_cmp should try _richcmp_ if _cmp_ failed

[jdemeyer]

With the current comparison implementation, defining __cmp__ in a Cython class can actually break cmp() if _cmp_ raises NotImplementedError. This is because defining __cmp__ causes Python to use __cmp__ and only __cmp__. Instead, __cmp__ (so really _cmp in Cython) should itself try _richcmp_ if _cmp_ didn't work.

To avoid an infinite loop, it's important that only _cmp calls _richcmp_, we should not change _cmp_ here.

[jdemeyer]

Last 10 new commits:

​1ad339b	Implement _rich_to_bool as inline function instead of member function
​17bd067	Merge tag '6.7.beta2' into t/17890/ticket/17890
​313a400	Optimize rich_to_bool_sgn
​629f6a5	Improve comparisons for permutation groups
​0d1e049	Improve _richcmp and documentation
​39273f1	Fix doctest formatting
​04570b3	Fix bad doctest in etaproducts
​3976f2c	Add pointers for special uses of __richcmp__
​1de420c	Merge commit '3976f2cc4ec59c9f978d9d02c7f697c81025c4ef' into t/18322/_cmp_should_try__richcmp__if__cmp__failed
​c1d1881	_cmp should try _richcmp_ if _cmp_ failed

[git]

Branch pushed to git repo; I updated commit sha1. This was a forced push. New commits:

​c70bd39

_cmp should try _richcmp_ if _cmp_ failed

[git]

Branch pushed to git repo; I updated commit sha1. This was a forced push. New commits:

​621bd10

_cmp should try _richcmp_ if _cmp_ failed

[git]

Branch pushed to git repo; I updated commit sha1. This was a forced push. New commits:

​cb07254

_cmp should try _richcmp_ if _cmp_ failed

[git]

Branch pushed to git repo; I updated commit sha1. This was a forced push. New commits:

​bb98fd0

_cmp should try _richcmp_ if _cmp_ failed

[vdelecroix]

1. The code of _cmp_ raises a NotImplementedError which has a custom error message which involves string formatting. As this error is catched in a critical part of the code (i.e. _cmp) I would either:
   • remove the message "comparison not implemented for %r"%type(left)
   • or use lazy strings

The second option looks better to me.

2. It would be nice to update this comment:

       ####################################################################
       # For a derived Cython class, you **must** put the __richcmp__
       ...
       # arguments have identical parents.
       ####################################################################
   

   with something like: implementing `_richcmp_` will automatically makes `cmp` works, see the implementation of Element._cmp.

3. Actually, the current implementation let met think that for elements we should:
   • implement _richcmp_ for more speed (e.g. == is cheap but < is not) or if _cmp_ does not make sense (e.g. we want to test for equality but raise errors on <)
   • implement _cmp_ otherwise

Am I right?

Vincent

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​d826520

Use LazyFormat for _cmp_ exception

[jdemeyer]

Replying to vdelecroix:

3. Actually, the current implementation let met think that for elements we should:
   • implement _richcmp_ for more speed (e.g. == is cheap but < is not) or if _cmp_ does not make sense (e.g. we want to test for equality but raise errors on <)
   • implement _cmp_ otherwise

I think so, yes.

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​ac9ca1c

Update comment