str.is* implementation seem suboptimal for single character strings

[gdementen]

BPO	17559
Nosy	@birkenfeld, @vstinner, @benjaminp, @ezio-melotti, @serhiy-storchaka

^{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 2013-03-28.09:56:32.591>
created_at = <Date 2013-03-27.14:41:09.281>
labels = ['interpreter-core', 'performance']
title = 'str.is* implementation seem suboptimal for single character strings'
updated_at = <Date 2013-03-28.09:56:32.589>
user = 'https://bugs.python.org/gdementen'

bugs.python.org fields:

activity = <Date 2013-03-28.09:56:32.589>
actor = 'georg.brandl'
assignee = 'none'
closed = True
closed_date = <Date 2013-03-28.09:56:32.591>
closer = 'georg.brandl'
components = ['Interpreter Core']
creation = <Date 2013-03-27.14:41:09.281>
creator = 'gdementen'
dependencies = []
files = []
hgrepos = []
issue_num = 17559
keywords = []
message_count = 6.0
messages = ['185338', '185392', '185393', '185394', '185407', '185410']
nosy_count = 6.0
nosy_names = ['georg.brandl', 'vstinner', 'benjamin.peterson', 'ezio.melotti', 'gdementen', 'serhiy.storchaka']
pr_nums = []
priority = 'normal'
resolution = 'works for me'
stage = None
status = 'closed'
superseder = None
type = 'performance'
url = 'https://bugs.python.org/issue17559'
versions = ['Python 3.4']

[gdementen]

In isspace, isalpha, isalnum and isdigit, I see code like:

/* Shortcut for single character strings */
if (PyString_GET_SIZE(self) == 1 &&
isspace(*p))
return PyBool_FromLong(1);

Is it intentional to not use:

if (PyString_GET_SIZE(self) == 1))
return PyBool_FromLong(isspace(*p) != 0);

which would be faster when the result is False (but a tad slower when it is True because of the extra comparison).

Also, is there a reason (other than historical) why the macros Py_RETURN_TRUE and Py_RETURN_FALSE are not used instead of their equivalent functions PyBool_FromLong(1) and PyBool_FromLong(0)?

See:
http://hg.python.org/cpython/file/e87364449954/Objects/stringobject.c#l3324

[benjaminp]

The shortcut seems fairly pointless to me.

[vstinner]

If you would like to improve Python, you have to focus on the development version which is Python 3.4. In this version, the code is different:

    if (length == 1)
        return PyBool_FromLong(
            Py_UNICODE_ISSPACE(PyUnicode_READ(kind, data, 0)));

I'm not sure that having a special case for string of 1 character provide any speed up...

[benjaminp]

There's still stuff in bytes_methods.c which looks like the old string code.

[gdementen]

Argl. I know I should have used 3.4... but that is what I thought I did.
I used http://hg.python.org/cpython then "browse", and assumed it was the default branch... I know realize that since the last commit at that time was on the 2.7 branch, that is what I got. And I didn't even realize my mistake by the string vs unicode differences. Now, I will just go hide somewhere in shame...

[birkenfeld]

Benjamin, if that comment means there's still something to be done, please reopen.