[Distutils] GCC versions and binary egg names

P.J. Eby pje at telecommunity.com
Mon Jul 26 19:42:21 CEST 2010

At 06:04 PM 7/26/2010 +0100, Chris Withers wrote:
>Hi All,
>In addition to the UCS2/4 problems already described by MAL and 
>which I've bumped into myself, I now have a problem with binary 
>linux eggs where the GCC version doesn't match that of the system 
>the egg is being installed on.
>Current egg metadata doesn't take either of these into account or 
>even provide a way for the nature of the binary egg to be recorded.
>How're people coping with this?
>What are the future plans in this area?

Egg binaries aren't a great idea on Linux; they were mainly intended 
for Windows and Mac, where the architectures are uniform and working 
compilers aren't just an apt-get or yum away.

(Ironically, this is a side effect of eggs being invented *before* 
easy_install; if I'd thought of easy_install *first*, I might not've 
bothered with eggs (as a distribution format) at all.  Certainly, 
sdists don't add a lot of installation overhead except on 
compiler-less platforms.)

Incidentally, if you look way, way back in the distutils-sig 
archives, you can see where I first raised the question of platform 
strings and addressing some of these issues, but at the time nobody 
was interested.  Since then, the topic gets raised periodically, but 
there has never been anyone with enough of both interest and 
knowledge to put together a concrete proposal for overhauling 
platform strings on *nix platforms.

OS/X and Windows, OTOH, have had proposals implemented either in 
distutils or setuptools over the last few years.  (Distutils needs 
code to *distinguish* platforms (by changing the strings), but 
setuptools only needs special code to be able to tell when the 
running platform supports running something built with a different 
platform string.  So, additions of UCS build info to distutils would 
probably not affect setuptools; addition of gcc info might.)

More information about the Distutils-SIG mailing list