universal_variant no

[Ryan Schmidt]

On Jun 21, 2011, at 17:41, Bradley Giesbrecht wrote:

> On Jun 21, 2011, at 1:57 PM, Daniel J. Luke wrote:
> 
>> On Jun 21, 2011, at 2:23 PM, Joshua Root wrote:
>>> 
>>> On 2011-6-22 04:01 , Bradley Giesbrecht wrote:
>>>> opendkim depends on libmilter which does not have a universal variant.
>>>> 
>>>> Does this alone mean that I must/should add "universal_variant   no" to my opendkim Portfile?
>>> 
>>> Depends on how the dependency is used. If it's an architecture-specific
>>> library that is linked against, which seems likely from the name in this
>>> case, then yes, you should also disable universal in the dependent.
>>> Otherwise users will just get an arch mismatch error when they try it.
>> 
>> it's probably linking with libmilter, so this is correct.
>> 
>>> If the dependency just provided a command which was executed by the
>>> dependent in a separate process, then the archs wouldn't need to match
>>> and you could put it in depends_skip_archcheck and leave universal enabled.
>> 
>> the best thing would probably be to provide a patch to enable universal for libmilter :)
> 
> Ya, I spent what little time I had this morning working toward that end but didn't get it done. Maybe this evening.
> 
> libmilter (sendmail) doesn't use configure and I haven't found an effective env var.
> 
> Would PortGroup muniversal be useful here? My muniveral knowledge is limited to the couple lines I read this morning.

muniversal is great for builds that respect the env vars MacPorts sets during the configure phase, but don't like building for multiple archs at once. But if libmilter doesn't use a configure script then that's going to be more difficult.