gh-83923: Added os.getdtablesize function to os module

[rruuaanng]

• Issue: Enhancement: add os.getdtablesize(..) to os (posix) module #83923

What's confusing me now is that it seems there's no equivalent to getdtablesize on Windows. If it's allowed, I could write an API to provide an estimated value for the Windows system.

And I'm a bit confused about the clinic tag. This is my first time submitting changes to the C-API. Could you point out any mistakes I might have made? Thanks!!

[rruuaanng]

cc @skirpichev

[rruuaanng]

I think the failure of document builind isn't important, because it is a C extension. So there is no Python definition.

[skirpichev]

No, it is. CI status should be green.

You should document new functions in sphinx docs as well.

[rruuaanng]

No, it is. CI status should be green.

You should document new functions in sphinx docs as well.

Okay, I'm adding os related documents, and eliminating CI failed.

[vstinner]

It sounds interesting to expose getdtablesize() in Python.

[vstinner]

It seems like the function is available on FreeBSD, macOS and Solaris, but not on OpenBSD nor NetBSD.

[vstinner]

It seems like the function is available on FreeBSD, macOS and Solaris, but not on OpenBSD nor NetBSD.

You should check if the function is available by adding getdtablesize to configure.ac in this list:

# checks for library functions
AC_CHECK_FUNCS([ \
  accept4 alarm bind_textdomain_codeset chmod chown clock closefrom close_range
  copy_file_range ctermid dup dup3 execv explicit_bzero explicit_memset \
  faccessat fchmod fchmodat fchown fchownat fdopendir fdwalk fexecve \
  ...

And then run make regen-configure.

[rruuaanng]

Okay, Thanks for pointing out my mistake, and thanks for your review!

[vstinner]

@serhiy-storchaka @corona10 @erlend-aasland: Do you think that it's worth it to expose getdtablesize() in the os module? The Linux manual page says:

Portable applications should employ sysconf(_SC_OPEN_MAX) instead of this call.

os.sysconf("SC_OPEN_MAX") is already available on Python.

[rruuaanng]

In #83923

getdtablesize({2,3}) is a wonderful library call/system call to have access to because it allows one to determine the descriptor limits at runtime in an easier manner than having to do the equivalent with os.sysconf(..)

I think this‘s necessary, and intuitive.

[serhiy-storchaka]

You can also use resource.getrlimit(resource.RLIMIT_NOFILE).

I do not think that we need to implement a third way.

[rruuaanng]

You can also use resource.getrlimit(resource.RLIMIT_NOFILE).

I do not think that we need to implement a third way.

If you're familiar with POSIX but not with Python, this can be pretty tough for you. Like me, for example.

[serhiy-storchaka]

getdtablesize() is not in POSIX, unlike sysconf() and getrlimit(). This is another argument not to add it.

[rruuaanng]

getdtablesize() is not in POSIX, unlike sysconf() and getrlimit(). This is another argument not to add it.

Oh sorry, it should be BSD. Just like #83923, It seems like nothing else works as well as calling someone by their name when it comes to making them feel good.

[rruuaanng]

Because of #ifndef MS_WINDOWS, CI appears declared but not defined. Maybe I should move #ifndef MS_WINDOWS into the os_getdtablesize_impl.

D:\a\cpython\cpython\Modules\clinic\posixmodule.c.h(1432,1): error C2129: static function 'PyObject *os_getdtablesize_impl(PyObject *)' declared but not defined [D:\a\cpython\cpython\PCbuild\_freeze_module.vcxproj]
         (compiling source file '../Modules/posixmodule.c')
             D:\a\cpython\cpython\Modules\clinic\posixmodule.c.h(1432,1):
             see declaration of 'os_getdtablesize_impl'

/*[clinic input]
os.getdtablesize

Return the maximum number of files a process can have open.
[clinic start generated code]*/

static PyObject *
os_getdtablesize_impl(PyObject *module)
{
    int size;
#ifndef MS_WINDOWS
    size = getdtablesize();
#endif  /* !MS_WINDOWS */
    return PyLong_FromLong(size);
}

or

#ifndef MS_WINDOWS
...
#else
static PyObject *
os_getdtablesize_impl(PyObject *Py_UNUSED(module)) { Py_RETURN_NONE; }
#endif  /* !MS_WINDOWS */

[vstinner]

@rruuaanng: I suggest you to pause the implementation until we decide if we should add the function or not.

[rruuaanng]

@rruuaanng: I suggest you to pause the implementation until we decide if we should add the function or not.

Okay, I think so.

[vstinner]

@serhiy-storchaka:

getdtablesize() is not in POSIX, unlike sysconf() and getrlimit(). This is another argument not to add it.

The glibc implements getdtablesize() by calling getrlimit(RLIMIT_NOFILE) on POSIX.

Let's close the issue, I agree with @serhiy-storchaka that this function is redundant and it's not worth it to add it. Sorry @rruuaanng. I close the issue.