x86-64 rawhide update obnoxiousness

[Paul Jakma]

On Sun, 16 Oct 2005, Michal Jaegermann wrote:

> I am curious how you would propose to resolve that taking into 
> account all existing packages and this detail that people making 
> original packages may be often unaware of the issue and even if 
> they are they may not have suitable installations for testing. 
> Multiply that by 'extras', external repositories, and so on.  If 
> you will get strict here then resulting pains will surely far 
> exceed your worst current hiccups.

True.

> AFAICT dependencies are really "swallowed" during an installation 
> if files in question are the same - which is easy to test.  At 
> least if the whole concept is not buggy It appears that you were 
> bitten by something else or you run into some traps in testing but 
> that what testing is for.

It was FC3testX I think. If it's now as you describe, then it's sane 
- within the constraints of not wanting to change too much to 
accomodate multi-arch.

> There are obviously open problems in a multilib packages _removal_ 
> but this is not exactly the same thing.

Indeed.

Ideally you'd split off the docs, etc into -common.noarch packages 
(lot of work) or add true multi-arch capability to rpm packages, such 
that a package would mark each file individually with it's 
architecture. Then possibilities open up like:

- x86 (or just x86-64) packages contain both common, i386 and x86_64
   arch files and one could install either one or both of the sets of
   arch-dependent files.

- the x86-64 package contains /just/ the x86-64 files, depends on the
   full package (whose i386 arch files are marked as such) and one
   uses the conglomeration of the two packages to install x86-64 (and
   also i386, if one desires)

Obviously, there's a space cost. But it solves the removal issues, 
and it's how other systems which have supported multi-arch for *far* 
longer than Linux have done it (IRIX), and it worked nicely there. 
(Also makes cross compiling very easy, eg installing x86-64 libs on 
i386 to allow compilation of x86-64 binaries).

But it'd need a lot of reworking of packages. :( Till then the 
"swallow same file" hack is acceptable I guess, if it does indeed now 
check the files are actually the same.

Thanks for clarifying this for me.

>   Michal

regards,
-- 
Paul Jakma	paul at clubi.ie	paul at jakma.org	Key ID: 64A2FF6A
Fortune:
You could live a better life, if you had a better mind and a better body.