Bug#638753: libhdf5-serial-1.8.7: threadsafe memory leaks

[Gilles Filippini]

Control: tags -1 + moreinfo

Hi,

On Sun, 21 Aug 2011 17:35:26 +0200 Jerome BENOIT <g6299304p at rezozer.net>
wrote:
> Package: libhdf5-serial-1.8.7
> Severity: wishlist
> 
> Hello:
> 
> Since a while I have noticed via valgrind some memory leaks let by the function `H5Zfilter_avail':
> the following Code reproduce the issue:
> 
> --8><---------------------------------------------------
> // gcc -o test_H5Zfilter_avail test_H5Zfilter_avail.c -lhdf5
> // valgrind --leak-check=yes --leak-check=full --show-reachable=yes ./test_H5Zfilter_avail
> //
> 
> #include <stdio.h>
> #include <hdf5.h>
> 
> // http://wiki.hdfgroup.org
> #ifndef H5Z_BZIP2
> #define H5Z_BZIP2 307
> #endif
> 
> int main(int nargs, char *args[]) {
> 
> 	fprintf(stderr,
> 			"filter with identifier `%d' is %s registered\n",
> 		H5Z_BZIP2,
> 		(H5Zfilter_avail(H5Z_BZIP2))?"already":"not"
> 		);
> 
> 	return (0); }
> --><8-----------------------------------------------------
> 
> A closer look shows that the memory leakes are let by the threadsafe machinery.

When trying to reproduce this bug I only get 'still reachable' class
memory leaks, with the very same small memory footprint no matter how
many times H5Zfilter_avail was called. Is that what you experienced?
I tend to consider 'still reachable' memory leaks not leaks. Especially
when there're stable.

Thanks,

_g.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 508 bytes
Desc: OpenPGP digital signature
URL: <http://lists.alioth.debian.org/pipermail/pkg-grass-devel/attachments/20171218/7d14f9a1/attachment.sig>