[azfoo]

🤖 Claude

Moves #621 (beat units and time signatures)'s beat pattern parser into the core, for the beat importer and for adding a beat timeline.

What changed

• packages/tilia-core/src/tilia_core/timemap/pattern.py: tilia/timelines/beat/pattern.py from feat: beat patterns, beat units, time signatures and quarter stamps #621, unchanged (the same blob, 236c76e5, at feat/beat-unit's head 0ee8ea45, at 07cb6d4f and at a17bbd83). It sits next to units.py, the other module from feat: beat patterns, beat units, time signatures and quarter stamps #621: the beat importer and adding a beat timeline will call it, and so will the app's beat timeline, its validators and the beat pattern dialog once the app imports the core's copy. It imports nothing from the app and doesn't log, so nothing had to change. The app keeps its own copy until it switches to the core's.
• packages/tilia-core/tests/timemap/test_timemap_pattern.py: feat: beat patterns, beat units, time signatures and quarter stamps #621's tests/timelines/beat/test_pattern.py, with only the import changed (first commit), plus a TestFormatBars class (second commit): runs collapsed, nested groups not recovered, and parse(format_bars(bars)) giving the same bars back. The moved tests cover parse and group_runs only.

Tests

• uv run --package tilia-core pytest packages/tilia-core -q, on the current v1-prerelease: 794 passed, 21 skipped, 8 xfailed (Python 3.11), 55 of them in test_timemap_pattern.py.
• The pattern tests on Python 3.10: 55 passed.
• python scripts/check_packages.py: all package checks passed (both wheels, fresh installs without PySide6 or the app, both test suites), on the current v1-prerelease.
• ruff check and black --check on both files: clean.

Guards: what removing each one does

Each check neutralised in turn, the pattern tests run, the file restored (its blob checked afterwards). Run on the module and tests as they are here:

Guard	Without it
MAX_BARS, after each item (parse_items)	1 failed, 54 passed
MAX_BARS, before a group is repeated (parse_item)	55 passed: untested, see below
MAX_DEPTH	3 failed, 52 passed
MAX_DIGITS	2 failed, 53 passed
ASCII digits only (_DIGITS, not str.isdigit)	4 failed, 51 passed
repeat count at least 1	1 failed, 54 passed
bar length at least 1	2 failed, 53 passed
leading zeros not counted as digits	1 failed, 54 passed
format_bars collapsing runs	6 failed, 49 passed

Untested: the check before a group is repeated, the one that keeps 999999[999999[4]] from exhausting memory. Without it, 999999999[999999999[4]] is still refused by the check after each item, but only once the list of 999,999,999 bars is built (22 s and several GB here), so every test passes. The same holds for #621's own tests. A cheap test that fails without it is TimeLineAnnotator/project-management#301 (test the large-group guard); it isn't in this PR, which moves the code and its tests as they are.

Part of TimeLineAnnotator/project-management#270

🤖 Generated with Claude Code