For your consideration: Secondary Architectures in Fedora

[David Woodhouse]

On Thu, 2007-06-14 at 07:37 -0400, Daniel Veillard wrote:
> Right, sorry. I will go down in my hole to avoid this in any possible future.
> I still think what you are suggesting in the remaining of your mail does not
> match what I suggested, bahh, no big deal.

You're right; it doesn't. Your suggestion (mostly) matches Spot's
original proposal, which is at
http://fedoraproject.org/wiki/TomCallaway/SecondaryArchitectures

My proposal is slightly different -- I believe that Spot's original
proposal is flawed, because if we make it so easy for the secondary
architectures to end up with old or missing packages, we might as well
not bother helping them at all. People have showed that they're
_already_ capable of running entirely asynchronously and building Fedora
for SPARC/Alpha/etc without any help from us at all. It's horrible, and
you're constantly fighting to keep stuff in sync with 'Fedora', but it
can be done. What Spot proposes really wouldn't be much better for them.

I think that people are overestimating the 'burden' on package
maintainers if we do synchronous builds. Note that I'm not talking about
how beehive used to do it, where you have to resubmit the build if it
fails. If the build fails on S390, you'd have the _option_ of declaring
that it's really an arch-specific problem, and you'd be able to ship the
already-completed packages anyway for the other architectures. You
wouldn't even have to wait for it to rebuild. But you _would_ be
expected to at least _glance_ at the failure and make a proper decision
about the failure, rather than shipping the partially-failed build
automatically. And while we wouldn't be _forcing_ package maintainers to
do a proper job of maintaining portability, it would at least help to
indicate expectations.

I also believe that people are underestimating the amount of time that a
build failure on one arch actually shows up a _generic_ problem which
just happens to bite in some situations but not others. Sudden build
failures on an arch which _used_ to build are something which the
package maintainer really _should_ be looking at.

-- 
dwmw2