Fedora Extras Development Build Report

Ralf Corsepius rc040203 at freenet.de
Sun May 8 10:04:34 UTC 2005


On Sun, 2005-05-08 at 10:10 +0100, David Woodhouse wrote:
> On Sun, 2005-05-08 at 06:45 +0200, Ralf Corsepius wrote:
> > Are you aware about how much effort had been required to get packages
> > compiled on x86_64, when x86_64 was new? It took years.
> 
> It'll never be that difficult again. We already have something 64-bit
> and something big-endian.
> 
> > Adding any new architecture will impose a similar footprint and will, if
> > a policy of "one arch blocks all" is implemented, this policy will block
> > packages on archs not being affected by the "blocking bug on one arch".
> 
> That result wouldn't be acceptable; you're right. But neither is it what
> I'm suggesting.
Could you elaborate?

It's how I understood your proposal and how Seth seems to have
implemented the FE build system: ATM, one build failure on one of i386,
x86_64 or ppc prevents building a package for the other architectures
and therefore disqualifies a package from being upgraded.

> Of _course_ we should be able to drop a problematic architecture from
> the build to allow new packages to get built -- I'm suggesting that it
> needs to be done _deliberately_ by use of ExcludeArch, rather than
> merely by acting on spurious build failures.
> 
> It needs to be a conscious decision on the part of the package
> maintainer, not random happenstance.
That's reasonable. We should try to provide the same packages for all
supported architectures, but we should not let packages failing for some
architectures block the package for those it does not fail.

In practice, developers will only have access to a limited set of
architectures, in some cases developers do not/do not want to support
some architectures, so ... there will always be "second class citizen
packages" on some architectures ... 

Ralf





More information about the fedora-extras-list mailing list