Tom Rockwell wrote:

> Perret Yannick wrote:
>
>> Perret Yannick wrote:
>>
>>> (...)
>>> For the second point, the only way should be to use 'rpm' to handle the
>>> i386 RPMs in the same way than the x64_64 ones.
>>> I think I will have a look to the current 'rpm' sources to see if this
>>> can
>>> be integrated...
>>>
>> Many many problems with that...
>>
>> First of all i386 and x86_64 RPMs conflict if they have at least a
>> common
>> file (i.e. a binary). So in order to install i386 and then x86_64 you
>> should
>> have to use '--force' for the last one.
>
>
> Actually, there is some logic in rpm to eliminate conflicts, seems to
> work reliably for binaries at least.  Do you have an rpm that this fails
> for on binaries? I can try it on my system.

I have to check... In fact I'm now waiting for the machine to be out of
the production farm so that I can play again with it (as I only have one
x86_64 machine installed).
If I remember well, most "library" packages work fine, and conflicts
really appears with packages holding other kind of files.

>
>>
>> And in any cases updates/erases failed to work fine because the two RPMs
>> share the same name, some files are shared and so one of the RPMs failed
>> to find the files, and uninstall scripts should not appreciate that
>> someone
>> remove some files previously.
>
>
> Given that there are some issues with installs, updates/erases are
> likely even less reliable...
>
I also find some other problems (not critical, but stupid): while trying
to install a i386 RPM, 'rpm' find that an other RPM is missing... and
suggest to install the x86_64 version of the missing RPM (instead of
the i386 one which is missing). And of course the x86_64 one is
still installed...

> Scripts seem even more problematic at the moment, due to all the odd
> things that rpm authors might do.
>
Yes.

>>
>> So a kind of external action should be used to manage properly the dual
>> installation... It's a pitty.
>
>
> It seems that the compatiblity system is reasonable; problems with
> individual rpm packages should be reported back to the packager...
>
It is the correct way, of course. But until all the packages are updated,
I would like to find a way (a not-too-ugly way) to make the full i386
compatibility (the full is important, as these machines are 64 bits but
should act exactly the same way than our i386 machines because
jobs running are exactly the sames).


--
Yannick