Implement PEP560 inheritance (__mro_entries__) #4005
Conversation
Fixes cython#3537 Both the C code and the tests are largely copied from CPython so hopefully should be pretty reliable. This doesn't backport perfectly so I've hard to disable a few tests of Python 2 (and may have to on earlier Python 3 versions too). I think that's OK. I haven't applied this at all to cdef classes. Maybe it could be applied to the second+subsequent bases but I don't think it's needed for the initial implementation.
Changed
The failures are the same as master (unless I broke it with the current lot of changes...)
#4006 clears up some simple bugs one of the tests for in 2.7 and 3.4/3.5 I think. The C++ tests are largely failing because something changed in Pythran (but I don't think there's a good fix for that yet...) But I think there's a few other things too |
Cython/Compiler/Nodes.py
Outdated
| def analyse_expressions(self, env): | ||
| if self.bases and not (self.bases.is_sequence_constructor and len(self.bases.args)==0): |
There was a problem hiding this comment.
Maybe not for this PR, but I wonder if we need to handle the "empty tuple" case at all. Isn't that something we can normalise away?
There was a problem hiding this comment.
I've not made any changes here.
As far as I can tell the self.bases check never fails (even for classes with no bases it's always assigned to an empty tuple). Therefore, I think that could go in principle (from both this and from existing code). It's possible I've missed a corner-case where it's useful though.
Explicitly handling an empty tuple is just a small optimization. We don't need to do it though - there's no issue with passing an empty tuple to __Pyx_PEP560_update_bases.
Co-authored-by: scoder <stefan_ml@behnel.de>
| @@ -4762,6 +4762,7 @@ class PyClassDefNode(ClassDefNode): | |||
| metaclass = None | |||
| mkw = None | |||
| doc_node = None | |||
| orig_bases = None | |||
There was a problem hiding this comment.
This also needs to be added to child_attrs to make sure it gets properly traversed. CloneNode blocks the traversal.
And please also document it in the comments above.
|
Thanks! |
Fixes #3537
Both the C code and the tests are largely copied from CPython
so hopefully should be pretty reliable.
This doesn't backport perfectly so I've hard to disable a few
tests of Python 2 (and may have to on earlier Python 3 versions too).
I think that's OK.
I haven't applied this at all to cdef classes. Maybe it could be
applied to the second+subsequent bases but I don't think it's
needed for the initial implementation.