Add FlagAction to argparse

[ericvsmith]

BPO	8538
Nosy	@vstinner, @ericvsmith, @merwok, @berkerpeksag, @matrixise, @wm75, @remilapeyre, @kenahoo
PRs	python/cpython#11478 python/cpython#11478 python/cpython#11478
Files	argparse_bool.py

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields: ```python assignee = None closed_at = created_at = labels = ['easy', 'type-feature', 'library', '3.9'] title = 'Add FlagAction to argparse' updated_at = user = 'https://github.com/ericvsmith' ``` bugs.python.org fields: ```python activity = actor = 'matrixise' assignee = 'none' closed = True closed_date = closer = 'matrixise' components = ['Library (Lib)'] creation = creator = 'eric.smith' dependencies = [] files = ['17100'] hgrepos = [] issue_num = 8538 keywords = ['patch', 'patch', 'patch', 'easy'] message_count = 19.0 messages = ['104259', '105032', '105620', '107758', '110610', '113479', '121530', '166165', '270269', '332621', '332625', '332627', '332628', '333022', '333071', '333325', '352259', '352277', '352278'] nosy_count = 11.0 nosy_names = ['bethard', 'vstinner', 'eric.smith', 'eric.araujo', 'Jeremiah.Jordan', 'berker.peksag', 'paul.j3', 'matrixise', 'wolma', 'remi.lapeyre', 'kenahoo'] pr_nums = ['11478', '11478', '11478'] priority = 'normal' resolution = 'fixed' stage = 'resolved' status = 'closed' superseder = None type = 'enhancement' url = 'https://bugs.python.org/issue8538' versions = ['Python 3.9'] ```

[ericvsmith]

From a python-dev email from Neal Becker, copied here so it won't get lost. ---------------------------- steven.bethard@gmail.com made a very nice module for me to enhance argparse called argparse_bool.py, which contains ConfigureAction. This will allow a boolean value to be set very like the gnu configure style:

--foo --with-foo --without-foo --no-foo --foo=yes --foo=no

I've been happily using it, and I think it would be of sufficient general interest to include it with the standard library.

[merwok]

Are we sure we want three or four ways to spell the same thing? --foo and --no-foo seem a useful couple to me, and --with-foo/--without-foo cater to different use cases. Perhaps it would be ok to have a SwitchAction (or FlagAction) for --thing/--no-hing and one ConfigureAction for the --with[out]-* style.

[aa3203a7-2d25-4619-bcf7-1a8f18a9cd6a]

I'm looking into making a patch from this for py3k, and have the following observations:

1. I agree with merwok, at the moment the monolithic ConfigureAction is a bit excessive for most uses (other than maybe emulating ./Configure...). I vote we split it to FlagAction and ConfigureAction.
2. I don't think this should be added as an argparse builtin/registered action (like action='store' or action='count'), but rather as a ready-made user-action in a separate namespace (from argparse.actions import FlagAction)

Given feedback that these two decisions seem sane, I'd be happy to produce a patch against recent py3k.

[merwok]

I think you can go ahead and produce a patch. Don’t start splitting argparse into a package now though, this is a sensible decision IMO but has to be agreed by the maintainer; just add FlagAction and ConfigureAction to argparse for now.

[merwok]

Some useful tips to make patches: http://www.python.org/dev/patches/

[5f77b749-174d-460c-805c-15f59f434233]

I think this should be updated so that nargs=0 is allowed, so that you can only do --foo/--no-foo and don't clutter up the help/interface with --foo [FOO] --no-foo=[FOO]

You can do this by adding nargs to the ConfigureAction.__init and passing that through to super, and then updating __call to check 'if value is None or value == []:' instead of the None.

[merwok]

I think FlagAction should implement strictly boolean options, that is --foo and --no-foo, without arguments at all.

For ConfigureAction, there is a precedent (unless I’m mistaken) in configure, which permits such things: --without-unicode --with-unicode=ucs4 --with-unicode (uses default value for arg)

I say we focus on the simple FlagAction for this bug and keep ConfigureAction for another patch.

Yaniv: Can you give us a status update?

[8955c213-fd54-471c-9758-9cc5f49074db]

On the off chance that someone was waiting for feedback from me, I'll say:

(1) A simple boolean --foo/--no-foo action seems useful to me. I would probably call it BooleanOptionalAction rather than FlagAction. (Almost anything could be considered a "flag".)

(2) At the moment, argparse doesn't supply any actions, so I don't think we should make a separate namespace for them. If it starts to grow a list of such actions, we can add a separate namespace later. (And given that the name will end with "Action", I think it should be pretty clear without the namespace.)

[vstinner]

I'm sorry but there is no activity since 4 years, so I guess that the feature is not common enough to require a builtin support in argparse.

[f2659757-e971-4d08-87ab-4c271cb73680]

@vstinner - I don't think that conclusion is correct, here is a very highly-upvoted answer on SO that indicates a lot of people are still looking for this:

https://stackoverflow.com/a/15008806/169947

I myself asked a related (more focused?) question where I was directed here:

https://stackoverflow.com/q/53937481/169947

I'm guessing the right thing to do now would be refocus the merge request in a new ticket - is this still the right tracker?

[7a064fe6-c535-4d80-a11f-a04ed39056c5]

Let me highlight something about

https://stackoverflow.com/a/15008806/169947

The original question was how to implement an Action that accepts 'True' or 'False' as an argument. Users often try type=bool, which doesn't work because of the normal behavior of the Python bool(astr) function. That's been the subject of several other bug/issues.

https://bugs.python.org/issue14392 https://bugs.python.org/issue26994 https://bugs.python.org/issue24754 https://bugs.python.org/issue21208

My answer in that SO question is

https://stackoverflow.com/a/19233287/901925

----

@mgilson's answer proposes a '--foo', '--no-foo' alternative. That is in line with this bug/issue.

    parser.add_argument('--feature', dest='feature', action='store_true')
    parser.add_argument('--no-feature', dest='feature', action='store_false')
    parser.set_defaults(feature=True)

So the question here is whether mgilson's simple answer is enough, or do we need to add Eric's ConfigureAction class?

On a casual reading the patch proposed here shouldn't have backward compatibility issues, since it is an addon class, and doesn't modify existing classes. But it lacks tests and documentation.

Documentation for argparse is a tough issue. While advanced users want more features and more documented details, most of the SO questions come from beginners, who's eyes glaze over when they read the existing documentation.

[ericvsmith]

Yes, this is the correct bug tracker.

And note that this code isn't mine, I just posted it here so it wouldn't be lost. It looks like the original message was from https://mail.python.org/pipermail/python-dev/2010-April/099704.html

[f2659757-e971-4d08-87ab-4c271cb73680]

Thanks, Paul and Eric, for your very quick replies.

You're quite correct, the original question in https://stackoverflow.com/a/15008806/169947 is indeed hoping for --foo TRUE and --foo False etc. to work. Personally I don't like that as much as the GNU-style --foo and --no-foo technique, because when you let people type TRUE or True or T or 1, etc., it gets a bit confusing about exactly what is accepted as a true value, what's false, is a zero interpreted as 0 or "0", whether a failure to parse the value as True or False will be reported as an error or not, and so on. The user typically can't really know these answers without reading the actual code, or running it to see what generates an error (or triggers whatever behavior they're looking for), which is certainly not ideal.

By contrast, with --foo and --no-foo, the options are strict and clear. And supplying an argument will trigger a parse failure.

For @mgilson's proposal, I think the main thing I find unsatisfactory (besides the fact that it takes 3 lines to define, and I'll have to come back to that SO answer every time to make sure I get them right...) is that the --help output can't be made clear. With the following specification:

    parser.add_argument('--foo', dest='foo', help="Do foo", action='store_true')
    parser.add_argument('--no-foo', dest='foo', help="Don't foo", action='store_false')
    parser.set_defaults(foo=True)

we get the following --help text (when using ArgumentDefaultsHelpFormatter):

--foo       Do foo (default: True)
--no-foo    Don't foo (default: True)

and that last line seems to be a contradiction, or at least very confusing. The only alternative I see is to turn off ArgumentDefaultsHelpFormatter, but I think the defaults are generally helpful information for the user.

To fix that --help issue seems to require quite a bit more complicated registration of arguments, so it's probably not going to happen in most people's scripts.

I should be clear: I haven't vetted the argparse_bool.py proposal in detail either, so I'm not specifically asking for it to be adopted. Just hoping for a nice resolution to the --foo --no-foo issue that codifies best-practices in a way that makes it trivial to get nice behavior in scripts.

As for documentation - I had poked around in the code a bit and seen the register method, but I thought since it wasn't documented, I'd better not use it. Is it for public consumption? If so, I'd say it's better documented than undocumented, even if it provides more info than most people need. My guess is that enough people are probably using it to make it impossible to eliminate, which is a good test for whether something should be documented too.

[vstinner]

I reopen the issue since it seems like there are people requesting the feature. (I also removed myself from the issue, I'm not interested to implement it.)

[23982c60-ed6c-47d1-96c2-69d417bd81b3]

I also removed myself from the issue, I'm not interested to implement it.

I would like to try and implement the change. I will open a PR shortly.

[23982c60-ed6c-47d1-96c2-69d417bd81b3]

I made a first proposal for this feature in PR 11478, I had to adapt argparse.Action to customize how the action is represented in the usage string by adding a format_usage() method that would default to the current behavior if not overridden.

Other methods to do this may be better but I did not find them.

[matrixise]

I am also interested by this feature.

[matrixise]

New changeset 6a517c674907c195660fa9178a7b561de49cc721 by Stéphane Wirtel (Rémi Lapeyre) in branch 'master': bpo-8538: Add support for boolean actions to argparse (GH-11478) https://github.com/python/cpython/commit/6a517c674907c195660fa9178a7b561de49cc721

[matrixise]

I have merged the PR 11478 and this feature will be included into 3.9.

Thank you for your proposal of Remi and thank you to Eric for the idea of this feature.