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

[Dmitry Gutov]

On 09.04.2020 22:16, Stefan Monnier wrote:
> > Thanks. I tried it (and mentioned in the parent thread), but it seems like
> > in addition to the downsides of f-b-i-s (imprecise scrolling, which
> > apparently makes it a bad candidate for default behavior) it adds some
> > extra flickering from time to time (I figured because some redisplays
> > occurred before the idle timer had the chance to run). It's not as
> > obvious in an 'emacs -Q' session, but with my custom config it's more
> > apparent. And that flickering occurs the most while I'm flooring C-v.
>
> I think the flickering is the direct effect of the defer'd jit-lock:

It's weird, though, that it's not so obvious with the 'emacs -Q' build.

> you
> first see the non-jit-lock'd text and only after that the idle timer
> runs jit-lock and updates the display accordingly.

And that it happens in the middle of scrolling. All the while C-v is 
still pressed.

And also: jit-lock-function supposedly has a provision for that sort of 
circumstances: jit-lock-defer-time=0 specially, doing fontification 
immediately, as long as input-pending-p=>nil.

> > Is it possible that the conditions for skipping redisplay might return
> > false while input-pending-p returns t?
>
> Very much so, yes: the tests are sufficiently different and the possible
> circumstances sufficiently varied that it's almost unavoidable.
>
> E.g. usually the redisplay is skipped based on the presence of input at
> the *beginning* of the command, whereas the `jit-lock-defer-time` thingy
> samples the input later, when jit-lock is called.

The condition for skipping redisplay based in the presence of inputs 
before and after, to be a tad more accurate. But the difference is 
there, to be sure.

So thus far fast-but-imprecise-scrolling seems to provide better 
behavior. I would love to remove that variable, though, and just tell 
people to use jit-lock-defer-time *if* the jit-lock behavior can be 
improved.