Is module __file__ attribute absolute or relative?

__file__ is absolute since Python 3.4, except when executing a script directly using a relative path:

Module __file__ attributes (and related values) should now always contain absolute paths by default, with the sole exception of __main__.__file__ when a script has been executed directly using a relative path. (Contributed by Brett Cannon in bpo-18416.)

Not sure if it resolves symlinks though.

Example of passing a relative path:

$ python script.py

Late simple example:

from os import path, getcwd, chdir

def print_my_path():
    print('cwd:     {}'.format(getcwd()))
    print('__file__:{}'.format(__file__))
    print('abspath: {}'.format(path.abspath(__file__)))

print_my_path()

chdir('..')

print_my_path()

Under Python-2.*, the second call incorrectly determines the path.abspath(__file__) based on the current directory:

cwd:     C:\codes\py
__file__:cwd_mayhem.py
abspath: C:\codes\py\cwd_mayhem.py
cwd:     C:\codes
__file__:cwd_mayhem.py
abspath: C:\codes\cwd_mayhem.py

As noted by @techtonik, in Python 3.4+, this will work fine since __file__ returns an absolute path.

__file__ can be relative or absolute depending on the python version used and whether the module is executed directly or not.

TL;DR:

• From python 3.5 to 3.8, __file__ is set to the relative path of the module w.r.t. the current working directory if the module is called directly. Otherwise it is set to the absolute path.
• Since python 3.9, __file__ is set to the absolute path even if the corresponding module is executed directly.

This behavior is explained in What's new In Python 3.9 (thanks to @user0's comment):

Python now gets the absolute path of the script filename specified on the command line (ex: python3 script.py): the __file__ attribute of the __main__ module became an absolute path, rather than a relative path. These paths now remain valid after the current directory is changed by os.chdir(). As a side effect, the traceback also displays the absolute path for __main__ module frames in this case.

Toy example: For instance, having the following setup (inspired by this answer):

# x.py:
from pathlib import Path
import y
print(__file__)
print(Path(__file__))
print(Path(__file__).resolve())

# y.py:
from pathlib import Path
print(__file__)
print(Path(__file__))

with x.py and y.py in the same directory. The different outputs after going to this directory and executing:

python x.py

are:

Python 3.5 - 3.8:

D:\py_tests\y.py
D:\py_tests\y.py
x.py
x.py
D:\py_tests\x.py

Python 3.9 - 3.11

D:\py_tests\y.py
D:\py_tests\y.py
D:\py_tests\x.py
D:\py_tests\x.py
D:\py_tests\x.py

With the help of the of Guido mail provided by @kindall, we can understand the standard import process as trying to find the module in each member of sys.path, and file as the result of this lookup (more details in PyMOTW Modules and Imports.). So if the module is located in an absolute path in sys.path the result is absolute, but if it is located in a relative path in sys.path the result is relative.

Now the site.py startup file takes care of delivering only absolute path in sys.path, except the initial '', so if you don't change it by other means than setting the PYTHONPATH (whose path are also made absolute, before prefixing sys.path), you will get always an absolute path, but when the module is accessed through the current directory.

Now if you trick sys.path in a funny way you can get anything.

As example if you have a sample module foo.py in /tmp/ with the code:

import sys
print(sys.path)
print (__file__)

If you go in /tmp you get:

>>> import foo
['', '/tmp', '/usr/lib/python3.3', ...]
./foo.py

When in in /home/user, if you add /tmp your PYTHONPATH you get:

>>> import foo
['', '/tmp', '/usr/lib/python3.3', ...]
/tmp/foo.py

Even if you add ../../tmp, it will be normalized and the result is the same.

But if instead of using PYTHONPATH you use directly some funny path you get a result as funny as the cause.

>>> import sys
>>> sys.path.append('../../tmp')
>>> import foo
['', '/usr/lib/python3.3', .... , '../../tmp']
../../tmp/foo.py

Guido explains in the above cited thread, why python do not try to transform all entries in absolute paths:

we don't want to have to call getpwd() on every import .... getpwd() is relatively slow and can sometimes fail outright,

So your path is used as it is.