Name a fingering by the measure's place, not the number printed on it

[isc]

A fingering is stored under a name for the note it belongs to. That name began
with the measure's number attribute, which is a label and not an identity, and
the library ships scores that prove it.

What was wrong

Score	Symptom
Gnossienne No. 1	No barlines, so all 11 measures export as number="0". 872 notes shared 86 names: a fingering written on one measure was drawn on all eleven, and could not be taken off any of them.
The Entertainer (two one-staff parts)	OSMD numbers staves across the sheet, the injection numbered them within the part. 963 notes of the lower staff were filed under names the injection gave to the upper staff's notes.
Chopin Ballade No. 1	Same, with two parts of two staves each: the second part's staves landed on the first's.
The Entertainer (1902)	A second ending printed number="X1". parseInt reads NaN, OSMD reads null — so 57 notes were stored under one name and looked for under another, and their fingerings never appeared at all.
14 more scores	Cue and print-object="no" notes: the three walks that produce a name disagreed about which to count. An index is a running total, so past the first such note every fingering in the measure slid onto the note next door.

Eighteen of the ninety-eight shipped scores, not the four the ambiguity is
visible on.

The name

m<measure index>:<sheet-wide staff>:<voice>:<note>, defined once in
public/js/fingeringKeys.js.

• measure — its position in the score, which is both OSMD's SourceMeasures
  index and the nth <measure> of a <part>. Repeats do not disturb it: it
  names the source measure, engraved once, and the playback sequence already
  maps passes onto sources.
• staff — counted across the sheet, from each part's <staves>.
• note — every <note> that is not a rest, cue and hidden and grace notes
  included. Each walk keeps its own reasons to leave a note out of what it
  builds; none of them may reach the counter.
• the m — old and new names can never be confused. That matters because a
  record syncs whole and last-write-wins, so a device still on the old build
  must show nothing rather than something wrong.

The walks

The click map no longer walks the sheet at all — it reads the note model, which
already holds the name — so three copies of the counting rule become two. The
two that remain cannot be merged: injection has to happen before osmd.load().
So test/fingering_key_scheme_test.rb holds them against each other across the
whole library instead, checking that each name means a note of the same
pitch in both: 114,903 notes, 98 scores, no disagreement. The same test fails
on thirteen scores against the code this replaces.

Migration

There is no migration system, and none is needed. Each score's extraction builds
Map<old name, new name[]> as it names the notes, so the translation is exact
rather than re-derived from the file — which could not work anyway, since the
stored names carry the editor's spelling of the measure number
(MeasureNumberXML, null) and the file only offers the injection's
(parseInt, NaN).

On the first load of a score whose record still holds old names, they are
rewritten and saved with a fresh updatedAt so the fix reaches the player's
other devices, and what the injection could not place is drawn by the light
re-render a newly entered fingering already takes.

Two judgement calls, both in the code's comments:

• A name several notes answered to is copied onto all of them. The score
  looks exactly as it did — the player sees the same fingerings in the same
  places — and each is now its own name, so the ten they never wrote can be
  deleted one at a time. Picking a single winner would move their fingering to a
  measure they never touched.
• A name no note answers to is dropped. Those are the NaN: names, which
  could never be drawn; nothing visible is lost.

Testing

• New browser test run 5× standalone, plus once through the whole suite.
• Full Ruby suite: 116 runs, 0 failures. JS suite: 358 tests, 0 failures.
• Verified live in Chrome: a legacy 0:0:0:4 on the Gnossienne becomes 11
  explicit names and all 11 fingerings are drawn where they were.

Note for #348

That PR mirrors this scheme deliberately. It will want to rebase onto
fingeringNotesInDocument, which is the injection walk lifted out so a second
consumer — and the invariant test — can share it.

🤖 Generated with Claude Code