launchpad-dev team mailing list archive

[Michael Hudson <michael.hudson@xxxxxxxxxxxxx>]

Julian Edwards wrote:
> Michael Hudson wrote:
>> Hm.  I was thinking that it would only build the source package, and
>> then the source package would get built by something like the existing
>> mechanism.  It doesn't have to be done this way, I guess, but it seems
>> sanest and easiest to me.
> 
> This is exactly what should be happening, yes.  The buildd-manager will
> dump the source package in the "incoming" directory from where the
> process-upload script will pick it up.

Great.

>> This is tangent, but how will scoring work?  I'm guessing the value
>> returned by IBuildFarmJob['score'] is stashed in the BuildQueue and then
>> it's the value in the BuildQueue that's examined by findBuildCandidates?
> 
> Noodles is going to sort that out next week.  But basically we need the
> same sort of scoring scheme across all job types so that none is
> favoured over others unless it's deliberately rescored.

I guess as a first approximation we give the recipe job the same score
as a i386 build of the resulting package would have?

There's something lurking in my brain about here to do with the fact
that we don't actually know for sure what source package a recipe will
build until it's done it.  I don't know what we should do in case where
it builds something unexpected -- bitching and moaning seems reasonable,
as we'll only want to allow people to trigger recipe builds if they have
permissions to upload the implied source package to the specified archive?

>> I guess it would help if I remembered how signing of binary packages
>> built by the current system worked...
> 
> They are not signed :)

So I know now!  I guess that makes it harder to reuse the existing
infrastructure...

Cheers,
mwh