On Sep 7, 2014, at 8:11 PM, Bradley Giesbrecht <[email protected]> wrote:

> 
> On Sep 7, 2014, at 4:56 PM, Thomas G Lockhart <[email protected]> wrote:
> 
>> On Sep 7, 2014, at 2:44 PM, Thomas G Lockhart <[email protected]> 
>> wrote:
>> 
>>> 
>>> On Sep 7, 2014, at 11:21 AM, Bradley Giesbrecht <[email protected]> 
>>> wrote:
>>> 
>>>> 
>>>> On Sep 6, 2014, at 10:44 PM, Thomas G Lockhart <[email protected]> 
>>>> wrote:
>>>> 
>>>>> I’m configuring a new machine (running Mavericks) and notice that some 
>>>>> packages I’m used to building from source are available as pre-built 
>>>>> binaries. That is great. But one package, py27-omniORBpy, wants to put 
>>>>> its python files under /opt/local/lib/ when installed using “port install 
>>>>> py27-omniORBpy”, whereas the prerequisite package omniORB wants to put 
>>>>> its python libraries under /opt/local/Library/
>>>>> 
>>>>> I’d expect both of these to install python code under 
>>>>> /opt/local/Library/, and in fact py27-omniORBpy *does* do that when built 
>>>>> from source using “port install -s py27-omniORBpy”.
>>>>> 
>>>>> Any ideas on what might be causing these differences?
>>>> 
>>>> A none default env var you have set?
>>> 
>>> Ah ha! Not exactly, but...
>>> 
>>> The omniORBpy software uses autoconf and the python discovery mechanisms in 
>>> those m4 macros. And it looks like it falls back to using something based 
>>> on the value of —prefix (/opt/local in our case?) if it does not find a 
>>> python executable. If it does find one it expects to look up the library 
>>> location using:
>>> 
>>> from distutils import sysconfig; sysconfig.get_python_lib()
>>> 
>>> I have my /opt/local/bin/python pointing at python2.7. And autoconf found 
>>> that so my py27-omniORBpy worked just as intended. If I remove the default 
>>> version of python:
>>> 
>>> sudo port select —set python none
>>> 
>>> then I see the same (incorrect) build behavior here. Autoconf can’t figure 
>>> out what to use so falls back to using a library path based on the —prefix 
>>> value.
>>> 
>>> So any suggestions for locking in the value of a PYTHON environment 
>>> variable or the actual python executable within a MacPorts variant? I would 
>>> have guessed that the python PortGroup would have magically handled that, 
>>> but that does not seem to be the case.
>> 
>> OK, I’ve got a fix which involves setting the variable configure.python to 
>> be equal to python.bin.
>> 
>> Bradley, it looks like you have matched this up with a two-year old ticket 
>> which was not on my radar even though I’m the one who opened it and provided 
>> a patch! That previous ticket already has a Portfile.diff which should 
>> probably be superseded by my new one (though the effect is identical, the 
>> new one is a bit cleaner imho).
>> 
>> Should I just replace the Portfile.diff and can we get this one closed out?
> 
> Sure, I'll take a look, do you have the numbers for ticket(s)?

#35818 and I’ve just posted a patch. There has been some activity on this 
ticket (which I missed; my email address has changed) but afaict the patches 
already applied plus my small patch file just posted should be enough to get a 
self-consistent build. btw trac would not let me supersede the Portfile.diff 
which was already there so I just posted an additional one.

Thanks for closing this one out!

                 - Tom

> 
> Regards,
> Bradley Giesbrecht (pixilla)
> 

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

Reply via email to