fix(score): rebuild viewer beat positions when a score is re-imported

[FelipeDefensor]

The bug

SvgViewer.load_svg_data only computed the beat-to-x mapping when viewer_beat_x was falsy, and nothing ever cleared it. Importing a second musicXML into a score timeline that already had one kept the first score's mapping.

Reproduced by loading an SVG whose markers sit at x=100/150 and then one at x=900/950: timeline.viewer_beat_x and viewer.beat_x_position still read {1.0: 100.0, 1.5: 150.0}. Everything downstream is then wrong for the new score — scroll_to_time, _get_closest_time, the measure tracker, and every stavenote's seek_x.

The fix

The mapping is rebuilt from the SVG on every load. That needs the data markers to still be there, so load_svg_data no longer writes the stripped SVG back over svg_data. Stripping is cheap and now runs on every load anyway; a stored mapping is still honoured as a fallback, for files saved before this change.

Two things made the pair awkward to decouple, and both are handled:

1. The viewer_beat_x setter ignored falsy values, so set_data("viewer_beat_x", {}) was a silent no-op and the mapping could not be cleared through the normal path. It assigns x_pos or {} now, and ScoreTimeline.clear() discards both halves together — keeping either one alone would leave the timeline describing a score that is no longer there, and would make an undo overwrite the value it just restored.

2. _restore_timeline_state restores SERIALIZABLE attributes one at a time, svg_data before viewer_beat_x, so the viewer reloads while only half of the pair has been applied. viewer_beat_x joins UPDATE_TRIGGERS and both updaters go through _reload_svg_view, so whichever attribute arrives second leaves the viewer consistent. An empty svg_data closes the viewer instead of being handed to etree.fromstring. SERIALIZABLE is untouched — the per-timeline hash built from it goes into the .tla file.

Plus one small related change: reset_svg no longer reaches the viewer through the svg_view property, which would build one — re-registering its toolbar commands against a throwaway widget — only to destroy it again.

The second commit removes _strip_beat_x_markers, which has no callers left now that every load goes through _get_beat_x_pos, and moves its tests over.

Tests

Eight new tests in TestViewerBeatPositions: clearing drops the mapping; a score imported after a clear uses its own beat positions; the same for a score imported over another without a clear; undo/redo across both transitions displays no error and leaves the viewer consistent; saving and reloading preserves the positions; a marker-less SVG falls back to the stored mapping; and an SVG with no source at all reports SCORE_SVG_CREATE_ERROR. Four of them fail on dev.

Full suite green (1835 passed). The tests/test_app.py failures under -n auto are the pre-existing xdist flake — all 90 pass serially.

Notes for the reviewer

• svg_data now keeps its markers, so .tla files get bigger. Files written before this change still load, through the stored-mapping fallback.
• SCORE_SVG_CREATE_ERROR was unreachable on dev: validate_pre_validated always returns True, so the success flag that guarded it was always true. This PR makes the error genuinely reachable — when neither markers nor a stored mapping are available — and covers it with a test.
• Overlaps with fix/472-score-clear-svg-viewer, which also discards svg_data on clear and closes the viewer. That branch sits on a different base where svg_view returns None rather than lazily building a viewer, so the two will conflict in the same three places.

🤖 Generated with Claude Code