This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: ACATS tests that fail valgrind check

Hi Arnaud, thanks for commenting.  I made no attempt so far to
analyse the cause of the valgrind warnings, in particular I didn't
try to work out if the warnings were correct or not.  As you correctly
point out, valgrind sometimes gets it wrong.  For example, it used to have
problems with multi-threaded programs and also signal handling, probably it
still does.

On Wednesday 12 January 2005 11:16, Arnaud Charlet wrote:
> > (1) "Conditional jump or move depends on uninitialised value(s)"
> I believe many of these are actually due to the fact that valgrind does not
> properly support trampolines (used to implement pointers to nested
> subprograms).

Right.  Valgrind does support nested functions, but not pointers to nested

> > Correct programs should run the same whether run under the control of
> > valgrind or not.

> Unless valgrind has bugs and/or does not support some features such as
> tranmpolines.

Notice the "should".  I thought of mentioning explicity that valgrind is not
perfect, but that seemed too obvious to mention!

> Since valgrind is not a tool that has been qualified in any way for Ada
> programs, you should take its output with a grain of salt, although 
> valgrind most likely detected some real problems here.

I plan to enter all these reports into bugzilla, then start analysing them to
see which ones are bogus (it is good to have bogus reports and a note saying
they are bogus in bugzilla, since it will save time if someone else sees the
same valgrind failures).

All the best,


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]