Buiildbots - Framework not found CoreMidi

[Ryan Schmidt]

On Jan 18, 2021, at 21:23, Craig Treleaven wrote:

> On Jan 18, 2021, at 10:01 PM, Ryan Schmidt wrote:
> 
>> On Jan 18, 2021, at 08:35, Craig Treleaven wrote:
>> 
>>>> The correct spelling is CoreMIDI, not CoreMidi. You get away with it on a case-insensitive filesystem.
>>> 
>>> Thanks.  I’ve pushed a commit that I hope fixes the CoreMIDI issue.  Although the buildbots don’t seem to be picking it up.
>> 
>> Most buildbot workers are still busy building the hundreds of ports that needed to be rebuilt after we switched from jpeg to libjpeg-turbo. They'll get to subsequent commits when they're done with that. You can observe buildbot activity at:
>> 
>> https://build.macports.org/waterfall
>> 
> Thanks.  What confused me is that I checked the waterfall before I pushed my change.  At that time, all the builders were idle and I thought I’d get results very quickly.  It took many minutes (60-90?) for the watchers to start issuing the jobs on the builders.  Apparently my push must have been *just* after the monster commit.  It was quite a bit later before I realized my place in line.  Oh well.  Looks to me like it will be another 24 hours or so to process all the libjpeg-turbo changes.  

Your commit was 5 hours after Josh's. I wasn't looking at the buildbot when your commit came in, but it's certainly possible that the "portbuilder" builders were idle, and if you had limited your waterfall to just the "portbuilder" tag it may have appeared that everything was idle. "portbuilders" don't get triggered until after their parent "portwatcher" has finished enumerating which ports need to be built and after the "mirror" task for those ports finishes. Enumerating hundreds of ports takes several minutes, while mirroring hundreds of ports can take hours. For example, the mirroring task triggered by the macOS 11 Apple Silicon portwatcher took almost 4 hours:

https://build.macports.org/builders/jobs-mirror/builds/236888

We run the mirroring task again for each portwatcher, i.e. for each build platform. This is inefficient and we want to improve it but we put it off until after we update to buildbot 2. See https://trac.macports.org/ticket/59286

We do maintain a cache of what has already been mirrored, so the subsequent jobs are quicker. The mirroring task triggered by the macOS 10.15 portwatcher took "only" 75 minutes since a lot of work could be skipped since it had already been done by the previous task:

https://build.macports.org/builders/jobs-mirror/builds/236897