PEP 338: Executing modules inside packages with '-m'

N

Nick Coghlan

Python 2.4's -m command line switch only works for modules directly on sys.path.
Trying to use it with modules inside packages will fail with a "Module not
found" error. This PEP aims to fix that for Python 2.5.

Previously, posting of a draft version of the PEP to python-dev and python-list
didn't actually generate any responses. I'm not sure if that's an indication
that people don't see the restriction to top-level modules as a problem (and
hence think the PEP is unecessary), or think the extension to handle packages is
obvious (and hence see no need to comment).

Or, it could just be a sign that Python 2.4 hasn't been out long enough for
anyone to care what I'm yabbering on about :)

Anyway, all comments are appreciated (even a simple "Sounds good to me").

Cheers,
Nick.

***********************************************************************
PEP: 338
Title: Executing modules inside packages with '-m'
Version: $Revision: 1.2 $
Last-Modified: $Date: 2004/12/11 20:31:10 $
Author: Nick Coghlan <[email protected]>
Status: Draft
Type: Standards Track
Content-Type: text/x-rst
Created: 16-Oct-2004
Python-Version: 2.5
Post-History: 8-Nov-2004


Abstract
========

This PEP defines semantics for executing modules inside packages as
scripts with the ``-m`` command line switch.

The proposed semantics are that the containing package be imported
prior to execution of the script.


Rationale
=========

Python 2.4 adds the command line switch ``-m`` to allow modules to be
located using the Python module namespace for execution as scripts.
The motivating examples were standard library modules such as ``pdb``
and ``profile``.

A number of users and developers have requested extension of the
feature to also support running modules located inside packages. One
example provided is pychecker's ``pychecker.checker`` module. This
capability was left out of the Python 2.4 implementation because the
appropriate semantics were not entirely clear.

The opinion on python-dev was that it was better to postpone the
extension to Python 2.5, and go through the PEP process to help make
sure we got it right.


Scope of this proposal
==========================

In Python 2.4, a module located using ``-m`` is executed just as if
its filename had been provided on the command line. The goal of this
PEP is to get as close as possible to making that statement also hold
true for modules inside packages.

Prior discussions suggest it should be noted that this PEP is **not**
about any of the following:

- changing the idiom for making Python modules also useful as scripts
(see PEP 299 [1]_).

- lifting the restriction of ``-m`` to modules of type PY_SOURCE or
PY_COMPILED (i.e. ``.py``, ``.pyc``, ``.pyo``, ``.pyw``).

- addressing the problem of ``-m`` not understanding zip imports or
Python's sys.metapath.

The issues listed above are considered orthogonal to the specific
feature addressed by this PEP.


Current Behaviour
=================

Before describing the new semantics, it's worth covering the existing
semantics for Python 2.4 (as they are currently defined only by the
source code).

When ``-m`` is used on the command line, it immediately terminates the
option list (like ``-c``). The argument is interpreted as the name of
a top-level Python module (i.e. one which can be found on
``sys.path``).

If the module is found, and is of type ``PY_SOURCE`` or
``PY_COMPILED``, then the command line is effectively reinterpreted
from ``python <options> -m <module> <args>`` to ``python <options>
<filename> <args>``. This includes setting ``sys.argv[0]`` correctly
(some scripts rely on this - Python's own ``regrtest.py`` is one
example).

If the module is not found, or is not of the correct type, an error
is printed.


Proposed Semantics
==================

The semantics proposed are fairly simple: if ``-m`` is used to execute
a module inside a package as a script, then the containing package is
imported before executing the module in accordance with the semantics
for a top-level module.

This is necessary due to the way Python's import machinery locates
modules inside packages. A package may modify its own __path__
variable during initialisation. In addition, paths may affected by
``*.pth`` files. Accordingly, the only way for Python to reliably
locate the module is by importing the containing package and
inspecting its __path__ variable.

Note that the package is *not* imported into the ``__main__`` module's
namespace. The effects of these semantics that will be visible to the
executed module are:

- the containing package will be in sys.modules

- any external effects of the package initialisation (e.g. installed
import hooks, loggers, atexit handlers, etc.)


Reference Implementation
========================

A reference implementation is available on SourceForge [2]_. In this
implementation, if the ``-m`` switch fails to locate the requested
module at the top level, it effectively reinterprets the command from
``python -m <script>`` to ``python -m execmodule <script>``. (There
is one caveat: when reinterpreted in this way, ``sys.argv[0]`` may not
actually contain the filename of ``execmodule``. This only affects
``execmodule`` itself, not the requested module).

``execmodule`` is a proposed standard library module that contains a
single function (also called ``execmodule``). When invoked as a
script, this module finds and executes the module supplied as the
first argument. It adjusts ``sys.argv`` by deleting ``sys.argv[0]``
and replacing the new ``sys.argv[0]`` with the module's filename
instead of its Python name.

The function ``execmodule`` is like ``execfile``, but uses the Python
module namespace to locate the script instead of the filesystem. It
has an additional optional argument ``set_argv0`` which causes the
filename of the located module to be written to ``sys.argv[0]`` before
the module is executed.

A hybrid C/Python implementation is used as the Python module is much
more flexible and extensible than the equivalent C code would be. It
also allows the ``execmodule`` function to be made available. Scripts
which execute other scripts (e.g. ``profile``, ``pdb``) have the
option to use this function to provide ``-m`` style support for
identifying the script to be executed.

The Python code for ``execmodule`` has also been posted as a
cookbook recipe for Python 2.4 [3]_.


Open Issues
===========

- choosing a name for the standard library module containing
``execmodule``. The reference implementation uses ``execmodule``.
An alternative name proposed on python-dev is ``runpy``.


Alternatives
============

The main alternative implementation considered ignored packages'
__path__ variables, and looked only in the main package directory. A
Python script with this behaviour can be found in the discussion of
the ``execmodule`` cookbook recipe [3]_.

This approach was not used as it does not meet the main goal of the
``-m`` switch -- to allow the full Python namespace to be used to
locate modules for execution.


References
==========

... [1] Special __main__() function in modules
(http://www.python.org/peps/pep-0299.html)

... [2] Native ``-m`` execmodule support

(http://sourceforge.net/tracker/?func=detail&aid=1043356&group_id=5470&atid=305470 )

... [3] execmodule Python Cookbook Recipe
(http://aspn.activestate.com/ASPN/Cookbook/Python/Recipe/307772)


Copyright
=========

This document has been placed in the public domain.

...
Local Variables:
mode: indented-text
indent-tabs-mode: nil
sentence-end-double-space: t
fill-column: 70
End:
 
A

alia_khouri

A useful feature that is a logical extension of current '-m' behaviour.
(I'm actually surprised it was left out in the first place)

This will definitely allow me and other python programmers to package
our scripts better

Sounds Good to me. (-;

Thank you for the PEP

AK
 
N

Nick Coghlan

A useful feature that is a logical extension of current '-m' behaviour.
(I'm actually surprised it was left out in the first place)

That seems to be a common reaction :)

It was dropped for 2.4 because I wasn't sure exactly how it should work, and 2.4
was already in beta at the time. So I went with the cookbook recipe instead.

Cheers,
Nick.
 
R

richard

A useful feature that is a logical extension of current '-m' behaviour.
(I'm actually surprised it was left out in the first place)

This will definitely allow me and other python programmers to package
our scripts better

Sounds Good to me. (-;

/me too :)


Richard
 
F

Fredrik Lundh

This will definitely allow me and other python programmers to package
our scripts better

puzzling.

I'd say that for a typical user,

$ python -m foo.bar arg

is a marginal improvement over

$ python -c "import foo.bar" arg

compared to

$ bar arg

</F>
 
N

Nick Coghlan

Fredrik said:
I'd say that for a typical user,

$ python -m foo.bar arg

is a marginal improvement over

$ python -c "import foo.bar" arg

This doesn't work. Any code protected by "if __name__ == '__main__':" won't run
in this context (since 'foo.bar' is being imported as a module, not run as a
script).

Even 'python -c "from foo.bar import _main; _main()" arg' isn't quite right,
since sys.argv[0] will be wrong (it will say '-c', instead of giving the
module's filename). There's also the problem that there is no standard idiom for
_main() functions.
compared to

$ bar arg

This is true, but it has its own problems, mainly in the area of namespace
conflicts on the packaging side:

1. Namespace conflicts between different Python versions
2. Namespace conflicts between different Python packages
3. Namespace conflicts between Python packages and other programs
4. Additional overhead to create an installed module that is usable as a script
a. Add a shebang line for *nix style systems
b. Think about how to deal with the previous 3 points
c. Update the installer to copy the file to the right place with a good name
d. Realise you're screwed on Windows, since you can't control the file
associations and the script will always run with the default interpreter.

An extended -m, on the other hand deals with all those problems automatically:

python -m foo.bar arg # Default interpreter, foo's bar
python -m bar.bar arg # Default interpreter, bar's bar
python24 -m foo.bar arg # Force Python 2.4, foo's bar
python24 -m bar.bar arg # Force Python 2.4, bar's bar
bar arg # Unrelated application called bar

Points 1, 3 & 4 were the justification for adding the current version of -m to
Python 2.4 (obviously, point 2 didn't apply, since the current version doesn't
work for modules insides packages). Specifically, it makes it trivial to get
hold of the right version of pdb and profile for the interpreter you're working
with.

For usability, you can hide all of the above behind a menu item or desktop
shortcut. However, the major target of the feature is Python developers rather
than the end-users of applications built using Python.

Cheers,
Nick.
 
F

Fredrik Lundh

Nick said:
This doesn't work. Any code protected by "if __name__ == '__main__':" won't run in this context
(since 'foo.bar' is being imported as a module, not run as a script).

I appreciate that you're taking the time to teach me about Python, but I can
assure you that it's not really needed.

as for the rest of your arguments, I have to assume that you were joking. (or
that you have no experience whatsoever of distribution of Python programs in
Unix and Windows environments).

</F>
 
J

Just

"Fredrik Lundh said:
I appreciate that you're taking the time to teach me about Python, but I can
assure you that it's not really needed.

Neither is the sarcasm.
as for the rest of your arguments, I have to assume that you were joking.
(or
that you have no experience whatsoever of distribution of Python programs in
Unix and Windows environments).

Whatever. You suggestion does not work in many cases. How about a
program that starts threads? Can't do that as a side effect of import.

Just
 
F

Fredrik Lundh

Just wrote:

Whatever. You suggestion does not work in many cases. How about a
program that starts threads? Can't do that as a side effect of import.

my suggestion was to make sure that the user can type "bar arg" to start a
Python program called "bar" with the argument "arg". that's trivial, on all
major platforms, despite what Nick says -- and yes, you can start threads
from a program named "bar". try it.

</F>
 
J

Just

"Fredrik Lundh said:
Just wrote:



my suggestion was to make sure that the user can type "bar arg" to start a
Python program called "bar" with the argument "arg". that's trivial, on all
major platforms, despite what Nick says -- and yes, you can start threads
from a program named "bar". try it.

This subthread was specifically about your

python -c "import foo.bar" arg

suggestion.

Just
 
N

Nick Coghlan

Fredrik said:
my suggestion was to make sure that the user can type "bar arg" to start a
Python program called "bar" with the argument "arg". that's trivial, on all
major platforms, despite what Nick says -- and yes, you can start threads
from a program named "bar". try it.

The command line switch came out of a pydev discussion regarding making pdb and
profile easy to run for developers using multiple versions of Python on the same
computer (e.g. someone running a Python 2.3 default installation, a Python 2.4
alpha alternate installation and a Python build from CVS).

Making a *single* version of an end-user application available on a platform is,
as you say, quite straightforward. Versioning issues may need to be considered,
but they're usually limited to making sure you get the 'right' Python out of any
which are available on the machine.

And for an application, especially one not aimed at developers, that's almost
certainly the right road to take. You'll have a name already, and presumably a
versioning system as well (I believe dist_utils provides something along those
lines).

Making a Python-version specific developer tool easily accessible isn't quite as
straightforward, since you need to deal with making multiple versions available
simultaneously, and each version has a different definition of the 'right'
interpreter. As happened in the standard library with pdb and profile, the
result is often to just not bother with it - leaving the developer to specify
the full path to the particular version they want to run. That is, quite
frankly, a pain - the same info is getting specified twice (once in selecting
the python version to run, and again in specifying the full path to the
associated tool)

One of the options explored in the original pydev discussion that led to '-m'
was standalone, executable scripts for pdb and profile. The major issue with
that idea was that it didn't scale very well - to use that solution for any
other potentially useful scripts in the standard library, we would have had to
come up with a decent name for each one, and then apply whatever versioning
mechanism we settled on.

Using the Python module namespace to find the script means that all the
versioning issues are already taken care of by the Python interpreters own
versioning system - whichever version of Python you specify in the command line
invocation, you get the correct version of the tool for that version of Python.
It has the added bonus of working not only for pdb and profile, but any other
scripts which are found to be useful (even those in extension packages, if this
PEP is accepted by the BDFL).

Cheers,
Nick.

P.S. As you might have guessed from the above, it's not the least bit
coincidental that the example scripts I generally use when discussing '-m' are
pdb, profile and pychecker.checker. This feature is a convenience for command
line junkies - which seems to be a fairly common trait amongst the developers I
know :)
 
F

Fredrik Lundh

Just said:
This subthread was specifically about your

python -c "import foo.bar" arg

suggestion.

in my original post, I said

I'd say that for a typical user, "A" is a marginal improvement over
"B", compared to "C".

which, I thought, tried to say that for a user expecting "C", neither "A" nor "B"
is good enough.

</F>
 
N

Nick Coghlan

Fredrik said:
in my original post, I said

I'd say that for a typical user, "A" is a marginal improvement over
"B", compared to "C".

which, I thought, tried to say that for a user expecting "C", neither "A" nor "B"
is good enough.

Ah, OK - that makes a lot more sense than the way I read it (it looked to me as
if you expected option B to do the same thing as option A. It didn't seem likely
you really believed that, but that was the only interpretation I saw at the time).

Anyway, as my other rambling message points out (eventually) - this feature is
intended for developers and Python version specific utility scripts like pdb,
profile and pychecker.checker, rather than launch scripts for full applications.

For end users, I agree with you wholeheartedly - applications should behave like
applications, no matter what language they're written in :)

Cheers,
Nick.
 
M

Martin Bless

[Nick Coghlan said:
Python 2.4's -m command line switch only works for modules directly on sys.path.

On my Windows machine this command line switch really makes my life so
much easier. I appreciate -m very much. Going further as proposed in
PEP 338 sounds good to me.

One thing I stumbled across with the current implementation:

Why doesn't "python -m abc" work with

../abc/
../abc/__init__.py

assuming ./abc/ is directly on the path? In analogy to normal module
import?

mb - Martin
 
F

Fredrik Lundh

Martin said:
On my Windows machine this command line switch really makes my life so
much easier.

if you use windows, chances are that what you really want is exemaker...

</F>
 
N

Nick Coghlan

Martin said:
[Nick Coghlan <[email protected]>]
One thing I stumbled across with the current implementation:

Why doesn't "python -m abc" work with

./abc/
./abc/__init__.py

assuming ./abc/ is directly on the path? In analogy to normal module
import?

It doesn't work because abc is a package, rather than a module. There are lots
of things that work as modules for import, but can't be used directly as scripts
(builtin modules, C extension modules, packages, frozen modules, etc)

The command line "python ./abc" doesn't work, either - the closest you get is
"python ./abc/__init__.py".

This is one of the things mentioned in the PEP - the restriction to Python
source code or compiled bytecode is explicit, in order to match the existing
command line behaviour. Even implementation of the PEP won't allow "python -m
abc" to work.

However, with PEP 338, "pep -m abc.__init__" will actually try to run the
package initialisation code as a script, simply due to the way imp.find_module
works. Whether that does anything sane is up to the package author. Something to
realise is that, in this case, __init__.py actually gets run twice - once for
the package import, and then again as a script.

Cheers,
Nick.
 

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments. After that, you can post your question and our members will help you out.

Ask a Question

Members online

No members online now.

Forum statistics

Threads
473,744
Messages
2,569,484
Members
44,903
Latest member
orderPeak8CBDGummies

Latest Threads

Top