bpo-39406: Implement os.putenv() with setenv() if available

[vstinner]

If setenv() C function is available, os.putenv() is now implemented
with setenv() instead of putenv(), so Python doesn't have to handle
the environment variable memory.

If setenv() is available, posix_putenv_garbage state is not needed.

https://bugs.python.org/issue39406

[vstinner]

I updated the PR to use SetEnvironmentVariable() on Windows.

@serhiy-storchaka: Would you mind to review the updated PR?

[vstinner]

On Windows, I removed the explicit check on _MAX_ENV:

    if (size > _MAX_ENV) {
        PyErr_Format(PyExc_ValueError,
                     "the environment variable is longer than %u characters",
                     _MAX_ENV);
        goto error;
    }

On my Windows 10 VM, _MAX_ENV is equal to 32767, but SetEnvironmentVariableW() seems to accept way longer values!

>>> value='x' * (32767); os.environ['long'] = value; os.environ['long'] == value, len(value)    
(True, 32767)
>>> value='x' * (32767+10); os.environ['long'] = value; os.environ['long'] == value, len(value) 
(True, 32777)
>>> value='abcd' * (32767); os.environ['long'] = value; os.environ['long'] == value, len(value) 
(True, 131068)

But I got ERROR_NO_SYSTEM_RESOURCES (error 1450: "Insufficient system resources exist to complete the requested service") for a value of 1 GB:

>>> import os
>>> value='x'*(2**30)
>>> os.putenv('long', value)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
OSError: [WinError 1450] Ressources système insuffisantes pour terminer le service demandé

The limit seems to be the memory rather than an hardcoded limit. A "shorter" value (64 MB) is fine:

>>> import os
>>> value='x'*(2**26)
>>> os.putenv('long', value)

[serhiy-storchaka]

Why rename it? This increases the size of the diff.

[serhiy-storchaka]

Suggested change

	#elif defined(HAVE_PUTENV)
	#elif defined(HAVE_SETENV) || defined(HAVE_PUTENV)

[serhiy-storchaka]

Suggested change

	#endif /* HAVE_PUTENV */
	#endif /* HAVE_SETENV || HAVE_PUTENV */

[vstinner]

Oh, a test fail on Windows:

2020-01-21T10:50:49.1283745Z ======================================================================
2020-01-21T10:50:49.1283887Z ERROR: test_unset_error (test.test_os.EnvironTests)
2020-01-21T10:50:49.1284029Z ----------------------------------------------------------------------
2020-01-21T10:50:49.1284166Z Traceback (most recent call last):
2020-01-21T10:50:49.1284296Z   File "D:\a\cpython\cpython\lib\test\support\__init__.py", line 695, in wrapper
2020-01-21T10:50:49.1284429Z     return func(*args, **kw)
2020-01-21T10:50:49.1284504Z   File "D:\a\cpython\cpython\lib\test\test_os.py", line 962, in test_unset_error
2020-01-21T10:50:49.1284628Z     self.assertRaises(ValueError, os.environ.__delitem__, key)
2020-01-21T10:50:49.1284764Z   File "D:\a\cpython\cpython\lib\unittest\case.py", line 799, in assertRaises
2020-01-21T10:50:49.1284895Z     return context.handle('assertRaises', args, kwargs)
2020-01-21T10:50:49.1285989Z   File "D:\a\cpython\cpython\lib\unittest\case.py", line 202, in handle
2020-01-21T10:50:49.1286157Z     callable_obj(*args, **kwargs)
2020-01-21T10:50:49.1286268Z   File "D:\a\cpython\cpython\lib\os.py", line 691, in __delitem__
2020-01-21T10:50:49.1286377Z     raise KeyError(key) from None
2020-01-21T10:50:49.1295463Z KeyError: 

[vstinner]

I made other changes in the meanwhile which introduced conflicts. So I wrote a new PR instead: PR #18128.