bug#56682: locked narrowing

[Gregory Heytings]

> > Consider this other scenario.  A buffer is buried, and modified in the 
> > background by some process.  That process adds a long line in that 
> > buffer, followed by enough non-long lines, with point after these 
> > non-long lines. Now the user asks for that buffer to be displayed 
> > again.  With your proposed approach, we would search around point if 
> > there are long lines. There aren't, so we do not activate long line 
> > optimizations.  Now the user scrolls the buffer.  The buffer has not 
> > changed in any way, so the detection is not triggered, and we reach the 
> > long line.  Kaboom!
>
> No "Kaboom!": when the user scrolls the buffer, point moves, so we can 
> detect that and trigger rescanning.

In which case we are back to the alternative solution, which is to trigger 
rescanning on each redisplay cycle.  Which might, after all, be the 
simplest and safest solution.