RichCompare and RichCompareBool

[Aaron Brady]

On Mar 2, 9:24 pm, Terry Reedy <tjre... at udel.edu> wrote:
> Gabriel Genellina wrote:
> > En Mon, 02 Mar 2009 17:54:09 -0200, Terry Reedy <tjre... at udel.edu>
> > escribió:
>
> >> Aaron Brady wrote:
> >>> Hi,
> >>>  In the source for 3.0.1, PyObject_RichCompareBool seems to perform an
> >>> extra check on identity that PyObjecct_RichCompare does not perform.
>
> >> To me, the existence of two functions suggests that they are
> >> *intended* to act differently.
>
> > Mmm, but then the documentation is misleading and incomplete, at least.
> > PyObject_RichCompareBool is not equivalent of "o1 op o2" as it claims,
> > and both functions differ in more than return value (the only documented
> > difference).
>
> Could be.  I believe the C API docs have had must less testing by naive
> users than the Python docs, and even those are not quite perfect yet.

RichCompareBool is used 62 times in the source, minus declaration.
It's partly (by eye) used to test identity with Py_EQ.  Then there
should be a different function, more like "IdCompareBool".

Duncan stated:
> Without looking at the code it sounds like a bug: identity doesn't always
> imply equality.

In such cases, should "a in [a]" be True or False?

Abstract example:

>>> class A:
...     def __eq__( self, oth ):
...         if self is oth:
...             return False
...         return object.__eq__( self, oth )
...
>>> a= A()
>>> a==a
False
>>> [a]==[a]
True

'list' does seem to break its contract.  However without a concrete
use, I can't make a judgment for a vote.

Also, did not receive Gabriel's post.