Python and Debian multi-arch

[Jeroen Demeyer]

I don't know whether this has ben mentioned already on this list, but
also Python has problems with Debian multi-arch. Python doesn't know
that it has to look in /usr/lib/i386-linux-gnu for libraries, so
building the crypt module fails and building some of the hash functions
for hashlib fails.

Setting LD_LIBRARY_PATH and LIBRARY_PATH doesn't help.

Symlinking libcrypt.so and libssl.so into $SAGE_LOCAL/lib does help.

[Julien Puydt]

Le lundi 23 avril, Jeroen Demeyer a écrit:

I already proposed :
http://trac.sagemath.org/sage_trac/ticket/12623
http://trac.sagemath.org/sage_trac/ticket/12624

Solving not-yet-existing problems isn't that pointless...

Snark on #sagemath

[Jeroen Demeyer]

We really have to patch every single Python package? That's awful.

If only distutils/setuptools wouldn't be so broken... why do they
hard-code the directories to look for .so files?

[Julien Puydt]

Le lundi 23 avril, Jeroen Demeyer a écrit:

> We really have to patch every single Python package? That's awful.

No, only python packages which insist on overriding lib paths, because
that means even a patched compiler/interpreter which knows where to
look, will still fail because of the override.

> If only distutils/setuptools wouldn't be so broken... why do they
> hard-code the directories to look for .so files?

[irony mode on] They know where things are supposed to be, so why not be
failsafe and only look there? [irony mode off]

Snark on #sagemath

PS: http://trac.sagemath.org/sage_trac/ticket/12627

[Jeroen Demeyer]

I hate Python's installer and distutils/setuptools.

[Jeroen Demeyer]

On 2012-04-23 14:33, Jeroen Demeyer wrote:
> I hate Python's installer and distutils/setuptools.

... but I hate it less than I hate SCons, which is broken by design, for
example by ignoring environment variables on purpose!

[Julien Puydt]

Le lundi 23 avril, Jeroen Demeyer a écrit:

Uh, like : http://trac.sagemath.org/sage_trac/ticket/12639 ?

Snark on #sagemath

[Volker Braun]

On Monday, April 23, 2012 7:59:11 AM UTC-4, Snark wrote:

No, only python packages which insist on overriding lib paths, because
that means even a patched compiler/interpreter which knows where to
look, will still fail because of the override

Yes thats awful.

It is yet another thing that could be solved once-and-for-all with the compilerwrapper ;-)

[Jeroen Demeyer]

On 2012-04-23 20:06, Volker Braun wrote:
> It is yet another thing that could be solved once-and-for-all with the
> compilerwrapper ;-)

Here, this is not the case. The problem is not with gcc here, the
problem is with Python "detecting" which libraries are installed. No
compilerwrapper can solve this.

[Volker Braun]

You are right, they don't seem to do autotools-style library detection. Thats messed up ;-)