Re: [gentoo-project] Re: Council meeting 2015-04-14: call for agenda items

[Rich Freeman <rich0@gentoo.org>]

On Tue, Apr 7, 2015 at 8:01 PM, Matt Turner <mattst88@gentoo.org> wrote:
> On Tue, Apr 7, 2015 at 4:38 PM, Rich Freeman <rich0@gentoo.org> wrote:
>> On Mon, Apr 6, 2015 at 6:22 PM, Matt Turner <mattst88@gentoo.org> wrote:
>>> I've wanted for a long time a system that can help reduce the time I
>>> have to sit in front of a computer waiting for things to finish
>>> compiling before I can test them. Trawl bugzilla, grab package lists,
>>> go build binpkgs. Just let me grab binpkgs and actually test the
>>> software...
>>
>> I approve, and I'm sure the rest of the Council will too.  When will
>> you have it ready?
>
> I don't think it's some pie-in-the-sky fantasy that you can quip "when
> will you have it ready?" about.

It doesn't have to be a fantasy to ask when it will be ready.  In
fact, the question is more meaningful if it can actually be done.

Maintainers are suffering with a problem now.  Unless we really think
a solution is about to emerge, it doesn't make sense to delay.  If you
have reason to think a solution is about to emerge, speak up.

>
> Isn't, in fact, this something like what Zorry is working on? He
> mentioned in 10 days ago in "[gentoo-dev] Cluster tinderbox poc"

This might be promising, but right now all it does is build packages.
Presumably you'll want it to build using the right libraries so that
the binpkg will run on stable, and make the binpkg downloadable, some
wrappers, and so on.  If it is using VMs and not emulated machines,
then you need to have multiple instances of this thing for each target
arch.  You'll want some kind of bugzilla tie-in if you want the thing
to have already built the stablereq package when you first go to look
at it, otherwise you're going to have to queue up a build, and if
you're going to do that, why not just build it on your own box?

Don't get me wrong - I think it is a good idea, and tinderboxing is
great for a lot of things.  I just don't know that it is really a
solution to the problem of "I don't want to build stablereq packages
myself" and even if it were it seems like this months away from being
a solution for something like sparc (anybody have a 64-core sparc
sitting around?).

>
> I'm lamenting the fact that ~30 emails in no one has even appeared to
> wonder out loud if there's actually a technical solution to the
> problem. Instead we're debating the merits of unkeywording whole
> subtrees.
>

Tinderboxes are useful for build tests, or for running automated
tests.  When I'm arch testing the build time is usually the least of
my concerns.  I can just open screen and launch emerge -1 pkg in each
of a bunch of windows, then in a bit go and test each one.  It is the
testing that really takes the time.  If you're just going to compile
test then the tinderbox is fine, but then you might as well bypass the
arch team entirely.  The one place where I wouldn't mind having binary
packages handy would be for stuff like chromium or libreoffice, or
maybe for testing a whole gnome release or something.

And if we're going to be building binary packages, we might as well
mirror them and let the users make use of them if they're sticking to
the targeted profile.

It just isn't a trivial problem.  I think it is great that you're
bringing it up, but you don't need to criticize others for not having
proposed it first.

-- 
Rich