This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: [patch] Fix testsuite annotate-quit race (PR 544)
> * I still face testsuite results flutter making the day-to-day regressions
> evaluations expensive. Most of the flutters are fixed in gdb-6.8 but still
> some of them remain, this fix is for one of them.
>
> * I do not care much if it FAILs or PASSes but it must not flutter.
>
> * I found it is a known PR 544 only after I fixed it.
>
> * I fixed it as it was a nuisance reported by the courtesy of Roland McGrath.
>
> * As the SIGINT blocking in RL_REDISPLAY fixes the race (1 of 10 cases to 0 of
> 350 cases during my test) I find it a proof there must be some race.
> In fact if you check that RL_REDISPLAY modifies a lot of global variables
> being tangled by SIGINT-handler-called RL_FREE_LINE_STATE it is clear some
> locking is missing there.
>
> * Sure I am fine to keep the patch in the local branch.
I find it to resolve such matters easier by talking person to person than
communicating by e-mail, but here goes:
You say "I am fine to keep the patch in the local branch." but from the
preceding points it sounds anything but fine. I don't have the understanding
of readline to make a judgement on your patch but I am just worried it will
break something for Emacs just as a previous change that you and Daniel made to
readline did.
It appear to me that the only problem this FAIL is causing is with the actual
running of the testsuite. I would be quite happy if this test was just
removed. Would that work for everyone else?
--
Nick http://www.inet.net.nz/~nickrob