valgrind vs glibc

[Simon Josefsson]

Hi, I'm looking at the gsasl builds, e.g, this one:

http://hydra.nixos.org/build/2657830/log

The only reason it fails is because of valgrind complains like this one:

==39998== Invalid free() / delete / delete[]
==39998==    at 0x4C250A2: free (in 
/...-valgrind-3.6.0/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==39998==    by 0x559B9CA: free_mem (in /...-glibc-2.12.2/lib/libc-2.12.2.so)
==39998==    by 0x559B571: __libc_freeres (in 
/...-glibc-2.12.2/lib/libc-2.12.2.so)
==39998==    by 0x4A21598: _vgnU_freeres (in 
/...-valgrind-3.6.0/lib/valgrind/vgpreload_core-amd64-linux.so)
==39998==    by 0x54BC41C: __run_exit_handlers (in 
/...-glibc-2.12.2/lib/libc-2.12.2.so)
==39998==    by 0x54BC4C4: exit (in /...-glibc-2.12.2/lib/libc-2.12.2.so)
==39998==    by 0x54A5C83: (below main) (in 
/...-glibc-2.12.2/lib/libc-2.12.2.so)
==39998==  Address 0x4024758 is not stack'd, malloc'd or (recently) free'd

This seems like a issue with the system installation of valgrind or
possibly glibc.  Does anyone have suggestion on how to silence this?
Could a system-wide valgrind suppression file be installed to silence
these?

I'm consdering disabling valgrind for the builds to avoid issues like
this.  My experience is that automatically enabling valgrind does not
work well -- false positives are too common, and typically humans are
required to properly parse the output.

/Simon