Default +universal variant for configure-based ports

[Ryan Schmidt]

On Feb 26, 2007, at 02:57, Paul Guyot wrote:

> You are right. This will not work on 10.3 simply because 10.3  
> installations are not capable of building universal binaries. I  
> have just added a warning (we could transform it into an error)  
> when the +universal variant is selected on machines where the  
> Universal SDK is not installed. This will cover machines running  
> 10.3 and incomplete 10.4 installs.

What is the difference between a warning and an error? Are you saying  
it just prints a message, then tries to continue? How could it  
succeed if the universal SDK is not present?

> I realize that this code should be modified when Leopard will be  
> out, for example, we should not specify 10.4 universal SDK on  
> Leopard if all the libraries there are universal. But I can only  
> speculate for now.

I do not believe that any such changes would need to be made when  
Leopard is out. Leopard should be able to run 10.4 universal  
binaries. And we should continue to make 10.4 universal binaries even  
when on Leopard, so that any universal binaries that were made there  
will still run on 10.4. Are we setting MACOSX_DEPLOYMENT_TARGET=10.4?  
If not, we should be.

> I am not aware of your exact needs, but please note that MacPorts  
> does not produce 10.3.9 PPC + 10.4 Intel universal binaries.

And I don't think we should try to support 10.3.9 PPC + 10.4 Intel  
universal binaries; it's too much of a pain. Unless we were to  
include a trick like the unify script from the mozilla project in the  
macports infrastructure somewhere. But even then, many software  
packages throw considerably more fits when being cross-compiled than  
when they are being universally compiled. (glib2: I'm lookin' at you...)