[09:09] <jbailey> doko: Around?
[09:31] <doko> jbailey: yes
[09:32] <jbailey> doko: Heya!  Is there any sort of regression checker in the gcc testsuite?
[09:33] <doko> no, there was one running upstream for checkins; usually I inspect the logs manually. elmo wrote something for binutils, which now runs in the build as well (comparing against the installed version).
[09:33] <jbailey> Hmm.
[09:33] <jbailey> I was thinking that the simplest one would be no FAILs
[09:33] <jbailey> And just mark everything that currently FAILs as XFAIL and move on.
[09:34] <jbailey> IIRC, dejagnu can give a different return code for that.
[09:34] <doko> It's difficult to automatically tell about regressions, new failing testcases show up as false positives
[09:38] <jbailey> Sure, I'm just thinking at the release point for gcc-4.1 and such.
[09:38] <jbailey> I don't think new test cases are usually added to the stable branch.
[09:41] <doko> they are.
[09:41] <jbailey> Ah, hmm.
[09:42] <jbailey> That could suck.
[09:43] <jbailey> In this case, the hppa patch is totally localised to the machine descriptor, but in general, I'm happier if I know that I'm not going to break something elsewhere.
[09:43] <doko> you would have to save the .log files instead of the .sum files for all tests, then you can detect added testcases as well
[09:46] <jbailey> Well.  I think I was hoping more that sometime in stage 3, we just submit a list of current fails, mark them as XFAIL and move on. =)