Re: Fatal error compiling large project (Win10/2.19.82)

[Aaron Hill]

On 2018-12-18 7:24 pm, Saul Tobin wrote:
> It looks like the Fatal realloc error happens after the score finishes
> compiling while it's being written to a temporary file. At that point, 
> the
> Lilypond process is using ~1.8 G of RAM. Is it possible I'm running 
> into a
> 32 bit size limitation in Guile or something like that?

Very likely.  There have been a number of issues reported before 
regarding memory pressure.  The Windows build of LilyPond is 32-bit 
only.  Also, I do not believe it is "large-address aware" either, so 
that means the address space is strictly 2GiB.  When you factor in DLLs 
and other resources, 1.8GiB is probably a reasonable upper bound for the 
working set.

Due to a number of factors (bit-ness being one of them), I switched to 
using the 64-bit Linux build of LilyPond, running it under the Windows 
Subsystem for Linux (WSL).  This strategy, however, is not for the faint 
of heart.  You can search the mailing list for some posts I made on the 
subject; but if you are not already using WSL, it may not make sense to 
set it up just for LilyPond.

To work within the limits of the existing Windows build, you will 
ultimately have to simplify the score in some way.  Manually segmenting 
it could allow you to compile each section without running out of 
memory, leaving you with the (hopefully) trivial task of stitching 
together the individual outputs into a single result.


-- Aaron Hill