[csw-maintainers] Buildbot

[Peter FELECAN]

"Maciej (Matchek) Blizinski" <maciej at opencsw.org> writes:

> On Fri, Jul 31, 2009 at 6:49 PM, Peter FELECAN<pfelecan at opencsw.org> wrote:
>> "Maciej (Matchek) Blizinski" <maciej at opencsw.org> writes:
>>
>>> [...]
>>>
>>> I've split the thread about the build systems into another one, and
>>> I'd like to go back to the buildbot topic. Are there any comments
>>> about the design or deployment plan?
>>
>> Split or not split, my comments are valid for this
>> discussion. Consequently, please take into account this design
>> requirements:
>>
>> 1. The buildbot deployment must be done on a separate hardware from that
>>   used for the regular, "manual", builds.
>
> This requirement might be a showstopper here. Can we trade it for a
> restricted set of packages to build, avoiding the large ones?

It *is* a showstopper if the infrastructure does not follow. If a
quality assurance process impact negatively the controlled process we are
in the presence of a counterproductive action.

The interest of a separate test environment resides in the capacity to
fully build it: you build the root of the graph dependency, install it,
you build the direct dependants of the previous package and you install
them, and you repeat this until the exhaustion of the dependency
graph. That will test more than the building of a package, it will also
validate the result as an installable package and that it can be used in
building the other packages and, if we push to the extreme, that its
behaviour is nominal. That could be an interesting process in helping to
deliver stable releases. But I don't think that we, as a group, have the
energy to implement something like that.

-- 
Peter