[C API] Add Py_NewRef() function to get a new strong reference to an object

[vstinner]

BPO	42262
Nosy	@vstinner, @erlend-aasland
PRs	bpo-42262: Add Py_NewRef() function #23152 bpo-42262: Py_NewRef() casts its argument to PyObject* #23626

^{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 2020-11-05.14:11:51.499>
created_at = <Date 2020-11-04.16:20:34.933>
labels = ['expert-C-API', '3.10']
title = '[C API] Add Py_NewRef() function to get a new strong reference to an object'
updated_at = <Date 2020-12-03.13:01:17.892>
user = 'https://github.com/vstinner'

bugs.python.org fields:

activity = <Date 2020-12-03.13:01:17.892>
actor = 'vstinner'
assignee = 'none'
closed = True
closed_date = <Date 2020-11-05.14:11:51.499>
closer = 'vstinner'
components = ['C API']
creation = <Date 2020-11-04.16:20:34.933>
creator = 'vstinner'
dependencies = []
files = []
hgrepos = []
issue_num = 42262
keywords = ['patch']
message_count = 4.0
messages = ['380337', '380339', '380409', '382407']
nosy_count = 3.0
nosy_names = ['vstinner', 'hodgestar', 'erlendaasland']
pr_nums = ['23152', '23626']
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = None
url = 'https://bugs.python.org/issue42262'
versions = ['Python 3.10']

[vstinner]

In C, the following pattern (A) is very common:

    Py_INCREF(sysdict);
    interp->sysdict = sysdict;

• (1) Increment the reference counter
• (2) Store a reference to the object somewhere

Similar pattern (B) using return:

    Py_INCREF(temp);
    return temp;

The problem in these patterns is that the object has to be written twice. I propose to add a simple new Py_NewRef() function which returns a new strong reference. In short, it's just:

static inline PyObject* Py_NewRef(PyObject *obj)
{
    Py_INCREF(obj);
    return obj;
}

(The actual implementation is just a little bit more complex, to also export it as a regular function.)

Pattern (A) becomes:

interp->sysdict = Py_NewRef(sysdict);

Pattern (B) becomes:

return Py_NewRef(temp);

Py_NewRef() might help to prepare a C extension to be converted to HPy which uses a HPy_Dup() function. HPy_Dup() is different than "Py_INCREF(obj), obj" for subtle reasons. I let you dig into HPy documentation for the rationale:

https://hpy.readthedocs.io/en/latest/api.html#handles

Even if you ignore HPy, Py_NewRef() avoids to repeat the object variable, and so makes the code simpler. Simple example:

-#define Py_RETURN_NONE return Py_INCREF(Py_None), Py_None
+#define Py_RETURN_NONE return Py_NewRef(Py_None)

[vstinner]

The problem in these patterns is that the object has to be written twice.

I'm talking about the variable name which has to be repeated (written twice) in the source code.

[vstinner]

New changeset 53a03aa by Victor Stinner in branch 'master':
bpo-42262: Add Py_NewRef() and Py_XNewRef() (GH-23152)
53a03aa

[vstinner]

New changeset 8b6c4a9 by Victor Stinner in branch 'master':
bpo-42262: Py_NewRef() casts its argument to PyObject* (GH-23626)
8b6c4a9