Re: [Emacs-diffs] trunk r116461: Connect electric-indent-mode up with CC Mode. Bug #15478.

[Alan Mackenzie]

Hello, Dmitry.

On Thu, Mar 27, 2014 at 10:02:49AM +0200, Dmitry Gutov wrote:
> On 26.03.2014 22:53, Alan Mackenzie wrote:

> > Indentation (in the sense of what `indent-line-function' does) only makes
> > sense in programming (etc.) modes.  In text modes, and the like,
> > indentation is, in practice, done with adaptive fill prefices.

> This could be considered a reason to improve the indent-line-function in 
> text-mode. `indent-relative' offers behavior that's pretty close. Maybe 
> it could be made to follow the behavior of auto-fill even closer.

Notice, here, how we're no longer talking about electric indentation, but
rather about newline-and-indent.  The two topics are distinct.

> > Having RET do `newline-and-indent' in Emacs Lisp Mode while doing
> > `newline' in Text Mode makes a lot of sense to me.  A trickier question
> > is to identify which of the non-programming modes really want this sort
> > of indentation.

> I'd rather make exceptions for specific "non-programming modes", where 
> indentation of the next line is really hard to guess.

Yes, but how?  fundamental-mode is a non-programming mode, so the global
key map needs RET set up for newline, C-j for newline-and-indent.  That
leaves lots of mode key maps to be set up.  At this point, your
suggestion and mine become the same.

> Many modes that don't inherit from prog-mode have something to do with 
> structured content, and often define their own specific indentation 
> functions (sgml-mode, markdown-mode inherit from text-mode, css-mode and 
> yaml-mode inherit from fundamental-mode).

> I'd really expect typing <div> and pressing RET in html-mode to offer 
> hard +2 indentation on the next line, but I wouldn't call it a 
> programming mode.

Yes.

> In Markdown, I'm often typing code blocks, and I expect RET to bring me 
> to the column which the previous line was indented to, so I don't have 
> to press TAB each time. And if I'm outside of a code block, the lines 
> usually either have no indentation (then indent-relative indents to the 
> 0th column as well), or they serve as continuation of a paragraph, and I 
> want each next line to have the same extra indentation until the 
> paragraph ends (and indent-relative does that well enough).

I usually think of html, markdown, and such like, as the "etc." in
"programming modes (etc.)".

> The last time I used text-mode, it was for a similar purpose (a couple 
> of code blocks, and the rest of the text is indented to column 0).

> `M-x fill-paragraph' would slaughter the code blocks, and it wouldn't 
> improve the indentation anywhere else.

If you edit the non-code blocks a lot in text mode, `fill-paragraph' is
_exactly_ what's wanted to restore the filling.  I think, in text mode,
M-q preserves existing indentation.

> >> There's a certain class of users who've been binding RET to
> >> `newline-and-indent' for a long time (myself included), and I
> >> haven't seen anyone mention only doing that in prog-mode, instead of
> >> globally.

> > That was what we collectively decided last Autumn when the topic came up.

> Okay: I haven't seen anyone mention doing it outside of emacs-devel.

> > Richard Stallman alluded to it in his disgust at bug #16156, when what
> > was bound to RET at the time zapped his indentation.

> > I personally would not be unhappy at leaving the traditional binding in
> > place for RET and C-j, but wouldn't mind them swapping in "indenting"
> > modes.  I'd object strongly to RET in text mode messing around with
> > indentation.

> Maybe text-mode by itself should be a special case. I don't use it often 
> enough to have a strong opinion.

I think RET should do the most natural sort of newline, and C-j the
subsidiary one, whatever they may happen to be for a particular mode.

> >> How hard can it be for a user to change the key bindings without a
> >> mode?

> > Middling, not very.  Such a minor mode might serve to damp down the
> > inevitable complaints the change in defaults will provoke.

> As long as this new mode is divorced from electric-indent-mode, I'd be 
> happy.

This is a key point.

> > By "makes sense" I think you mean "seems a sensible thing to do".  I take
> > issue with you here and say that, in Text Mode, it's a bizarre thing to
> > do.  I can't think of any normal circumstances where this behaviour would
> > be commonly desired; just how often in Text Mode do you not want RET just
> > to insert a new line when you type it at the beginning of a line?

> This specific behavior is a consequence of using `newline-and-indent'. 

No, not at all.  It's a consequence of electric behaviour getting
entangled with newline-and-indent.  Electric indentation doesn't belong
in Text Mode.  It's useful only where the indentation of a line of code
can be changed by what's in the line.

> One answer may be "Don't want that? Use `open-line'", which I sometimes 
> do, but special-casing indent-line-function in text-mode not to reindent 
> on the first invocation (when point is at bol followed by whitespace and 
> then non-whitespace on the same line) could well be another option.

;-)  That would be papering over the cracks.  Banishing electric
indentation from major modes where it's silly is what we really want.

> What makes sense to me, is using `newline-and-indent' itself. Richard 
> doesn't like

> foo
> |  bar

> turning into

> foo

> |bar

> I can understand that, but I don't like

> foo
>    bar|baz

> turning into

> foo
>    bar
> |baz  (note the missing indentation)

> Which situation do you think occurs more frequently?

The two are completely non-competing situations.  RMS's happened because
electric indentation was active where it shouldn't be.  Your situation is
a matter of binding (RET C-j) to ('newline 'newline-and-indent) the
appropriate way round.  Don't confuse these.

I think Text Mode is about typing natural language text in paragraphs.
The "indentation" will be the adaptive fill prefix inserted by
auto-fill-mode.  Normally, you'll be typing RET to start a new paragraph,
and this will either be at column 0, or at some column not dependent on
the previous text.  newline-and-indent isn't the Right Thing here.

-- 
Alan Mackenzie (Nuremberg, Germany).