gh-112919: Speed-up datetime, date and time.replace()

[eltoder]

Use argument clinic and call new_* functions directly. This speeds up these functions 6x to 7.5x when calling with keyword arguments.

Before:

$ python -m timeit -s "from datetime import datetime; dt = datetime.now()" "dt.replace(microsecond=0)"
500000 loops, best of 5: 501 nsec per loop

$ python -m timeit -s "from datetime import date; d = date.today()" "d.replace(day=1)"
1000000 loops, best of 5: 273 nsec per loop

$ python -m timeit -s "from datetime import time; t = time()" "t.replace(microsecond=0)"
1000000 loops, best of 5: 338 nsec per loop

After:

$ python -m timeit -s "from datetime import datetime; dt = datetime.now()" "dt.replace(microsecond=0)"
5000000 loops, best of 5: 65.9 nsec per loop

$ python -m timeit -s "from datetime import date; d = date.today()" "d.replace(day=1)"
5000000 loops, best of 5: 45.6 nsec per loop

$ python -m timeit -s "from datetime import time; t = time()" "t.replace(microsecond=0)"
5000000 loops, best of 5: 51 nsec per loop

• Issue: datetime.replace() is very slow #112919

[serhiy-storchaka]

In general LGTM, but please generate also __replace__ methods.

[serhiy-storchaka]

There is an issue here. DATETIME_DATE_REPLACE_METHODDEF can be changed with different options of Argument Clinic (for example when we experiment with the limited C API). The generated datetime_date_replace can no longer be compatible with manually written entry for __replace__.

The simplest way to fix this is to add separate definition for __replace__:

/*[clinic input]
datetime.date.__replace__ = datetime.date.replace
[clinic start generated code]*/

static PyObject *
datetime_date___replace___impl(PyDateTime_Date *self, int year, int month,
                               int day)
/*[clinic end generated code: output=bfb97f926e24f201 input=1b2cac755dfa9b40]*/
{
    return new_date_ex(year, month, day, Py_TYPE(self));
}

[eltoder]

If I understand correctly, this means that the bodies of all replace() functions need to be duplicated. This means that all future changes to that code will need to be done twice, which is a bad thing. (The functions may seem trivial, but I'm already working on a further change to address #89039.) This will also generate the argument parsing code twice, which will bloat executable size. A better approach would be to add some syntax to argument clinic to generate the second METHODDEF.

Also, note that I copied this approach from code.__replace__, so if this is not right, code.__replace__ needs to be changed as well.

[eltoder]

@serhiy-storchaka thank you for the review

[serhiy-storchaka]

LGTM.

Well, let merge it as is, and make the code more independent from Argument Clinic changes in other issue. Otherwise we risk to forget about this PR.

[miss-islington-app]

Thanks @eltoder for the PR, and @serhiy-storchaka for merging it 🌮🎉.. I'm working now to backport this PR to: 3.11.
🐍🍒⛏🤖

[miss-islington-app]

Thanks @eltoder for the PR, and @serhiy-storchaka for merging it 🌮🎉.. I'm working now to backport this PR to: 3.12.
🐍🍒⛏🤖

[miss-islington-app]

Sorry, @eltoder and @serhiy-storchaka, I could not cleanly backport this to 3.11 due to a conflict.
Please backport using cherry_picker on command line.

cherry_picker 1f515e8a109204f7399d85b7fd806135166422d9 3.11

[miss-islington-app]

Sorry, @eltoder and @serhiy-storchaka, I could not cleanly backport this to 3.12 due to a conflict.
Please backport using cherry_picker on command line.

cherry_picker 1f515e8a109204f7399d85b7fd806135166422d9 3.12

[bedevere-app]

GH-115344 is a backport of this pull request to the 3.12 branch.

[erlend-aasland]

Thanks, nice change!