Update/fix the build system for Python 3, remove some old/broken/useless code

[K900]

1. Strongly depend on Cython. Cython.Distutils confirmed present on
   win32/linux2 python2.7/python3.3 with cython 0.18.
2. Remove setup3k.py, setup.py is python2/python3 compatible now.
3. Hack around the patch.py step by defining DL_IMPORT at runtime.
   TODO: find out when exactly DL_IMPORT was removed and update version
   checks accordingly

[shizmob]

I'd just like to say that Cython (and by extension Cython.Distutils) is not installed by default, which is the whole reason for the weak dependency on Cython.

[K900]

I'm pretty sure it is now.
Edit: it is installed if you have distribute before installing Cython.

[shizmob]

Yes, but the point is that @bastienleonard doesn't want to force people to install Cython in the first place.

[K900]

I'm not too sure it's that much of an issue, because Cython can be installed pretty much everywhere now without any issues. Even if Cython should stay a weak dependency, it needs to be done in a different way, e.g. by trying to import Cython.Distutils and only falling back to Cython-less builds if it's not present.

[K900]

Also, shipping pregenerated files with the package might break the build, especially on older Python versions. That's just a theory for now, but there can be API mismatches.

[shizmob]

Oh, I'm not saying I disagree; I'm just stating the reasoning for the weak dependency in the first place.
Personally I'd much prefer just relying on Cython.

[K900]

I'm just stating my reasoning for relying on Cython. Also, it seems the (old) Arch package was already broken because the pregenerated stuff in there was for Python 2.7. Oops.

[K900]

DL_IMPORT(RTYPE)=RTYPE in at least Python 2.7 and up, though it's probably better to find the version where it was removed completely and only override from there up.

[bastienleonard]

I don't provide Cython-generated .cpp files anymore, but I'm going to wait a bit before removing that feature.
Also, I don't really want to use “Leonard” instead of “Léonard”, as pedant as it seems. Is there really no way to do that in a Python-independant way?
Thanks for the macro trick, I think it was one of the biggest annoyances for users.
Sorry but it's too late here to test and push those changes now, I'll have to do that later.

I'm just stating my reasoning for relying on Cython. Also, it seems the (old) Arch package was already broken because the pregenerated stuff in there was for Python 2.7. Oops.

Code generated by Cython should work with Python 2 and Python 3, as far as I know.

[K900]

Well, we can keep “Léonard” with some Unicode escape magic, but I'm not really sure it's worth it, it'll be like 10 lines of code just to replace one character... Also, Cython uses some version-specific hacks, especially for the older ones like 2.4 or 2.5.

[shizmob]

10 lines?

author=u'Bastien Léonard' if sys.version_tuple.major < 3 else 'Bastien Léonard'

[K900]

That won't work, it needs to parse the whole thing.

[shizmob]

% python3 -c "print(u'foo')"
foo

[K900]

Fixed.

[K900]

@shizmob: Python 3.3 has unicode literals, previous 3.x versions don't. http://www.python.org/dev/peps/pep-0414/

[shizmob]

What about 'Bastién Leonard'.decode('utf-8') if sys.version_tuple < 3 else 'Bastién Leonard'?

[K900]

Works on 2.7/3.3, but I'm not sure about some of the earlier versions.

[jstasiak]

I'd say 'Bastién Leonard'.decode('utf-8') if str is bytes else 'Bastién Leonard'

or even

author = 'Bastién Leonardi'
if bytes is str:
    author = author.decode('utf-8')

Works in Python 2.6, 2.7 and 3.3, should work in any 3.x version.

[K900]

Can anyone test with 2.5?

[jstasiak]

It doesn't work in 2.5: http://codepad.org/cNMHLTxU

However 'Bastién Leonard'.decode('utf-8') if int(sys.version_info[0]) < 3 else 'Bastién Leonard' seems to be working: http://codepad.org/YIfIhijN

[K900]

Then I guess the latter is safe to use. Updating now.

[jstasiak]

👍