Quite often the names of subcommands are quite long so that their meaning is clear.  However, the downside to this is the increased typing the user of the command must make.  A compromise between clarity and brevity can be realized if abbreviation of subcommands is allowed with the proviso that the abbreviation must be long enough to remove any ambiguity.

So that this feature is not foisted on every programmer that uses argparse it could be made an option to the ArgumentParser class with the (possible) additional refinement of specifying a minimum number of characters for the abbreviations.

Interesting idea.  I know that with Mercurial for example, I use abbreviations and aliases all the time.  Note that argparse already has aliases (or there is an open feature request about it).

Steven: What do you think of this request?

I just made such a change to Python 2.7's argparse.
If there is interest I'll post a patch here.

Unfortunately I can't find the description how to produce a proper patch.
The link I found (http://www.python.org/dev/patches/) gives an error 404.

http://docs.python.org/devguide/patch.html

I attached a patch against argparse.py from Python 2.7.1.

Subcommmands can now be abbreviated as long as they are unambiguous.
Otherwise an error message 'ambigous choice' will be thrown (like the 'invalid choice' one).

(It's my first patch, so hopefully I did it right.)

BTW: Éric mentioned aliases for argparse. 
For that I have done a patch too. 
I'll try to find the appropriate issue number here to post it.

Since there seems to be no means to edit (my last) message a little followup regarding aliases:

I found http://bugs.python.org/issue9234 
"argparse: aliases for positional arguments (subparsers)"

That one is for version 3.2 and already closed. 
So I'll stop bugging here with aliases.

Yes, I checked and aliases are already supported.

You seem to have overlooked the Versions field on the top of this page: new features don’t go in stable releases, so your patch would have to apply to default, a.k.a. 3.3.

Just for my understanding:
Even "the current development/testing version: Python 3.2.2 rc 1 (August 14, 2011)" isn't new enough here?

So I'll have to find out how to install Python 3.3 in parallel to my 2.7.1 (on Fedora 15).

Starting with "hg clone http://hg.python.org/cpython" now.

Until later (after reading http://docs.python.org/devguide/ by myself of course ;-)

> Even "the current development/testing version: Python 3.2.2 rc 1
> (August 14, 2011)" isn't new enough here?
No.  (Where does the quote come from?)  The current development version is 3.3; its version number is currently 3.3.0a0, to mean it was not released yet.  3.2 is a stable release now; 3.2.2c1 is the candidate release of the next bugfix version of 3.2, which is one of the stable versions (with 2.7).

> So I'll have to find out how to install Python 3.3 in parallel to my 2.7.1
> Starting with "hg clone http://hg.python.org/cpython" now.
Yep, there’s no need to install.  Cloning http://hg.python.org/cpython#default will get you 3.3.  If you already have a clone for 3.2, you can also pull the default branch in your existing clone; feel free to ask for anything that’s not in the devguide or not clear.

>> Even "the current development/testing version: Python 3.2.2 rc 1
>> (August 14, 2011)" isn't new enough here?
>
> No.  (Where does the quote come from?) 

http://python.org/download/releases/

> Cloning http://hg.python.org/cpython#default will get you 3.3

I have 3.3 compiled and running now. :-)

> feel free to ask for anything that’s not in the devguide or not clear.

Thanks for the offer!

> http://python.org/download/releases/
Ah, I understand: the release candidate for 3.2.2 is a testing release, but not a development one :)

Attached a patch against Python 3.3

All the existing tests are still running successfully.

But I wasn't able to add additional tests for the new functionality. 
Writing proper tests is a bit over my head for now - still learning.

test_argparse has a proper test infrastructure so that people don’t have to write the same boilerplate and assertion code over and over again.  If you want, you can probably learn by copy-pasting things.

What you want to test is basically that shortened forms work and conflicting abbreviations report an error.

After some interesting hours of work (learning about tests) I created a new patch, including my original code changes as in the former patch and the new additional tests as well.

Reviewed.  Steven: the patch is complete, pending a few doc editions.

Modulo the comments already on the patch by others, this approach looks fine to me.

You could consider just a small refinement: in the first loop in _SubParsersAction.__call__ where you look for the abbreviation, you can just set parser_name = p and break. Then the logic just below that can stay as it is: all you've done is morphed the abbreviation to the full command name.

Ouch, I really missed this one for a long time. :-(
(I didn't understand the workflow correctly and overlooked the reviews.)
I apologize to everyone who has been involved!

Finally I'm back here and re-did my patch for 3.4 this time.

I followed the comments of the reviewers (and Vinay's suggestion) as far as possible.

The latest patch seems okay to me.  I've successfully applied it and ran the test suite and everything passed.  Could someone please take a look with a view to getting this committed, thanks.

New changeset 97b28115bd08 by Vinay Sajip in branch 'default':
Closes #12713: Allowed abbreviation of subcommands in argparse.
https://hg.python.org/cpython/rev/97b28115bd08

There's a bug in this patch, first reported at https://github.com/python/benchmarks/issues/1

Attached patch causes test_parse_args_abbreviation to show the failure:

test.test_argparse.ArgumentParserError: ('SystemExit', '', "usage: PROG [-h] [--foo] bar {1,2,3,lost,long,longer} ...\nPROG: error: ambiguous choice: 'long' could match long, longer\n")

I haven't read the discussion in full, but it looks like this patch was added without any recent discussion or testing.

Previously if we had `choices=['a','abc']`, any exact match would be accepted, and partial matches rejected.  With this change the only choice that will work is the longest, 'abc'.  The other will be rejected as ambiguous.

In contrast with optionals flags, an exact match has priority over (possibly ambiguous) abbreviations.  See parser._parse_optional

I also noticed when testing this that for regular 'choices', the abbreviated string is put in the Namespace.  

     add_argument('--foo', choices = ['one','two'])

can produce `Namespace(foo='on')`.  I suspect most developers would want `Namespace(foo='one')`, regardless of whether it was an exact match or partial.  You don't list choices if you are ok with partial matches.

This isn't a problem with `subparsers`, since an abbreviation match still invokes the right subparser, and even puts the right name in 'subparsers' dest.  But for regular choices the behavior is highly debatable.

This needs to be reverted.  It may be fixable, but it needs more testing and discussion.  For now it's an enhancement that is causing backward compatibility problems.

ps

I missed this issue when I made an effort to find all argparse issues several years ago.  I contributed to https://bugs.python.org/issue14365, the other abbreviations issue mentioned in the recent github thread.

Another failure case:

parser.add_argument('--int', type=int, choices=[10,15,25])

'--int 15` puts 'int=15' in the Namespace.

'--int 2' puts 'int=2' in the Namespace, because it matches the 'str(25)'.

'--int 1' raises an error in the ambiguity error formatting code: 

     'choices': ', '.join(ac)}

This last error could be corrected with:

    'choices': ', '.join(map(repr, ac))

=================

The initial message for this issue stated:

    So that this feature is not foisted on every programmer that uses  
    argparse it could be made an option to the ArgumentParser class with 
    the (possible) additional refinement of specifying a minimum number 
    of characters for the abbreviations.

That was ignored in the subsequent discussion and patches.

Other Choices issues

http://bugs.python.org/issue16468 - argparse only supports iterable choices
http://bugs.python.org/issue16418 - long choices
http://bugs.python.org/issue16977 - choices='abc' case

I recommend closing this issue, and depend on aliases for subparsers abbreviations.  The interaction with 'choices' is just too complicated to handle as proposed here.

Mean culpa - I have  jumped the gun on this. Sorry to all for the inconvenience. I have just got back after being away for a few days, and will revert the change shortly, if no one beats me to it.

*Mea culpa. Autocorrect :-(

New changeset 3310ea7dbe30 by Vinay Sajip in branch 'default':
Issue #12713: reverted fix pending further discussion.
https://hg.python.org/cpython/rev/3310ea7dbe30