New submission from Edward Catmur:
Suppose we have a class C with MRO (C, B, A, object). C virtual-inherits an
ABC V, while B virtual-inherits an unrelated ABC W:
object
/ | \
A W |
| .` /
B` V
| .`
C`
Recalling that per PEP 443 singledispatch prefers concrete bases to virtual
bases, we would expect the following composed MRO:
C, B, V, A, W, object
However what actually happens is the composed MRO depends on the order of the
haystack; if W is processed first the result is correct, but if V is processed
first then (because V does not subclass W) W is inserted in the MRO *before* V:
C, B, A, object
C, B, V, A, object
C, B, W, V, A, object
This results in ambiguity between V and W.
Suggested fix is a slight change to the MRO composition algorithm, considering
whether the items already placed in the MRO are concrete base classes.
----------
components: Extension Modules
hgrepos: 200
messages: 191350
nosy: ecatmur
priority: normal
severity: normal
status: open
title: singledispatch: When virtual-inheriting ABCs at distinct points in MRO,
composed MRO is dependent on haystack ordering
versions: Python 3.4
_______________________________________
Python tracker <[email protected]>
<http://bugs.python.org/issue18244>
_______________________________________
_______________________________________________
Python-bugs-list mailing list
Unsubscribe:
http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
