On 9/7/2016 6:53 AM, Paul Jakma wrote: > On Tue, 6 Sep 2016, Paul Jakma wrote: > >> I have another patch that makes valgrind clean for me on the bncs, but >> I left the bncs in place while the peer is defined. See patch posted >> just there. > Ok, my variation on your fix (hope you don't mind me re-doing that not at all.
> - I > noted what the leak was from your patch, humm, don't think this is right as all our patches are run against valgrind prior to submitting. Now, it does look like this came from an integration problem as we added the AFI and someone else added the new per afi processing. As both sets of patches were sitting in the patch list the current process has the integrator (release master) needing to find / fix these. This goes to process and how we need patches submitted against an active dev branch so that the submitter (who ever submitted last in this case) does the integration and can resolves any problems/leaks. This would be both more scalable and efficient as a single integrators (release master) doesn't need to learn all the changes/interactions. I'll be happy to resubmit (including pull request) our outstanding patches against the dev/automerge branch once /8 is done to ensure it integrates cleanly and passes martin's tests. > but started again to try work > out the refcounting) + your other leak fix + your AFI_ETHER patch (which > I havn't tested) + my OPEN/FSM fixes (CI with previous hit some > collision handling problem in bringing up sessions): > > - valgrinds clean from start to finish with some simple BGP sessions for > me. > > - passes Martin's torture testing > > Which is on top of the other stuff from r8 and NHT, etc. This is great. I'll run our basic ~15 minute regression with valgrind against it and let you know the results. Lou > regards, _______________________________________________ Quagga-dev mailing list [email protected] https://lists.quagga.net/mailman/listinfo/quagga-dev
