Release files lists all build architectures rather than architectures on a given mirror

[Wookey]

http://archive.ubuntu.com/ubuntu/dists/maverick/Release contains:
Architectures: amd64 armel i386 ia64 powerpc sparc

and

http://ports.ubuntu.com/dists/maverick/Release contains:
Architectures: amd64 armel i386 ia64 powerpc sparc

Neither of those seem correct (the latter has no i386 packages, the
former no armel ia64 or sparc packages). But perhaps I misunderstand
the meaning of that field and there is a good reason for them to say
what they do?

Currently this problem prevents this file/field being used to determine whether a particular apt source provides support for a particular architecture or not.

[Julian Edwards (julian-edwards)]

This is related to the empty Packages files for ia64 and sparc I think.

[William Grant (wgrant)]

It's actually unrelated to the ia64 and sparc removals. It's because the archive split is done after Release is generated -- the published Release file is for the unified archive on cocoplum

[Julian Edwards (julian-edwards)]

On Monday 27 September 2010 14:25:53 you wrote:
> It's actually unrelated to the ia64 and sparc removals. It's because the
> archive split is done after Release is generated -- the published
> Release file is for the unified archive on cocoplum

What I meant is that the empty Packages files are probably caused by dirty
pockets - the same thing that's causing the Release file to list ia64 and
sparc. But it's just a hunch.

[William Grant (wgrant)]

This bug will still exist after sparc and ia64 are gone. ports.ubuntu.com will list i386 and amd64 while not having their packages.

[Julian Edwards (julian-edwards)]

On Monday 27 September 2010 15:07:02 you wrote:
> This bug will still exist after sparc and ia64 are gone.
> ports.ubuntu.com will list i386 and amd64 while not having their
> packages.

I agree on that part.

[Robert Collins (lifeless)]

So, the splitting code is outside of LP. I think an RT is needed to address this.

[Robert Collins (lifeless)]

wgrant notes that this requires signing on a different machine than does the split. I'm going to mark this low - its a nuisance, yes, but its not abnormal (any partial mirror of a debian archive can have this situation and be unable to resign the releases file).