fix: nan-variant reducers drop behavior and attrs#4093
Open
henryiii wants to merge 4 commits into
Open
Conversation
henryiii
added a commit
that referenced
this pull request
Jun 10, 2026
24 tests covering: behavior dict and attrs preservation across nanmin, nanmax, nanmean, nanvar, nanstd; correctness on nan-free data vs non-nan variants; correct NaN exclusion; highlevel=False semantics for nanvar/nanmean. Assisted-by: ClaudeCode:claude-fable-5
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files
|
ikrommyd
requested changes
Jun 11, 2026
ikrommyd
left a comment
ikrommyd
left a comment
Collaborator
There was a problem hiding this comment.
I searched a bit for ak_nan_to_none in the code and I notice that the arguments we pass when it's called inside another highlevel function are inconsistent. Does this PR fix all of them?
Intuitively I'd think we want highlevel=False inside the highlevel function implementation cause it's already unwrapped and we'd want to wrap it once at the end. highlevel=True should do extra wrapping/unwrapping when its called inside another highlevel function.
Collaborator
There was a problem hiding this comment.
This file has test classes. I'm pretty sure we never do that here.
Pass highlevel=True, behavior, attrs into the intermediate nan_to_none call in ak_min, ak_max, ak_mean, and ak_var so that custom behavior dicts and attrs are not silently dropped during nan filtering. Also reorder the dtype-tuple in ArgMax.apply to match ArgMin and the kernel spec (parents, offsets, starts instead of parents, starts, offsets). Assisted-by: ClaudeCode:claude-fable-5
24 tests covering: behavior dict and attrs preservation across nanmin, nanmax, nanmean, nanvar, nanstd; correctness on nan-free data vs non-nan variants; correct NaN exclusion; highlevel=False semantics for nanvar/nanmean. Assisted-by: ClaudeCode:claude-fable-5
Rename test file from hyphens to underscores (required by validate-test-names.py). Replace 24-test class-based suite with a single parametrized function covering all five nan-variant reducers; the new tests actually verify result.behavior and result.attrs on array-valued output, which is where the original bug was observable. Assisted-by: ClaudeCode:claude-sonnet-4-6
henryiii
force-pushed
the
henryiii/fix-nan-reducers
branch
from
fedb22a to
55c5bca
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
🤖 AI text below 🤖
Summary
Fix a bug where nan-variant reducers (
nanmin,nanmax,nanmean,nanvar) lost the input array'sbehaviordict andattrsduring computation because the intermediatenan_to_nonecall was invoked withhighlevel=Falseand/orNonefor behavior/attrs.src/awkward/operations/ak_min.py:147—nan_to_none._impl(array, False, None, None)→(array, True, behavior, attrs). Matches siblingsnansum,nanprod,nanstd,nanargmin,nanargmax.src/awkward/operations/ak_max.py:147— same fix asak_min.py.src/awkward/operations/ak_mean.py:179—nan_to_none._impl(x, False, behavior, attrs)→(x, True, behavior, attrs).src/awkward/operations/ak_var.py:174—nan_to_none._impl(x, highlevel, behavior, attrs)→(x, True, behavior, attrs). Passing the user'shighlevelflag into the intermediate conversion changed propagation semantics whenhighlevel=Falsewas requested.src/awkward/_reducers.py:249-256—ArgMax.apply's kernel-lookup dtype tuple had(result, fromptr, parents, starts, offsets)whereasArgMinand the kernel spec use(..., parents, offsets, starts). Reordered to match.Regression tests are in
tests/test_4093_nan_reducer_behavior_attrs.py— one parametrized function over all five nan-variant reducers, verifyingresult.behaviorandresult.attrsare preserved on array-valued output (axis=1), which is the exact code path where the bug was observable.AI assistance
This fix was generated using Claude Code (automated multi-agent review, model: claude-fable-5). The findings were pre-verified against sibling functions (
nansum,nanstd, etc.) that already implement the correct pattern.Test plan
tests/test_4093_nan_reducer_behavior_attrs.py(5 parametrized cases, one per nan-variant reducer)prek -a --quietpasses (no lint issues)🤖 Generated with Claude Code