Fix #1308: Remove dead dask.config.set() at module import time

[khushthecoder]

Closes #1308.

Summary

ag3.py:10 executed the following at module import time:

# silence dask performance warnings
dask.config.set(**{"array.slicing.split_native_chunks": False})

This line was added in an older dask era to silence a PerformanceWarning emitted by certain slicing operations. On current dask it is dead code — removing it directly fixes the bug (import malariagen_data no longer mutates global dask configuration).

Why just remove the line

Verified empirically on dask 2025.11.0 (what the project resolves to since dask = "*" in pyproject.toml):

• dask.config.get("array.slicing.split_native_chunks", "DEFAULT") returns "DEFAULT" — the config key is no longer recognised.
• from dask.array.slicing import PerformanceWarning raises ImportError — the warning class has been removed.
• da.take / da.compress / fancy indexing with out-of-order indices emit zero warnings with the config set to True, False, or unset.

So the line suppresses a warning that no longer exists, gated by a config key that no longer exists. Scoping it to context managers (the previous approach in this PR) preserved dead code unnecessarily; deleting it is the cleaner fix.

Diff

1 file changed, 4 deletions.

-import dask
 import pandas as pd
 ...
-# silence dask performance warnings
-dask.config.set(**{"array.slicing.split_native_chunks": False})

Test plan

• python -c "import malariagen_data; import dask; assert dask.config.get('array.slicing.split_native_chunks', 'DEFAULT') == 'DEFAULT'" — passes (import no longer mutates dask config).
• pytest tests/anoph/test_snp_data.py — 193 passed, no PerformanceWarning emitted.
• pytest tests/anoph/test_frq.py tests/anoph/test_cnv_data.py tests/anoph/test_hapclust.py tests/anoph/test_dipclust.py tests/anoph/test_hap_data.py tests/anoph/test_aim_data.py — 133 passed.
• pre-commit run --files malariagen_data/ag3.py — ruff, ruff-format, trailing-whitespace, end-of-file all pass.
• mypy malariagen_data/ag3.py — no issues.
• CI matrix on PR (linting, coverage, tests across Python 3.10 / 3.11 / 3.12 × numpy ==2.0.2 and >=2.0.2,<2.1).

Backwards compatibility

• No public API changed.
• If anyone pins a very old dask version where the warning still fires, they may see it again; the project does not currently pin a minimum dask version in pyproject.toml (dask = "*"), so modern dask is the supported target.

[khushthecoder]

Hi @jonbrenas — gentle follow-up. I've force-pushed a simpler approach to this PR than the
original scoped-context-manager version; quick summary so the change isn't confusing.

New approach: just deletes the 4 lines in ag3.py. No scoped context managers added anywhere.

Why this turned out to be the right shape: on current dask (2025.11.0, what poetry.lock
resolves to), the config key array.slicing.split_native_chunks no longer exists in dask's
schema — grep across the installed dask package returns zero matches, and
from dask.array.slicing import PerformanceWarning raises ImportError. So the original line
in ag3.py was silently setting a dead config key to suppress a warning class that no longer
exists. Wrapping dead code in context managers (the earlier approach) felt like overkill once I
verified that, so the PR is now just the deletion.

Diff: 1 file, 4 deletions, 0 additions.

Verified locally:

• import malariagen_data no longer mutates dask.config.
• pytest tests/anoph/test_snp_data.py (193 tests) plus test_frq, test_cnv_data,
  test_hapclust, test_dipclust, test_hap_data, test_aim_data — all pass, zero
  PerformanceWarning emitted from any da.take / da.compress call site.
• pre-commit + ruff + mypy clean.

CI is green on the new commit across all 10 checks. No rush on review — whenever it fits
your queue.