2011/8/16 Kamil Dudka wrote:

> When using up2date versions of gcc/glibc with optimization turned on, valgrind
> reports a lot of false positives on string routines and other optimized stuff
> [1,2].  On some Linux setups valgrind refuses to start [3].  Upstream often
> does not provide any applicable solution [4].  I would not call valgrind
> generally available on linux systems, or at least not generally usable.

Fixes mentioned in this thread are not about forcing valgrind usage anywhere.

It is only related with lib/curl builds with TrackMemory feature
enabled, and preventing that this specific feature fools valgrind when
it is used to verify correctness of such a build when running the
testsuite or alone.

Once that said,...

We have 28 linux autobuilds, in i686 and x86-64 flavours using gcc and
icc, that are running with valgrind in use, every day, for at least a
couple of years. We get some noise from third party libraries (SSL
ones), and for that reason some of our test definitions instruct the
test harness to run that specific test without valgrind.

So for our purposes, valgrind mostly works. But it could help us even
more, if we were not fooling it with TrackMemory feature initializing
every single malloc().

-- 
-=[Yang]=-
-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette:  http://curl.haxx.se/mail/etiquette.html