Issue 1201 in lilypond: GUB regtest building flaky?

[lilypond]

Status: Accepted
Owner: ----
Labels: Type-Build Priority-Critical

New issue 1201 by percival.music.ca: GUB regtest building flaky?
http://code.google.com/p/lilypond/issues/detail?id=1201

Neil wrote:
> There's something fishy about the latest comparisons: a number of .log
> changes suggest broken regtests, and if you check collated-files.html,
> they are visibly broken (e.g., bookparts.ly:
> http://lilypond.org/doc/v2.13/input/regression/collated-files.html#bookparts.ly);
> but they appear to be false positives (at least for the mingw binary
> and master).

A bit of poking around on my system confirms this.  Building bookparts.ly  
with 2.13.29-2 (not uploaded) works just fine; it does 1 page for each of  
the 3 sections.  But the log file of
target/linux-x86/build/lilypond-git.sv.gnu.org--lilypond.git-release-unstable/out/lybook-testdb/52/lily-23a75a14.log
shows:
Processing  
`/main/src/gub/target/linux-x86/build/lilypond-git.sv.gnu.org--lilypo
nd.git-release-unstable/out/lybook-testdb/52/lily-23a75a14.ly'
Parsing...
Renaming input to:  
`/main/src/gub/target/linux-x86/src/lilypond-git.sv.gnu.org--
lilypond.git-release-unstable/input/regression/bookparts.ly'
Interpreting music...
Preprocessing graphical objects...
Finding the ideal number of pages...
Fitting music on 1 or 2 pages...
Drawing systems...
Finding the ideal number of pages...
Fitting music on 1 page...
Drawing systems...
Finding the ideal number of pages...
Fitting music on 4 or 5 pages...
Drawing systems...
Writing header field `texidoc' to  
`/main/src/gub/target/linux-x86/build/lilypond
-git.sv.gnu.org--lilypond.git-release-unstable/out/lybook-testdb/52/lily-23a75a1
4.texidoc'...

2.13.29-1 (which was uploaded, and is still visible on the web) was built  
with GUB 33eecb067c1cf07efc94c3c0cc99819a7c76a23a , which is exactly what  
2.13.28-1 (which seemed to be ok) used.  So the difference probably comes  
from lilypond git... it's just possible that 2.13.28-1 was polluted from a  
previous compilation on my computer.  I usually run a nuke-gub.sh script,  
which cleans out any *lilypond* in the build dirs, but I can't swear that I  
remember to run this, since .28 was built in my 1 day between Bordeaux and  
Glasgow.  But the only difference between .28 and .26+.26 is Patrick's  
three NSIS+lilypad patches.

I don't know where to go from here.  Anybody feel like looking for anything  
suspicious in lilypond git between .28 and .29 ?