[Angstrom-devel] Maintenance of unstable feed

[Paul Sokolovsky]

Hello Rod,

Saturday, January 5, 2008, 2:40:32 PM, you wrote:

> Paul Sokolovsky wrote:
>>    But if we don't invite people to cooperate on one well-maintained
>> repo, we push them to setup their own feeds and organize feed
>> graveyards like http://handhelds.org/moin/moin.cgi/IpkgFeeds

> I must disagree with this statement.  As a real counter-example, the
> NSLU2-Linux project has tens of thousands of users, and they *all* use
> packages from official feeds built by a single autobuilder.  There are
> no unofficial feeds that we know of, and there are no binary uploads, by
> anyone, ever.

  That's what we're looking for for Angstrom too, asymptotically.
Except that you don't say here how you solve problem of testing,
especially, community testing (and that's 80-90% of the entire
testing an OpenSource project can afford).

  But later, you say...

> Yes, our policy for slugos-packages is that if it compiles then it can
> be enabled for the feed.  Since the autobuilder works directly from the
> latest OE head, then our developers have direct control over what the
> autobuilder builds, with no middle-man involved.

...that you don't have any testing at all, except for purely formal
buildability one! The way it likely works for NSLU2 is, like Koen
points, that the package base is by now well saturated and tested. I'm not
sure that Angstrom wants such procedure at all, and its package base
is for sure different. While it's mostly easy for a console app to
"just work" after build, it's not so for GUI apps and resource-limited
devices. So, we don't want to throw in 100 apps which build, but 70 of
which don't run well due to memory required and screen dimensions.

  Hence, these talks about unstable feed as a kind of community
boil-test labs. And we could setup autobuilder for unstable too, but:
1) we have other places to improve automation first; 2) the whole
matter is to get humans involved, not robots. Humans will test it
after all, and that's the hardest part. So, if few humans put their
efforts into initial discovery of nice package, testing build on one
system, then on another, then uploading package to common unstable
feed, then announcing it, - hopefully then they will have motivation
to test it, or at least keep and eye on its future fate.

  Of course, that not necessarily will work - even for that some
initial critical mass of people is required. But I really have no idea
how to try to bootstrap it otherwise - because again, we first of all
need people, working in comfort on the common goal beneficial to all
of them.
  

[]


-- 
Best regards,
 Paul                            mailto:pmiscml at gmail.com