PyVectorcall_NARGS(): change return type to Py_ssize_t

[vstinner]

BPO	39611
Nosy	@vstinner, @encukou, @jdemeyer

^{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 = <Date 2020-02-18.12:20:27.391>
created_at = <Date 2020-02-11.16:50:02.061>
labels = ['expert-C-API', 'invalid', '3.9']
title = 'PyVectorcall_NARGS(): change return type to Py_ssize_t'
updated_at = <Date 2020-02-18.13:58:06.526>
user = 'https://github.com/vstinner'

bugs.python.org fields:

activity = <Date 2020-02-18.13:58:06.526>
actor = 'vstinner'
assignee = 'none'
closed = True
closed_date = <Date 2020-02-18.12:20:27.391>
closer = 'petr.viktorin'
components = ['C API']
creation = <Date 2020-02-11.16:50:02.061>
creator = 'vstinner'
dependencies = []
files = []
hgrepos = []
issue_num = 39611
keywords = []
message_count = 4.0
messages = ['361824', '362198', '362199', '362215']
nosy_count = 3.0
nosy_names = ['vstinner', 'petr.viktorin', 'jdemeyer']
pr_nums = []
priority = 'normal'
resolution = 'not a bug'
stage = 'resolved'
status = 'closed'
superseder = None
type = None
url = 'https://bugs.python.org/issue39611'
versions = ['Python 3.9']

[vstinner]

I propose to change PyVectorcall_NARGS() return type from unsigned size_t to signed Py_ssize_t.

Currently, the function is defined as:

static inline Py_ssize_t
PyVectorcall_NARGS(size_t n)
{
    return n & ~PY_VECTORCALL_ARGUMENTS_OFFSET;
}

But in CPython code base, the result is always stored in a *signed* Py_ssize_t:

    Py_ssize_t nargs = PyVectorcall_NARGS(nargsf);

Sometimes, this nargs is passed to _PyObject_MakeTpCall() which expects nargs to be Py_ssize_t, so it's consistent.

In general in CPython, a size uses type Py_ssize_t, not size_t. Example: PyVarObject.ob_size type is Py_ssize_t.

[encukou]

The current return type already is Py_ssize_t, exactly for the reason you mention – compatibility with all other "argument count" values in Python. (It would be more correct to use unsigned, but that ship has sailed.)

The *argument* type is unsigned size_t, though: unsigned is the correct type for for bit fields. Also, the "nargsf" value should never be directly used as argument count; making it a different type tends to trigger nice compiler warnings.

[encukou]

Closing; please reopen if we're somehow misunderstanding each other :)

[vstinner]

Alright, I was confused by how PyVectorcall_NARGS() is supposed to be used.