Re: Scrolling commands and skipping redisplay, was: Re: emacs rendering comparisson between emacs23 and emacs26.3

[Dmitry Gutov]

On 17.04.2020 09:11, Eli Zaretskii wrote:

> > > I think you will find that many more do than you seem to assume.  Even
> > > just redisplaying a window does this in many cases (to find the proper
> > > place for window-start position).  C-n and C-p do as well.  And when
> > > scroll-conservatively is in effect, almost every command that moves
> > > point does.
> >
> > Would that be... almost all of them?
>
> I don't think I understand the question.  Please elaborate.

I think the majority of commands either move point, or have the 
potential to do so. Even those that do completing-read, for example.

> > (Indeed, the fact that (setq scroll-conservatively 1) doesn't work as
> > well as one would expect is a real bug IMO).
>
> I'm not aware of any bugs in that, so please describe what you think
> is a bug in more detail.

1. (setq scroll-conservatively 1)

2. Visit xdisp.c.

3. Lean on C-n.

4. From time to time, you will see point jump into the middle of the 
screen, when font-lock takes too long, apparently (or maybe it 
corresponds to GC pauses).

This shouldn't happen because next-line only moves by one line.

> > In any case, if there is a way to dynamically detect these cases and
> > disable redisplay skipping for them, I'd like to try that out.
>
> Yes, jit-lock-defer-time is that way.  But you already know that.

No, it speeds up redisplay. It doesn't make *sure* that redisplay is not 
skipped.

To do that, need to modify the condition of skipping redisplay.

> If that's not what you meant, then what exactly would you like to be
> able to detect?  The move_it_* functions are the entry point to the
> display simulation code, and they are used all over the place.

Perhaps detect the cases when move_it_* was used inside the command's 
implementation?

> > Even more if there was a way to put the results of "simulated redisplay"
> > to use in the "real" redisplay later.
>
> That happens automatically, since the 'fontified' property they
> produce is left on the buffer text.

That might be sufficient if we assume that font-lock is always the 
slowest part of redisplay (for some commands it might not always be 
true, but I'm guessing here).

But with that assumption, the extra condition of not skipping redisplay 
could be simplified (?) to only checking whether the current screen has 
been fontified already (i.e. 'fontified' is non-nil). If it's fontified, 
we won't skip redisplay.