http://bugs.winehq.org/show_bug.cgi?id=14851
--- Comment #11 from Joe Mattiello mail@joemattiello.com 2009-03-31 15:27:23 --- (In reply to comment #10)
(In reply to comment #9)
A reverse regression test between 0.9.37 (crash on load) and 0.9.38 (no crash, but no connect) for the 'crash on open' bug yielded:
========================================== dd8a736c3fc8ebcbbce1d4c548ac0e57d5c3feff is first bad commit commit dd8a736c3fc8ebcbbce1d4c548ac0e57d5c3feff Author: Rob Shearman rob@codeweavers.com Date: Mon May 21 16:45:17 2007 +0100
ole32: The identity functions can be used from apartments other than theone the proxy was unmarshaled into.
In this case, don't use the cached This->remunk and don't cache the unmarshaled object in This->remunk.:040000 040000 5220efd5cf61185d34f51c92fb59de3f8ba431de bba76338655538612a9f2aad75635a66a3fbdeb6 M dlls ==========================================
git checkout dd8a736c3fc8ebcbbce1d4c548ac0e57d5c3feff
This yielded a build that would still crash on open.
Please advise.
Run two bisects, one to find the crash fix patch, one to find the no connect regression.
Once you've identified the crash fix patch, apply it each time before testing for no connect.
I thought I was doing the former. The crash was fixed between 0.9.37-38. So I bisected and marked 38 as bad and 37 as good and rebuilt. Doing the normal, bisect, build, compile pipeline from there, except marking a crash as 'good' and successful open as 'bad'.
This is what brought me to dd8a736c3fc8ebcbbce1d4c548ac0e57d5c3feff. This "should" be the patch that fixed it, or so I thought, alas checking out that version still crashes.
I'm going to re-read the reverse regression article to see if I skipped something. It could be that I am just not familiar with git and may have a contaminated tree between tests.