Re: Issue 1475 in lilypond: unknown crash on a large score

[lilypond]

Comment #4 on issue 1475 by address@hidden: unknown crash on a large  
score
http://code.google.com/p/lilypond/issues/detail?id=1475

I added 15 instances, for a total of 16 instances, and ended up using 3Gb  
of real memory and 3Gb of real memory before getting the error

lilypond (2769( malloc: *** mmap(size=2097152) failed (error code=12)
*** error: can't allocate region

I got 9 of the same error message, then

terminate called after throwing an instance of 'std::bad_alloc'

Working on OSX 10.5


14 instances gave me the same error.

13 gets me up to 2GB real memory and 2.7GB Virtual memory, but so far no  
crash.  It finally finished, and gave 8 instances of
programming error: insane spring min_distance requested, ignoring it
continuing, cross fingers

12 used about 600 MB real and 1.1 GB Virtual and compilation succeeded  
after about 3 minutes.

Eliminating the markups between the scores and going up to 30 instances  
took the page count to 9 or 10 pages, but everything ran really quickly.

Going to 45 instances of the score took the page count to 14 or 15, and  
gave 1 instance of
programming error: insane spring min_distance requested, ignoring it
continuing, cross fingers.

But it still completed quickly.

60 instances of the score took the page count to 19 or 20.  I got 3 insane  
springs, but it finished quickly.

With this information, I started to suspect it's the interaction between  
the markup and the score, so I went to a much more simple test:


\score {
  \repeat unfold 8 {c''4 c c c}
}

\markup { t }


When repeated 16 times, I got the memory allocation error.

13 repeats of this test also gave me huge memory usage 2GB real and 3 GB  
Virtual, but it succeeded.

12 repeats kept me down in the 500 MB Real and 1.5 GB Virtual memory, where  
I had success in the past.  And it succeeded this time as well.

It looks to me like there is some kind of memory usage and calculation that  
is exponential  when we have markups between scores.  For my system,  
somewhere between 12 and 13 is the limit.