On Mar 11, 2010, at 01:13, vincent habchi wrote:

>> I don't believe any attempt was ever made to build two separate versions and 
>> lipo them together e.g. using the muniversal portgroup, though that would 
>> indeed be the required strategy for atlas, since it uses a non-Apple version 
>> of GCC which cannot build for multiple architectures simultaneously. You're 
>> welcome to see if switching atlas to the muniversal portgroup fixes the 
>> issue.
> 
> I faintly remember having tried that a while ago, but it fails. There are 
> some issues at stake, I'll dig deeper if you want.

You can pursue this however you like, but I recommend trying to use the 
muniversal portgroup rather than inventing yet another new universal build 
method.

>> I also don't know whether any attempt has yet been made to verify whether 
>> atlas will in fact build for the non-default build_arch. I assume it builds 
>> 32-bit on a GCC that defaults to 32-bit (e.g. Leopard), and 64-bit on a GCC 
>> that defaults to 64-bit (e.g. Snow Leopard), but I don't know what happens 
>> if the user changes build_arch.
> 
> First, one must alter the gcc-4.4 or gcc-4.3 portfile to allow building with 
> the --multilib option, that generates libraries both for 32 and 64bits. I've 
> done that, and it works perfectly, so I don't really know why this option was 
> disabled (as it seems to have been removed at some point).

I don't know anything about it. 

I see you filed a ticket 5 months ago and got no response from the maintainer. 
Perhaps ask him again.

http://trac.macports.org/ticket/22116

Multilib is also mentioned here.

http://trac.macports.org/ticket/22148

> Then, I vaguely think that atlas should be build somehow as a metaport, 
> calling twice the Portfile with a different bitsize, then lipo'ing everything 
> together via an external script.

We don't have any other port that builds universal that way; I'm not sure how 
you would even do that. (Portfiles are not in control of "calling" other 
portfiles.) I would recommend trying to make the muniversal portgroup work, 
since that is the existing build-and-lipo method we are using successfully in 
other ports.


_______________________________________________
macports-dev mailing list
[email protected]
http://lists.macosforge.org/mailman/listinfo.cgi/macports-dev

Reply via email to