Server Migration: buildbot

[Ryan Schmidt]

On Sep 17, 2012, at 16:24, Clemens Lang wrote:

> On Mon, Sep 17, 2012 at 04:08:21PM -0500, Ryan Schmidt wrote:
>> And *that* would probably put too much strain on the buildbots. It
>> already sometimes takes them hours to build a certain set of ports.
>> (Whenever I update the php port, for example.)
> 
> I see no other way to have the buildbot auto-detect broken packages to
> rebuild them (which would free us from having to revbump manually, which
> is something we want, right?)

I don't think we ever want to stop revbumping ports. Not everyone has rev-upgrade turned on, so increasing the revision is the only way to ensure people rebuild. Having an automated way to revbump all affected ports would be good however, and rev-upgrade could play a part of that.


> The process could probably be optimized to activate sets of
> non-conflicting packages at the same time (I'm thinking graph coloring).
> 
> The time-consuming part would probably be activating and deactivating,
> since updating the binary database would not be necessary (the
> information is already in the registry on the buildbot) and scanning
> should be rather quick, since only these packages and their dependencies
> are installed. So in conclusion, the whole process would probably be
> I/O-bound.
> 
> The case that rev-upgrade actually finds problems and triggers rebuilds
> would probably be rather uncommon (how many of the changes you commited
> required revbumps?)

It happens occasionally. libpng 1.5 recently. Every time libffi's version increases. I've been putting off updating libgsasl because it'll need revbumps on other ports.