buildbot questions

[William Siegrist]

On Aug 2, 2011, at 8:52 AM, Joshua Root wrote:

> On 2011-8-3 01:37 , Marko Käning wrote:
>> Hi,
>> 
>> only a few days ago I learnt about MacPorts' buildbot. And I have a few questions.
>> 

I don't think it has been widely announced since it is still a work in progress. 

>> for ports which FAILED. I guess it would be valuable for a port maintainer to know about this. Only because of this hint I see now that soprano cannot be installed due to a not responding git server… (which might be temporary, or only valid for the buildbot for some reason, but who knows)
>> 
>> Could it be that the buildbot is blacklisted by a.k.o?
>> 
>> 	anongit.kde.org[0: 184.106.129.124]: errno=Connection refused
> 
> The buildslave is behind a restrictive firewall, so it can't fetch from
> svn://, git:// or ftp:// URLs.
> 

The buildbot will eventually email committers and maintainers when something fails.  And for now, the only slave we have is the one behind the firewall, like Josh explains above, so those will be expected failures. 


>> 2) To find the output of a certain port's build procedure one would have to download compile-stdio (which get be fairly large). It would be cool to have the possibility to download the stdio output for the respective ports. But I guess this is all buildbot stuff and can't be so easily modified for MacPorts, right?
> 
> Buildbot is designed to be extended. It would be possible to upload all
> the logs to the build master and make them available. You could also
> just force a build for the port you're interested in by setting a build
> property called 'portlist' with the port name as its value.
> 

Once we are past the initial hump of getting all ports built once, the individual runs will be much more reasonable to view. 


>> 4) I guess. because of point 3 above it could be worthwhile to have a 2nd buildports-buildbot which could faster process changed ports. In parallel the 1st buildports-buildbot would always keep trying to build "all". (But who knows, it looks like "all" regularly fails after 15 to 30 hrs, so that the small chunks of changed ports get their chance after this time span - which might be indeed enough?)
> 
> We only need to get through 'all' once.
> 

What Josh said, plus what I said above. When we're caught up, the runs will be 1-per-commit and thus only touch a small number of ports. 

>> Anyway, It looks like this facility is essential to bring binary port packages to MacPorts. Really cool stuff!
>> 

Yes, plus catch build, checksum, and base failures. 

-Bill