[Pythonmac-SIG] easy_install question

Ronald Oussoren ronaldoussoren at mac.com
Mon Apr 23 15:30:41 CEST 2007


 
On Monday, April 23, 2007, at 02:46PM, "Ulysses Known" <gardenmadman at gmail.com> wrote:
>On 4/23/07, Ronald Oussoren <ronaldoussoren at mac.com> wrote:
>> On 23 Apr, 2007, at 3:44, Ulysses Known wrote:
>> > Hi,
>> >
>> > I just used easy_install to install a package on OSX 10.4.9 (MacBook
>> > Pro) and noted that the built egg has the following name:
>> >
>> > FOO-py2.5-macosx-10.3-fat.egg
>> >
>> > How do I configure easy install to build Intel only binaries for 10.4?
>>
>> Why do you want to do that?
>
>Ronald, thanks for your response.
>
>Well, since easy_install pulled down and built the package on this
>machine (a very slick and easy install indeed!) I see no need to have
>PPC binaries on it, since I would use that same install process on a
>PPC box.
>
>> Anyway, you'll have to change the setup.py to build intel-only
>> binaries. You have to add extra_link_args and and extra_compile_args
>> to all, both of them ['-arch', 'i386'].
>
>So I found setup.py and sections that build native code INSIDE the
>distributed tarball.  I see how this would work, but this makes the
>slick and easy install not so easy.

But unless your disk is almost completely full building a "thin" binary instead of a universal one isn't worth the trouble :-). Even buildtime shouldn't be noticable faster on intel macs as gcc will run the PPC and x86 builds in parallel.


>
>I was wondering more along the lines of distutils.cfg so that I would
>set it once for this system and then have it affect all the installs
>thereafter.
>
>Perhaps this is not possible, however I do see the following egg names
>through google so I wonder if it is possible:
>
>numpy-1.0.2.dev3507-py2.5-macosx-10.4-i386.egg
>
>Thanks for your help.

I'm not sure why numpy doesn't distribute a universal binary egg, that's rather user unfriendly if you ask me. Universal binaries are the best way to package software, users shouldn't have to worry about what CPU type they have in their machine.

Ronald


More information about the Pythonmac-SIG mailing list