llvm dev - Nov 2015 - Fwd: buildbot failure in LLVM on sanitizer-ppc64-linux1

On 10/29/2015 07:40 AM, Bill Seurer wrote:
> On 10/28/15 23:47, Philip Reames via llvm-dev wrote:
> > This long running bot failed on a unused variable warning. Given that
> > several other bots cover the warnings, any chance we could get this
one
> > configured to not fail the build on warnings?  Doing so would make it
> > more likely to actually get to the "sanitizer" part of the
testing
> process.
> >
> > Philip
>
> Currently the option for Werror is not configurable and is hard coded 
> into the cmake invocation.
This seems less than ideal.  The other option would be to do a staged 
build (i.e. have slower builders grab the most recent binary from a 
faster builder so that they're only running the "interesting"
parts on
an otherwise clean build.)
>
> The bot does a run in 30-40 minutes so I am not sure why you think 
> that is "long running". The similar X86 bot takes about twice as
long.
I didn't mean to imply that this was one of the longest running bots we 
had.  Sorry if it came across that way.

However, a 30-40 minute build cycle is fairly long.  In practice, I tend 
to devote the first couple of minutes after a patch submission to 
watching for problems (while checking email say), but after 20-30 
minutes without problems, I've moved on to something else and might not 
notice a problem report.  I'd see it eventually, but that might be 
several hours later or even the next day.  Anything we can do to reduce 
cycle time increases the odds that problems will get seen and addressed 
quickly.

Philip

On 11/02/15 20:04, Philip Reames wrote:
> However, a 30-40 minute build cycle is fairly long.  In practice, I tend
> to devote the first couple of minutes after a patch submission to
> watching for problems (while checking email say), but after 20-30
> minutes without problems, I've moved on to something else and might not
> notice a problem report.  I'd see it eventually, but that might be
> several hours later or even the next day.  Anything we can do to reduce
> cycle time increases the odds that problems will get seen and addressed
> quickly.
I have been tweaking the buildbot settings and moving bots between 
systems and in the past month I've reduced the run times by 50-75% for 
most of the power bots.  Unfortunately now I just need more hardware and 
that's just not going to happen in the near term.
-- 

-Bill Seurer

On 11/03/2015 08:14 AM, Bill Seurer wrote:
> On 11/02/15 20:04, Philip Reames wrote:
>> However, a 30-40 minute build cycle is fairly long.  In practice, I
tend
>> to devote the first couple of minutes after a patch submission to
>> watching for problems (while checking email say), but after 20-30
>> minutes without problems, I've moved on to something else and might
not
>> notice a problem report.  I'd see it eventually, but that might be
>> several hours later or even the next day.  Anything we can do to reduce
>> cycle time increases the odds that problems will get seen and addressed
>> quickly.
>
> I have been tweaking the buildbot settings and moving bots between 
> systems and in the past month I've reduced the run times by 50-75% for 
> most of the power bots.  Unfortunately now I just need more hardware 
> and that's just not going to happen in the near term.
Thank you for making the effort.  It's much appreciated.

Philip