[lmeyerov]

Summary

• Fixes GFQL 0.61.2: timezone-aware datetime equality loses matches for plain date strings #2233.
• Localizes naive temporal equality bounds to an aware datetime column’s timezone; offset-bearing bounds continue to compare as instants.
• Ambiguous/nonexistent localized wall times match no rows instead of silently choosing a fold.
• Retains naive-column behavior, duration controls, and malformed-bound E302 handling.

Tests

• Compact R1 matrix covers literal, parameter, and serialized JSON forms; UTC/non-UTC midnight; equivalent offsets; wrong-instant, noon, and next-day negatives; ambiguous/nonexistent bounds; null-only and empty typed frames.
• python -m pytest -q graphistry/tests/compute/gfql/test_temporal_leak_family_1915.py graphistry/tests/test_compute_filter_by_dict.py -k 'not cudf' --no-header → 95 passed, 21 deselected.
• ruff check graphistry/compute/filter_by_dict.py graphistry/tests/compute/gfql/test_temporal_leak_family_1915.py → passed.
• python -m mypy graphistry/compute/filter_by_dict.py graphistry/tests/compute/gfql/test_temporal_leak_family_1915.py → same 14 pre-existing errors.
• DGX cuDF 26.02.01, exact tracked source at fdb24db92e0db93c7f98dc650b74fb4fb0cd1432 → 116 passed, 0 skipped.
• Exact-head CI run 40053270807, CodeQL, and RTD are green.

[aucahuasi]

I tested this branch at d1a27c40b against the repro in #2235. Setup: pandas 3.0.6, Python 3.12, default (pandas) engine, strict=True. The comparison runs used the same environment with 0.61.1 and 0.61.2.

Equality is fixed, once pytz is installed:

Column tz	Query	0.61.2	this branch
UTC	n.ts = '2024-01-01'	[]	[a]
UTC	n.ts = '2024-01-01T00:00:00'	[]	[a]
UTC	JSON filter_dict {"ts": "2024-01-01"}	[]	[a]
America/New_York	n.ts = '2024-01-01'	[]	[a] (midnight local)
America/New_York	n.ts = '2024-01-01T05:00:00'	[]	[c] (05:00 local)
America/New_York	n.ts = '2024-01-01T05:00:00Z'	[a]	[a] (instant)
America/New_York	MATCH (n {ts: '2024-01-01'})	[]	[a]

Timezone-naive columns and <> are unchanged.

1. pytz is not a declared dependency. Without it, the fixed path raises instead of matching:

MATCH (n) WHERE n.ts = '2024-01-01' RETURN n.nid AS id
-> ModuleNotFoundError: No module named 'pytz'

(JSON filter_dict raises the same.)

• Where: filter_by_dict.py:44 does import_module("pytz.exceptions").
• Why it's missing: pandas 3 lists pytz only under its timezone/all extras, and graphistry's metadata doesn't require it. A pandas 2.x CI environment hides this, because pandas 2 still depends on pytz.
• Suggested fix: import pytz lazily with a fallback, catching the exceptions pandas itself raises for ambiguous or nonexistent local times, or add pytz as a dependency.

2. IN / IsIn with offset-less strings still match nothing on tz-aware columns. This is pre-existing, the same on 0.61.1, 0.61.2 and this branch:

Column tz	Query	all three versions
UTC	n.ts IN ['2024-01-01', '2024-02-01']	[] (expected [a, b])
UTC	JSON {"type": "IsIn", "options": ["2024-01-01", "2024-02-01"]}	[]
UTC	n.ts IN ['2024-01-01T00:00:00Z']	[a] (correct)

It's the same localization gap as equality, so it might be worth covering here too.

The other items in #2235 (2 to 7) are unchanged on this branch, as expected.