Why does 2.3a2 change "slots" in complex?

[Greg Chapman]

On Fri, 28 Feb 2003 22:03:45 -0500, mertz at gnosis.cx (David Mertz) wrote:

>One of the unit tests involves adding extra attributes to descendents of
>standard data types.  It turns out, that test breaks with Python 2.3a2,
>because of a change in the implementation of complex.
>
>Here is a brief test case that shows the behavior.  I am interested in
>why this change was made.  Is it a bug or a feature?  If a feature, what
>benefit does this have?  If a bug... well, I guess that's the point of
>an alpha version.

If you check the type of c in your example, you'll find that it is complex under
2.3a2 (rather than __main__.Complex).

I believe this is definitely a bug.  It appears this problem was introduced by a
change which causes the complex constructor (complex_new) to return its argument
if it is of the exact type complex (rather than creating a new complex
instance).  Instead, I believe complex_new should also check to see if the type
being constructed (i.e., the type parameter) is in fact exactly complex; if not,
then it should try to create the new instance no matter what arguments it gets.
I believe the correct fix would be to change this test:

	if (PyComplex_CheckExact(r) && i == NULL) 

to this:

	if (PyComplex_CheckExact(r) && i == NULL && PyComplex_CheckExact(type))

I'll file a bug report.

---
Greg Chapman