Copy #621's units.py into tilia_core.timemap again, after its four fixes

[azfoo]

🤖 Claude

Copies #621 (beat units and time signatures)'s units.py into tilia_core.timemap again, after four more changes there. #732 (copy units.py again) made the previous copy.

• tilia_core/timemap/units.py is feat: beat patterns, beat units, time signatures and quarter stamps #621's tilia/timelines/beat/units.py at 0ee8ea45, unchanged: the same git blob (537e3be). Since 07cb6d4f, the commit Copy #621's units.py into tilia_core.timemap again #732 copied, feat: beat patterns, beat units, time signatures and quarter stamps #621 has made four changes:
  • beat_unit_id is the id of the beat that carries the governing unit in the app too, so the field has one meaning in both copies and one docstring. The core's meters_of already gave beat ids, so nothing else here changes.
  • A unit longer than MAX_UNIT_LENGTH is refused with "'1111…' is longer than 32 characters.", instead of "is not a whole number or fraction".
  • MeasureMeter turns units and beat_values into tuples itself, so a meter built from lists can be hashed too.
  • quarter_length makes only an int a Fraction; Fraction() would also have read text and floats.

Tests

• test_timemap_units.py ports feat: beat patterns, beat units, time signatures and quarter stamps #621's changed and new cases:
  • the two messages for a long unit;
  • TestMeasureMeter: a meter built from lists holds tuples and hashes like one built from tuples, and quarter_length("1/2") raises TypeError.
• I wrote these first: they failed against the previous copy (5 failures) and pass with this one.
• uv run --package tilia-core pytest packages/tilia-core -q: 796 passed, 21 skipped, 8 xfailed, on v1-prerelease as of 2026-10-19. python scripts/check_packages.py passes. black and ruff are clean.

Guarantees. I neutralised the code behind each one, ran the suite, then restored it:

Guarantee	Code neutralised	Tests failing
a meter turns lists into tuples	__post_init__'s body	1
quarter_length reads no text	Fraction() for anything	1
a long unit is called too long	the old message	3

Part of TimeLineAnnotator/project-management#247

🤖 Generated with Claude Code

[azfoo]

🤖 Claude

Checked against #621 (beat units and time signatures) at 0ee8ea45: no problems found.

• The copy. units.py here and #621's tilia/timelines/beat/units.py at 0ee8ea45, its branch's head, are the same blob (537e3be).
• The ported tests are #621's new and changed cases, case for case: the two long-unit messages and TestMeasureMeter. #621's other test changes are about the app's own beat_unit_id, which the core's tests already check as a beat id.
• The guarantees table reproduces exactly: 1, 1 and 3 failing tests, and 676 passed again once restored.
• With #736 (time signatures in the measure table), merged since this branch was cut: the commit applies cleanly on today's v1-prerelease, and 741 tests pass, the shared examples included. So a rebase merge needs no new push.
• #621's quarter stamps and meters didn't change with these fixes: re-running the script in #736's description against #621 at 0ee8ea45 gives the same expected values for all 18 shared examples.
• CI is green on all three systems.

[azfoo]

🤖 Claude

90c3223f copies #621 (beat units and time signatures)'s units.py again, after one more change there. This updates the description above: the copy is now #621's tilia/timelines/beat/units.py at e5da86a9, unchanged, the same git blob (2ef3be79).

What changed in #621 since 0ee8ea45: parse_units refuses a fraction whose denominator, in lowest terms, is above 128 (MAX_DENOMINATOR, the limit of a beat unit's own denominator). It uses the message of any unreadable unit: '1/129' is not a whole number or fraction such as 1/2. Whole numbers are unchanged. Sums of units, such as a row's ts_numerator and the running quarter stamps of #747 (quarter stamps), then have denominators that divide one fixed number. Before, units such as 1/(10**29 + i) in measure i made them grow without limit.

Tests. test_timemap_units.py ports #621's new cases:

• 1/128, 127/128, 1+1/128 and 2/256 (read as 1/128) are read;
• 1/129, 3/7919, 1+1/129 and 1/100000000000000000000000000001 are refused with the message above;
• 129, 7919 and 258/2 are read.

Runs:

• uv run --package tilia-core pytest packages/tilia-core -q: 807 passed, 21 skipped, 8 xfailed (796 before, plus the 11 new cases).
• python scripts/check_packages.py passes. black and ruff are clean.

Guarantee. With the new check neutralised in the core's copy, 4 tests fail: the 4 refusals. The other 7 new cases pass either way. I restored it after the run.

Checked on #747 (quarter stamps), in a throwaway checkout of its head f93e17dc with this units.py. The input was 8,000 two-tap measures, measure i with the beat unit 4 / 1/(10**29 + i).

	build_rows	time_map_for + one qstamp	last row's repr and JSON
v1-prerelease	0.20–0.21 s	0.21–0.23 s	work
#747 as it is	8.7 s	49.7 s	raise ValueError
#747 with this units.py	0.29–0.35 s	0.41 s	work

• With this units.py, the units are refused, so the measures keep the assumed 4/4 and the last row's qstamp is 15998.
• The remaining gap to v1-prerelease is linear: 0.21 s for 4,000 measures and 0.98 s for 16,000. Part of it is the log line written for each refused unit. With plain 1 units, feat(core): quarter stamps, distances and lengths on the time map #747 builds the same rows in 0.17 s.
• With every allowed denominator in turn (1/1 … 1/128), the last qstamp's denominator has 176 bits, the rows take 0.19 s, and repr and JSON work.
• feat(core): quarter stamps, distances and lengths on the time map #747's own core suite, with this PR's units.py and test_timemap_units.py, gives 891 passed, 3 skipped, 1 xfailed.
• I didn't change feat(core): quarter stamps, distances and lengths on the time map #747's branch.

Part of TimeLineAnnotator/project-management#247