[Debian-med-packaging] migration exception for mhap

[Emilio Pozuelo Monfort]

Hi,

On 25/07/16 11:13, Jonathan Wiltshire wrote:
> On 2016-07-24 23:11, Afif Elghraoui wrote:
>> not that it's supposed to be usable on
>> every single one.
> 
> It still has to be installable.
> 
>> Barring porting libssw to i386 or reducing functionality of the package
>> to remove dependency on libssw, would you prefer that I declare mhap as
>> arch:any so that it only builds for architectures where it will be
>> installable? Then we'll have multiple copies of the exact same package
>> for amd64, kfreebsd-amd64, and x32.
> 
> That doesn't help you at all, because it's still uninstallable on anything that
> is not amd64.
> 
>>> For performance reasons britney only tests installability on amd64 and
>>> i386 (hence the message), otherwise the list would be much longer.
>>>
>>> A package cannot migrate if it is not installable on the test
>>> architectures.
>>>
>>
>> For the purposes of mhap, it is a package for scientific research and
>> would probably not be usable on i386 even if it could be installed
>> there. It requires more powerful processors than anything that is i386
>> that I am aware of (besides am64 CPUs posing as such, but the package
>> works on x32 anyway).
> 
> Maybe you want "arch:any-amd64 x32" then?

I'm not sure about that. mhap is arch:all. It just happens to be uninstallable
on some architectures because one of its dependencies isn't available
everywhere. Whenever that dependency gets support for those architectures, then
mhap will be installable.

This isn't different to sspace, circlator, pbalign, console-setup-freebsd,
python-pbcore, python-pbgenomicconsensus... to name a few.

Maybe we should change our policy here or fix some stuff (including how britney
handles arch:all packages), but we should carefully think about it and then be
consistent about it.

Cheers,
Emilio