5.11.0: NPE in Voices.refineSystem (RHYTHMS) on systems with varying staff counts

[jdburrage]

Summary

On a score whose systems hold different numbers of staves — an orchestral score that
omits the staves of resting instruments — the RHYTHMS step throws
NullPointerException in Voices.refineSystem, and the affected sheets never reach PAGE.

Environment

Audiveris 5.11.0, build 9e1e55cd2746037d059345881c53e6a6754bffbd
Windows 11 10.0, amd64
OpenJDK 64-Bit Server VM (build 25.0.3+9-LTS, mixed mode)
Tesseract OCR 5.5.2

Input

Haydn, Symphony No. 45 "Farewell", Timothy Lanfear 2020 engraving, 33 pages — freely
downloadable from IMSLP as file #633504:
https://imslp.org/wiki/Symphony_No.45_in_F-sharp_minor,_Hob.I:45_(Haydn,_Joseph)

This symphony is the natural test case for varying staff counts: the finale is written so
that the players leave the stage one at a time, and this engraving prints that literally.
Page 33 alone steps 6 → 5 → 3 → 3 → 2 staves system by system.

Command

Audiveris.exe -batch -export -save -output <dir> -- IMSLP633504-PMLP51750-045-imslp.pdf

Actual

Sheets 30, 32 and 33 stop at step 18 of 20; sheet 31 at step 17. Exit code 1, and because
no sheet completes the book, Created scores: [] and no MusicXML is written at all — the
other 27 sheets, which transcribe correctly, produce nothing.

Sheets 30, 32, 33:

WARN [#30] Book.java:2044 | Error processing stub java.lang.NullPointerException:
  Cannot invoke "org.audiveris.omr.sheet.rhythm.Measure.purgeVoices()" because "measure" is null
java.util.concurrent.ExecutionException: java.lang.NullPointerException: ...
	at org.audiveris.omr.sheet.SheetStub.doOneStep(SheetStub.java:519)
	at org.audiveris.omr.sheet.SheetStub.reachStep(SheetStub.java:1405)
	at org.audiveris.omr.sheet.Book.reachBookStep(Book.java:2035)
	at org.audiveris.omr.sheet.Book.transcribe(Book.java:2626)
	at org.audiveris.omr.sheet.Book.export(Book.java:593)
	at org.audiveris.omr.CLI$ProcessingTask.processBook(CLI.java:927)
Caused by: java.lang.NullPointerException:
  Cannot invoke "org.audiveris.omr.sheet.rhythm.Measure.purgeVoices()" because "measure" is null
	at org.audiveris.omr.sheet.rhythm.Voices.refineSystem(Voices.java:430)
	at org.audiveris.omr.sheet.rhythm.PageRhythm.processRanges(PageRhythm.java:288)
	at org.audiveris.omr.sheet.rhythm.PageRhythm.process(PageRhythm.java:248)
	at org.audiveris.omr.sheet.rhythm.RhythmsStep.doit(RhythmsStep.java:192)

Sheet 31, same step, different symptom:

WARN [#31] Book.java:2044 | Error processing stub java.lang.RuntimeException:
  java.lang.IndexOutOfBoundsException: Index 8 out of bounds for length 8

Expected

Either the sheets transcribe, or they fail individually without costing the other 27
sheets their export.

Notes

Voices.refineSystem appears to assume that a measure exists at a given index in every
part of the system. Where a system omits the staves of resting instruments, the part/measure
grid is not rectangular, and the lookup returns null.

The sheets that fail are exactly the ones whose systems vary in staff count, read off
book.xml (<part> count per <system>):

sheets  3..29   mostly uniform
sheet   30      [10, 11]
sheet   31      [9, 10]
sheet   32      [6, 7]
sheet   33      [1, 3, 5, 6]

A related but distinct failure in the same area affects export even when no sheet crashes;
filed separately.

[hbitteur]

I tried to process your example
It immediately failed at load time:

INFO  [IMSLP426348]                      Book 560  | 34 sheets in D:\soft\cases\Issue-970\IMSLP426348-PMLP51750-1242-Haydn-Symphonie45-000-Score.pdf
INFO  [IMSLP426348#1]                 SheetStub 1278 | Loaded image #1 9700x12916 from D:\soft\cases\Issue-970\IMSLP426348-PMLP51750-1242-Haydn-Symphonie45-000-Score.pdf
WARN  [IMSLP426348#1]                 SheetStub 411  | Too large image: 125 285 200 pixels (vs 20 000 000 max)

This is way too large for a sheet image!

I used this page to download:
https://s9.imslp.org/files/imglnks/usimg/6/67/IMSLP426348-PMLP51750-1242-Haydn-Symphonie45-000-Score.pdf

Are we talking of the same file?
I picked up the first one on the page you referenced. Perhaps a different one would exhibit smaller pages.
Could you double-check?

[hbitteur]

In fact, I should have taken the last one on the web page,
that is IMSLP633504-PMLP51750-045-imslp.pdf '0.735 MB) instead of IMSLP426348-PMLP51750-1242-Haydn-Symphonie45-000-Score (4.5 MB)

[hbitteur]

Audiveris is able to cope with systems with varying numbers of staves.

But here, on sheet 30, we have a different problem:

Some staves have missing measures, see staff 1 and staff 5, within the same system.

In a system, Audiveris accepts staves to be present or absent, but the present staves must be complete.

In other words: in a system, all staves must exhibit the same number of measures.