[Python-Dev] python 2.7.9 regression in argparse?

[Terry Reedy]

On 1/6/2015 7:39 AM, Victor Stinner wrote:
> More context:
>
> 2014-12-19 12:43 GMT+01:00 anatoly techtonik <techtonik at gmail.com>:
>> https://github.com/nickstenning/honcho/pull/121
>
> The link mentions the following changeset:
> ---
> changeset:   93122:1a3143752db2
> branch:      2.7
> parent:      93112:927cca0b9337
> user:        R David Murray <rdmurray at bitdance.com>
> date:        Fri Oct 17 20:07:08 2014 -0400
> files:       Lib/argparse.py Lib/test/test_argparse.py Misc/NEWS
> description:
> #9351: set_defaults on subparser is no longer ignored if set on parent.
>
> Before, if a default was set on the parent parser, any default for that
> variable set via set_defaults on a subparser would be ignored.  Now
> the subparser set_defaults is honored.
>
> Patch by Jyrki Pullianinen.
>
>
> diff -r 927cca0b9337 -r 1a3143752db2 Lib/argparse.py
> --- a/Lib/argparse.py   Fri Oct 17 16:20:15 2014 -0500
> +++ b/Lib/argparse.py   Fri Oct 17 20:07:08 2014 -0400
> @@ -1089,7 +1089,14 @@ class _SubParsersAction(Action):
>           # parse all the remaining options into the namespace
>           # store any unrecognized options on the object, so that the top
>           # level parser can decide what to do with them
> -        namespace, arg_strings = parser.parse_known_args(arg_strings,
> namespace)
> +
> +        # In case this subparser defines new defaults, we parse them
> +        # in a new namespace object and then update the original
> +        # namespace for the relevant parts.
> +        subnamespace, arg_strings = parser.parse_known_args(arg_strings, None)
> +        for key, value in vars(subnamespace).items():
> +            setattr(namespace, key, value)
> +
>           if arg_strings:
>               vars(namespace).setdefault(_UNRECOGNIZED_ARGS_ATTR, [])
>               getattr(namespace, _UNRECOGNIZED_ARGS_ATTR).extend(arg_strings)
> ---
>
> Which is related to http://bugs.python.org/issue9351
>
> Maybe argparse just became more strict? I don't understand the issue.

Steven Bethard, the argparse maintainer, defined the old behavior of 
ignoring subparser defaults (where there are also top level defaults) as 
a bug "counter to what people probably expect".  If the old behavior had 
been documented, changing it in a bug-fix release would have been a 
mistake.  But looking at the patch, the doc seems to have been silent on 
the issue.

This is not the first time someone considered a 'bug fix' to be a 
'regression', which it might be from their viewpoint.  The last comment 
on the github thread suggests that an easy fix was found.

-- 
Terry Jan Reedy