This is the mail archive of the gdb-patches@sourceware.org mailing list for the GDB project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [RFA/testsuite] Reduce the default test timeout to 30 seconds for Unix.


> > gdb/testsuite/ChangeLog:
> > 
> >         * config/unix.exp (timeout): Change value to 30.
> > 
> > Tested on x86_64-linux.  No regression.
> 
> I think that for this to be meaningful, you should really test this on
> a slower platform though.  Something like a native linux config on arm
> should do the trick.

I agree.  The problem is that I had to test on at least one platform,
and I don't have access to that many machines that are very noticeably
slow, particularly the system that you suggest.  I could tell you about
one of our old x86-solaris machines, but I am under surveillance and
will get killed if I even attempt the gdb-testsuite there.

I think the slowest I have is ppc-aix, but even that I think is way too
fast to trigger the timeout (I consider that, for the programs in our
testsuite, anything other than an "immediate" response is a bug).

All in all, I think the fairest test would be to try with a configuration
where communication with the target is really slow, something over the
serial line.  But we don't have any such configuration at AdaCore.

That being said, it's a shame to potentially penalize almost everyone
just because some systems might be too slow for the default 10 seconds
timeout. I thought that 30 secs was a good compromise for today's
equipment (meaning today's outdated equipment).  If there are still
some such systems out there where 30secs is not enough, they can very
easily change the timeout duration back.

It's fine if we decide to be conservative and not apply this patch
without more testing.  But I think we should just give it a try, and see
what happens.  We have very little to lose. It's really easy to further
adjust the timeout if some of the developers notice that 30 secs is
too small, or even revert.  But I suspect/hope that this will only ever
happen with specific non-unix configuration, where they already use a
board file that they'll instantly update to get a longer timeout.

-- 
Joel


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]