Vectorcall implementation should conform to PEP 590.

[markshannon]

BPO	39978
Nosy	@vstinner, @jkloth, @encukou, @markshannon

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = None
created_at = <Date 2020-03-16.14:09:28.404>
labels = ['3.9']
title = 'Vectorcall implementation should conform to PEP 590.'
updated_at = <Date 2020-03-16.15:14:03.284>
user = 'https://github.com/markshannon'

bugs.python.org fields:

activity = <Date 2020-03-16.15:14:03.284>
actor = 'vstinner'
assignee = 'none'
closed = False
closed_date = None
closer = None
components = []
creation = <Date 2020-03-16.14:09:28.404>
creator = 'Mark.Shannon'
dependencies = []
files = []
hgrepos = []
issue_num = 39978
keywords = ['3.9regression']
message_count = 2.0
messages = ['364325', '364331']
nosy_count = 4.0
nosy_names = ['vstinner', 'jkloth', 'petr.viktorin', 'Mark.Shannon']
pr_nums = []
priority = 'normal'
resolution = None
stage = None
status = 'open'
superseder = None
type = None
url = 'https://bugs.python.org/issue39978'
versions = ['Python 3.9']

[markshannon]

The implementation of PyObject_Vectorcall adds unnecessary overhead to PEP-590, which undermines its purpose.
The implementation was changed in #17052, which I objected to at the time.

The change has a negative impact on performance as it add calls to both
PyThreadState_GET() and _Py_CheckFunctionResult().

This is practically an invitation for callers to skip PyObject_Vectorcall and access the underlying function pointer directly, making PyObject_Vectorcall pointless.

#17052 should be reverted.

[vstinner]

The change has a negative impact on performance

How did you notice the slowdown? I ran benchmarks in #17052 and I didn't notice any performance overhead.

[markshannon]

It's too late to revert #17052 now.