Hi Evan,
On 07/26/2012 04:44 AM, Evan Huus wrote:
On Wed, Jul 25, 2012 at 7:03 PM, Martin Mathieson
<martin.r.mathieson@xxxxxxxxxxxxxx> wrote:
On Wed, Jul 25, 2012 at 6:12 PM, Evan Huus<eapache@xxxxxxxxx> wrote:
I've been playing around in my head for a while now with the idea of
running fuzz tests under valgrind. I notice the fuzz-test script
already sets a bunch of environment variables for memory checking, so
I'm not sure if valgrind would really add anything.
How extensive are the memory checks turned on by the fuzz-test script?
Would valgrind be a useful addition, or would it be mostly redundant?
Thanks,
Evan
I'm not sure either, but out of the 3 problems reported by valgrind that I
just fixed:
- 1 was writing outside of allocated data (which the canary checks may have
picked up on)
- 2 were reading/displaying/branching-based-upon uninitialized data (which I
don't expect other types of checks would spot)
Martin
Basic support added in revision 44024, so it's there if people want to
use it. I don't know if it would be worth turning on for the
build-bot, as it does slow down the check considerably, but it will
probably catch a few more errors.
I think it is a good idea. Maybe we could have a fuzz-test run with Valgrind
once a week.
The memory corruption I fixed recently has been identified using Valgrind as well.
Cheers,
Balint
Evan