[Nano-devel] trading speed against correctness

[Benno Schulenberg]

So... what do others think about this tradeoff?  Should
the patch be applied so that syntax colouring is always
recalculated and correct upon every single key stroke?
Or is that too much of a cost?

Or should we make the recalculation conditional?  For
example, up to ten thousand lines always recalculate,
but beyond that let it go, give up correctness?


On Thu, May 21, 2015, at 20:40, Alexandre Detiste wrote:
> Le mercredi 20 mai 2015, 13:15:58 Benno Schulenberg a écrit :
> > Without the recalculation patch it gives me this:
> > 
> > Function 805ef96 took 17 milliseconds.
> > Function 804d564 took 21 milliseconds.
> > 
> > *With* the recalculation patch this:
> > 
> > Function 805efba took 840 milliseconds.
> > Function 804d57c took 848 milliseconds.
> > 
> > That is on a C file of a 185 thousand lines.
> > Quite a considerable difference.
> > 
> > How do you fare with your 30 thousand lines of Python?
> 
> 400 milliseconds for both insert and removal with the patch,
> 
> 2ms per insert and 12ms per removal without the patch;
> all of this with a very small deviation from the mean.

Benno

-- 
http://www.fastmail.com - Access your email from home and the web