GNU bug report logs - #76205
31.0.50; 31.0.50; python.el completion broken for non-builtin objects in IPython 8.30.0 & Python 3.13.2

Previous Next

Full log

Message #38 received at 76205 <at> debbugs.gnu.org (full text, mbox):

From: Visuwesh <visuweshm <at> gmail.com>
To: kobarity <kobarity <at> gmail.com>
Cc: Liu Hui <liuhui1610 <at> gmail.com>, Eli Zaretskii <eliz <at> gnu.org>,
 76205 <at> debbugs.gnu.org
Subject: Re: bug#76205: 31.0.50; 31.0.50; python.el completion broken for
 non-builtin objects in IPython 8.30.0 & Python 3.13.2
Date: Sun, 16 Feb 2025 19:32:36 +0530

[ஞாயிறு பிப்ரவரி 16, 2025] kobarity wrote:

> Visuwesh wrote:
>> 
>> [சனி பிப்ரவரி 15, 2025] kobarity wrote:
>> 
>> >> [...]
>> >> So...what do we do from here?  Unfortunately, I know next to nothing
>> >> about this area to file a sensible bug report myself.
>> >
>> > I'm not sure either.  As ipython --simple-prompt does not support
>> > completion, I don't know if using rlcompleter with ipython
>> > --simple-prompt is supported or not.
>> >
>> > The direct cause of this problem seems to be that rlcompleter.__main__
>> > is not as expected.  So the following steps enable completion on
>> > ipython --simple-prompt on Python 3.13.
>> >
>> > import readline
>> > import rlcompleter
>> > import __main__
>> > rlcompleter.__main__ = __main__
>> > readline.parse_and_bind("tab: complete")
>> >
>> > Attached is a test patch that does something similar to this.
>> 
>> I can confirm that completion works for non-builtin objects with this
>> patch.
>
> Thank you for testing the patch.
>
> Here's what's happening.  ipython --simple-prompt on Python 3.13 is
> indirectly importing rlcompleter, which sets its Completer().complete
> as the completer.  Completer() without the namespace argument uses
> __main__.__dict__ as the source of global symbols to complete.  The
> __main__ here is the name in rlcompleter.py namespace, which is bound
> at import time of rlcompleter.py.  However, __main__ used in the REPL
> of ipython --simple-prompt is different from __main__ used at the
> import time of rlcompleter.py.  Therefore, changes in globals()
> (i.e. __main__.__dict__) are not reflected in
> rlcompleter.__main__.__dict__ and the completion will fail.
>
> Prior to Python 3.12, the problem does not occur because rlcompleter
> was not imported indirectly.  When we import rlcompleter in the native
> completion setup code, __main__ is already the same as the one used in
> the REPL.
>
> Since IPython does not use rlcompleter, it seems difficult to consider
> this problem a bug in IPython.  It would be better if rlcompleter
> reads sys.modules["__main__"] instead of __main__, but it would not be
> easy to change a long standing specification.  So I think it would be
> better for python.el to address this issue.
>
> To set up a completer that works correctly, it is sufficient to
> specify the namespace argument:
>
> readline.set_completer(rlcompleter.Completer(globals()).complete)
>
> However, we need to make sure that we do not override the customized
> completer.  The problem is that it is difficult to distinguish between
> the completer that is set up by importing rlcompleter and the
> customized completer.
>
> So it seems to me that it would be reasonable to override
> rlcompleter.__main__ so that the default completer works properly.  I
> think the attached revised patch is slightly better than my previous
> patch, since it does not add "__main__" to globals.  This change
> affects only IPython.  I added some comments and the commit message.

Thanks for the explanation.  The attached patch works on my end too.

> I would like to hear your opinions.

I do not have background knowledge on these parts so I hold no technical
opinions.  Please go ahead with a solution that you're satisfied with.
I am happy as long as the completion works as expected!

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.