__main__.py is used for python programs in zip files. The __main__.py file will be executed when the zip file in run. For example, if the zip file was as such:

test.zip
     __main__.py

and the contents of __main__.py was

import sys
print "hello %s" % sys.argv[1]

Then if we were to run python test.zip world we would get hello world out.

So the __main__.py file run when python is called on a zip file.

If your script is a directory or ZIP file rather than a single python file, __main__.py will be executed when the “script” is passed as an argument to the python interpreter.

You create __main__.py in yourpackage to make it executable as:

$ python -m yourpackage

What is the __main__.py file for?

When creating a Python module, it is common to make the module execute some functionality (usually contained in a main function) when run as the entry point of the program. This is typically done with the following common idiom placed at the bottom of most Python files:

if __name__ == '__main__':
    # execute only if run as the entry point into the program
    main()

You can get the same semantics for a Python package with __main__.py, which might have the following structure:

.
└── demo
    ├── __init__.py
    └── __main__.py

To see this, paste the below into a Python 3 shell:

from pathlib import Path

demo = Path.cwd() / 'demo'
demo.mkdir()

(demo / '__init__.py').write_text("""
print('demo/__init__.py executed')

def main():
    print('main() executed')
""")

(demo / '__main__.py').write_text("""
print('demo/__main__.py executed')

from demo import main

main()
""")

We can treat demo as a package and actually import it, which executes the top-level code in the __init__.py (but not the main function):

>>> import demo
demo/__init__.py executed

When we use the package as the entry point to the program, we perform the code in the __main__.py, which imports the __init__.py first:

$ python -m demo
demo/__init__.py executed
demo/__main__.py executed
main() executed

You can derive this from the documentation. The documentation says:

__main__ — Top-level script environment

'__main__' is the name of the scope in which top-level code executes.
A module’s __name__ is set equal to '__main__' when read from standard
input, a script, or from an interactive prompt.

A module can discover whether or not it is running in the main scope
by checking its own __name__, which allows a common idiom for
conditionally executing code in a module when it is run as a script or
with python -m but not when it is imported:

if __name__ == '__main__':
     # execute only if run as a script
     main()

For a package, the same effect can be achieved by including a
__main__.py module, the contents of which will be executed when the module is run with -m.

Zipped

You can also zip up this directory, including the __main__.py, into a single file and run it from the command line like this – but note that zipped packages can’t execute sub-packages or submodules as the entry point:

from pathlib import Path

demo = Path.cwd() / 'demo2'
demo.mkdir()

(demo / '__init__.py').write_text("""
print('demo2/__init__.py executed')

def main():
    print('main() executed')
""")

(demo / '__main__.py').write_text("""
print('demo2/__main__.py executed')

from __init__ import main

main()
""")

Note the subtle change – we are importing main from __init__ instead of demo2 – this zipped directory is not being treated as a package, but as a directory of scripts. So it must be used without the -m flag.

Particularly relevant to the question – zipapp causes the zipped directory to execute the __main__.py by default – and it is executed first, before __init__.py:

$ python -m zipapp demo2 -o demo2zip
$ python demo2zip
demo2/__main__.py executed
demo2/__init__.py executed
main() executed

Note again, this zipped directory is not a package – you cannot import it either.

Some of the answers here imply that given a "package" directory (with or without an explicit __init__.py file), containing a __main__.py file, there is no difference between running that directory with the -m switch or without.

The big difference is that without the -m switch, the "package" directory is first added to the path (i.e. sys.path), and then the files are run normally, without package semantics.

Whereas with the -m switch, package semantics (including relative imports) are honoured, and the package directory itself is never added to the system path.

This is a very important distinction, both in terms of whether relative imports will work or not, but more importantly in terms of dictating what will be imported in the case of unintended shadowing of system modules.

---

Example:

Consider a directory called PkgTest with the following structure

:~/PkgTest$ tree
.
├── pkgname
│   ├── __main__.py
│   ├── secondtest.py
│   └── testmodule.py
└── testmodule.py

where the __main__.py file has the following contents:

:~/PkgTest$ cat pkgname/__main__.py
import os
print( "Hello from pkgname.__main__.py. I am the file", os.path.abspath( __file__ ) )
print( "I am being accessed from", os.path.abspath( os.curdir ) )
from  testmodule import main as firstmain;     firstmain()
from .secondtest import main as secondmain;    secondmain()

(with the other files defined similarly with similar printouts).

If you run this without the -m switch, this is what you’ll get. Note that the relative import fails, but more importantly note that the wrong testmodule has been chosen (i.e. relative to the working directory):

:~/PkgTest$ python3 pkgname
Hello from pkgname.__main__.py. I am the file ~/PkgTest/pkgname/__main__.py
I am being accessed from ~/PkgTest
Hello from testmodule.py. I am the file ~/PkgTest/pkgname/testmodule.py
I am being accessed from ~/PkgTest
Traceback (most recent call last):
  File "/usr/lib/python3.6/runpy.py", line 193, in _run_module_as_main
    "__main__", mod_spec)
  File "/usr/lib/python3.6/runpy.py", line 85, in _run_code
    exec(code, run_globals)
  File "pkgname/__main__.py", line 10, in <module>
    from .secondtest import main as secondmain
ImportError: attempted relative import with no known parent package

Whereas with the -m switch, you get what you (hopefully) expected:

:~/PkgTest$ python3 -m pkgname
Hello from pkgname.__main__.py. I am the file ~/PkgTest/pkgname/__main__.py
I am being accessed from ~/PkgTest
Hello from testmodule.py. I am the file ~/PkgTest/testmodule.py
I am being accessed from ~/PkgTest
Hello from secondtest.py. I am the file ~/PkgTest/pkgname/secondtest.py
I am being accessed from ~/PkgTest

---

---

Note: In my honest opinion, running without -m should be avoided. In fact I would go further and say that I would create any executable packages in such a way that they would fail unless run via the -m switch.

In other words, I would only import from ‘in-package’ modules explicitly via ‘relative imports’, assuming that all other imports represent system modules. If someone attempts to run your package without the -m switch, the relative import statements will throw an error, instead of silently running the wrong module.