

Notes

• Collected node counts came from uv run pytest --collect-only -q against the vendored files on this repo checkout.

• test_arrow.py and test_geoarrow.py are currently gated entirely by the missing pyarrow extra, so their weight is known conceptually but not yet collected in this environment.

• test_file.py and test_file_geom_types_drivers.py are disproportionately large because they cross product engines, drivers, geometry mixes, and file suffixes.

• test_extension_array.py is the largest single compatibility slice and is the best denominator for Phase 6a progress because it stresses pandas alignment, dtype behavior, reshaping, casting, and groupby semantics.

• test_sindex.py plus test_sjoin.py form the clearest bridge from Phase 3 indexing kernels into user-visible GeoPandas behavior.

Recommended Promotion Order

1. test_config.py remains the smoke gate for vendored import health.

2. test_array.py should be the first geometry-buffer contract slice promoted once Phase 2 starts because it is core, medium-sized, and default-deps only.

3. test_extension_array.py should be the first major Phase 6a promotion target because it is large but still free of heavyweight I/O extras.

4. test_sindex.py and test_sjoin.py should anchor Phase 3 and 4 promotion.

5. test_overlay.py, test_clip.py, and test_geom_methods.py should stay behind Phase 4 and 5 kernel milestones.

6. Arrow/GeoArrow and SQL slices should remain explicitly optional until the upstream-optional dependency group is installed in CI or local benches.

Runtime Tiers

• light: under 50 collected nodes or no external system dependency

• medium: 50 to 150 collected nodes, mostly default dependencies

• medium-heavy: 150 to 300 nodes or moderate fixture/data pressure

• heavy: 300+ nodes, strong parametrization, or optional/external stack needs

Verification

Use these commands to refresh the inventory when vendored tests or dependency groups change:

uv run pytest --collect-only -q tests/upstream/geopandas/tests/test_extension_array.py
uv run pytest --collect-only -q tests/upstream/geopandas/tests/test_sindex.py tests/upstream/geopandas/tools/tests/test_sjoin.py
uv run pytest --collect-only -q tests/upstream/geopandas/tests/test_overlay.py tests/upstream/geopandas/tools/tests/test_clip.py
uv run pytest --collect-only -q tests/upstream/geopandas/io/tests/test_file.py tests/upstream/geopandas/io/tests/test_file_geom_types_drivers.py