Bug 693339

Summary: Ghostscript still has sections that use malloc/free
Product: Ghostscript Reporter: Robin Watts <robin.watts>
Component: GeneralAssignee: Sebastian Rasmussen <sebastian.rasmussen>
Status: CONFIRMED ---    
Severity: normal CC: sphinx.pinastri
Priority: P4    
Version: master   
Hardware: All   
OS: All   
Customer: Word Size: ---

Description Robin Watts 2012-09-18 16:09:29 UTC 
Some sections of the ghostscript code still call direct to malloc/free rather than using the ghostscript memory functions.

One such example of this is the luratech jbig2 wrappers.

We should grep the source for a list of all such places, document it here, then fix it.

Initially at least I am assigning this bug to Alex as he 'owns' the luratech wrappers.
Comment 1 Robin Watts 2012-09-18 16:09:49 UTC 
Assigning to Alex.
Comment 2 Alex Cherepanov 2012-10-08 00:00:53 UTC 
Calls to malloc() have been eliminated in JPX code.

http://git.ghostscript.com/?p=ghostpdl.git;a=commitdiff;h=6cbb5969cc2edab6417c673194f1c1d019d275ef

There are still malloc() calls in OpenJpeg library that has no API to
introduce a custom memory allocator.
Comment 3 Ray Johnston 2012-10-08 18:56:34 UTC 
This is very low priority since the only ones we've seen that really care
about controlling the allocator are embedded customers and they use (or should
use) the Luratech decoder.

About the only impact of not using the gs allocators is that the -K limit and
the memory use reporting will not take into account openjpeg allocations (and
any leaks in the openjpeg code will not be as evident without using OS tools).
Comment 4 Ken Sharp 2013-06-12 12:26:24 UTC 
Passing to Henry as openjpeg is a decoder. Not sure of v2 allows us a custom memory allocator or not.
Comment 5 Robin Watts 2013-10-01 06:46:05 UTC 
v2 does not fix this.

We *could* add some small tweaks in opj_malloc.h to add a new predefine that would force everything through opj_{malloc,calloc,realloc,free,aligned_{malloc,free}} (indeed PERF_MALLOC_OPT almost does this, but ends up mapping aligned_malloc onto opj_malloc), but this doesn't help us as we have no way of getting a gs_memory_t * into those functions.

We could fudge it with a static variable, and wrapping all the opj calls in a lock, but that's horrid.

The API changes to fix openjpeg are equally horrid though :(
Comment 6 Peter Cherepanov 2021-01-03 00:59:08 UTC 
The OpenJpeg library that is bundled with Ghostscript still has no memory allocation callbacks.