[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Bug#596899: Please unblock ia32-libs/20101012



Michael Gilbert <michael.s.gilbert@gmail.com> writes:

> On Tue, 9 Nov 2010 16:21:19 +0100, Julien Cristau wrote:
>> On Tue, Nov  9, 2010 at 15:41:56 +0100, Thijs Kinkhorst wrote:
>> 
>> > As for ia32-libs, I would be willing to sponsor it but I don't think we
>> > should be making uploads for such trivial cleanup operations, is this
>> > really necessary to get ia32-libs unblocked?
>> > 
>> No.  I just didn't want to unblock it until the wine issue was resolved.
>> 
>> I'm still not convinced ia32-libs-dev is a useful/sane thing to ship.
>> Providing a working runtime environment for 32bit programs is one thing.
>> Providing a build environment is another entirely, and the way it has to
>> mangle .la files (and now .pc too) makes me wonder what other sort of
>> brokenness it lets through.  
>
> Why is it that ia32-libs provides all of these 32-bit libs as a
> monolithic package anyway?  Wouldn't the saner solution be to provide
> each desired 32-bit lib from the original source package for that lib
> (for example bzip2 provides lib32bz2, lib32bz2-dev, etc)?  In that case
> ia32-libs is could just be a metapackage, rather than the mess it is
> currently.  Obviously this solution will need to be deferred to wheezy
> (perhaps as a release goal?) since time is short for squeeze.

1) bzip2 compiles a 32bit flavour on amd64. On ia64 it is included in
ia32-libs (lenny) or ia32-lib-core (squeeze). No 32bit compiler on ia64.

2) Providing the same binary package from different source packages on
different architectures is bad. Confuses the BTS and other
things. Providing a lib32bz2 on ia64 not build from bzip2 would be bad.
So it would have to be named something liike ia32-libbz2.

3) Ftp-master (Ganneff) rejected a split of ia32-libs into 54 source
packages some while back. This was so that each source change would only
require that source to be uploaded, preferably by the original
maintainer. Also dependencies between libs would have been tracked
correctly.

4) Ftp-master (Ganneff) removed ia32-apt-get from Debian. Ia32-apt-get
generated the lib32* packages on-the-fly on the users system. It
provided 32bit support for basically every library in Debian (or any
repository) with instant (security) updates and also support for 3rd
party apt repositories with only i386 (e.g. skype) to be directly
installable.

5) We now have several conflicting packages that need 32bit
support. E.g. nss-ldap and jackd. They should have been split out of
ia32-libs already to allow installing the different flavours of them but
that has to wait for post squeeze. Dependencies on them might require
more splits. At some point doing the full split down to actual source
package will be simpler. Someone might have to convince ftp-master to
reverse their decision on 3 or 4.

MfG
        Goswin



Reply to: