64bit variant naming scheme

[Ryan Schmidt]

On Dec 8, 2008, at 15:29, Bryan Blackburn wrote:

> So if someone wants to have 64bit support from a port, they'll need  
> to build
> it +universal?  This would have to require people adding the requisite
> setting to universal_archs in macports.conf as well right, since  
> trunk still
> specifies only 'ppc i386'?
>
> Also, what about ports where building 64bit is easier than  
> universal, if
> there are such ports?

To chime in: I also don't like having a variant for 64-bit support  
and also think it should be part of the universal mechanism. But I  
also don't like how setting the architectures you want in  
macports.conf is the only way to influence what archs a port builds for.

I would like each port to be able to declare which architectures it  
can be built for in universal mode. If a port does not specify, it is  
assumed it can be built for all architectures. This way, a port can  
specify that it only works on, say, i386 and ppc (if 64-bit doesn't  
work), or that it only works on i386 and x86_64 (for ports like wine  
that are Intel-only). Ports that are arch-agnostic would be able to  
indicate that in some way too. All this would replace  
"universal_variant no".

This would be in addition to the setting in macports.conf. For  
example, a user requests in macports.conf that they prefer all 4  
archs, and then individual ports can decide to only build a subset of  
those.

MacPorts should also record which architectures a port was built for,  
not just that it was built "universal". Otherwise it's painful to  
upgrade from, say, a 2-arch universal to a 4-arch universal build of  
a port.

MacPorts needs to be able to ensure that if you try to build port X  
universal, and it depends on port Y, then the set of architectures  
port X gets built for can be at most the set of architectures port Y  
was built for.

I'm sure there was a previous discussion on the topic where I said  
much the same thing, but I haven't tried to find it in the archives.