On 6 February 2018 at 03:13, Jeff King <peff@xxxxxxxx> wrote: > On Mon, Feb 05, 2018 at 08:28:10PM +0700, Duy Nguyen wrote: >> I learned SANITIZE=leak today! It not only catches this but also "dst". >> >> Jeff is there any ongoing effort to make the test suite pass with >> SANITIZE=leak? My t2038 passed, so I went ahead with the full test >> suite and saw so many failures. I did see in your original mails that >> you focused on t0000 and t0001 only.. > > Yeah, I did those two scripts to try to prove to myself that the > approach was good. But I haven't really pushed it any further. > > Martin Ågren (cc'd) did some follow-up work, but I think we still have a > long way to go. Agreed. :-) > My hope is that people who are interested in > leak-checking their new code can run some specific script they're > interested in, and maybe fix up one or two nearby bits while they're > there (either by fixing a leak, or just annotating via UNLEAK). Then we > can slowly converge on correctness. :) Yeah. My experience is that it's easy -- or was for me, anyway -- to get carried away trying to fix all "related" leaks to the one you started with, since there are so many dimensions to search in. Two obvious aspects are "leaks nearby" and "leaks using the same API". This is not to suggest that the situation is horrible. It's just that if you pick a leak at random and there is a fair number of "clusters" of leaks, chances are good you'll find yourself in such a cluster. Addressing a leak without worrying too much about how it generalizes would still help. ;-) Martin
