bug#24766: 26.0.50: [PATCH] Confusing behaviour for indent-relative-maybe

[Eli Zaretskii <eliz@gnu.org>]

> From: Alex <agrambot@gmail.com>
> Cc: 24766@debbugs.gnu.org
> Date: Sat, 22 Oct 2016 15:49:57 -0600
> 
> Eli Zaretskii <eliz@gnu.org> writes:
> 
> >> I took UNINTENDED-OK to mean that "if non-nil, nothing is done in the
> >> case that there are no appropriate indentation positions. If there are
> >> appropriate indentation positions, then it should indent as usual."
> >> 
> >> The docstring could be improved to state that.
> >
> > Definitely.  Would you like to give it a try?
> 
> Sure. That would describe the behaviour that my diff brings -- does that
> mean that you're okay with the proposed change?

No, I meant to improve the doc string to describe the current
behavior.

> >> The optional argument is implicitly mentioned as "unless
> >> this command is invoked with a numeric argument, in which case it
> >> does nothing."
> >
> > Well, that's not how we document such functions, right?  Both the
> > effect of the argument, when used from Lisp, and the fact that it's
> > the prefix argument in interactive invocation, should be stated.
> 
> Indeed. Though it seems, according to commit
> 1fd63d9b9bc249488ec12a49cc3a576baf8c788a, that you were the one to
> document it. ;-)

Yes, but my change just described one subtlety of that function, it
didn't change anything about the rest.  Most probably I bumped into
the aspect of the behavior I described, and found it not documented at
all.

> > No, it's definitely not a no-op.  It is only a no-op if the previous
> > non-blank line has no white space at its beginning, or the current
> > column is already past that first indentation point.  IOW,
> > indent-relative-maybe only ever indents to the first indentation
> > point, and only when that indentation point is preceded by whitespace.
> 
> Oh, I see. Sorry about missing that. For some reason I was only testing
> lines that started with non-whitespace.
> 
> I feel more hesitant to change such old behaviour now, but I still think
> that it should be done. Here are the functions that call
> indent-relative-maybe or call indent-relative with an argument:
> 
> * add-change-log-entry
> * mh-letter-next-header-field-or-indent
> * A few functions in AUCTeX (ELPA)
> 
> I'm not sure if they should be changed, but if they should, a new
> function could be made to match the previous indent-relative-maybe
> behaviour:
> 
> (let ((first-indent (save-excursion
>                       (re-search-backward "^[^\n]")
>                       (backward-to-indentation 0))))
>   (when (< (current-column)
>            first-indent)
>     (indent-to first-indent)))

Like I said, I don't think the behavior should be changed, only the
documentation, which is somewhat misleading.  If we want some
different behavior, we could have a new function, or a new value of
the argument to indent-relative.

Thanks.