Author vstinner
Recipients methane, serhiy.storchaka, vstinner
Date 2017-01-24.09:42:35
SpamBayes Score -1.0
Marked as misclassified Yes
Message-id <1485250955.42.0.234625059413.issue29360@psf.upfronthosting.co.za>
In-reply-to
Content
Serhiy Storchaka: "The assertion was valid while all keywords did came from a constant keywords tuple in CALL_FUNCTION_KW. But if the FASTCALL protocol is extended for var-keyword arguments, keywords can be arbitrary and the assertion can fail. The assertion on the next line can fail too."

The second _PyStack_AsDict() assertion is:

   assert(PyDict_GetItem(kwdict, key) == NULL);

This assertion fails if kwnames (tuple) argument of _PyStack_AsDict() contains duplicated keys, or at least two keys which have the same hash value and are equal.

If someone pass kwnames with duplicates keys on purpose (for whatever) reasons, the assertion fails. I'm not sure that the assertion makes sense since it's a feature of Python to be able to replace dictionary values:

>>> def func(**kw): print(kw)
... 
>>> func(**{'x': 1, 'x': 2})
{'x': 2}
>>> {'x': 1, 'x': 2}
{'x': 2}

Are you suggesting to remove "assert(PyDict_GetItem(kwdict, key) == NULL);" too?


Note: In Python 3.6, CALL_FUNCTION_EX already uses FASTCALL for C functions using METH_FASTCALL. In this case, _PyCFunction_FastCallDict() uses _PyStack_UnpackDict(). For example, open() already uses METH_FASTCALL in Python 3.6. In there an issue with _PyStack_UnpackDict()? (I don't think so.)
History
Date User Action Args
2017-01-24 09:42:35vstinnersetrecipients: + vstinner, methane, serhiy.storchaka
2017-01-24 09:42:35vstinnersetmessageid: <1485250955.42.0.234625059413.issue29360@psf.upfronthosting.co.za>
2017-01-24 09:42:35vstinnerlinkissue29360 messages
2017-01-24 09:42:35vstinnercreate