Re: memory (leak) debugging

On Tue, 31 Mar 2009, Dan Fandrich wrote:

[CC'ed over to the libcurl list as this is now more libcurl than libssh2]

> On Tue, Mar 31, 2009 at 10:16:44AM -0400, Daniel Johnson wrote:
>> I did notice that the libssh2-related tests explicitly disable valgrind
>> use, so it wouldn't have helped in this case.
>
> That was to avoid the OpenSSL false positives that result with the
> OpenSSL-using tests. Those explicit disables ought to be removed and
> replaced with a smarter test suite that filters the valgrind results like
> the other OpenSSL tests. Those problems shouldn't even happen if using
> GnuTLS with libssh2.

I actually commented out the ability to disable valgrind in runtests.pl when I
ran test 600 to 637 with curl only to see what valgrind says. It's actually
really useful to let curl's memory leak system trigger for a leak and then
read the valgrind log output for the exact allocating function call. That's
especially true when curl catches a leak caused by libssh2 since it does all
its mallocs from the single spot within libcurl.

So I think this disabling thing can be removed without causing any problems.

-- 
  / daniel.haxx.se
------------------------------------------------------------------------------
_______________________________________________
libssh2-devel mailing list
libssh2-devel_at_lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/libssh2-devel