Re: [PATCH] diff: Support diff.color-words config option

> Johannes Schindelin <Johannes.Schindelin@gmx.de> writes:
> 
> >> By the way, wouldn't it make sense to optimize the precontext of that 
> >> hunk by doing _something_ like:
> >> 
> >> 	if (!o->word_regex && strcmp(one->path, two->path))
> >>         	o->word_regex = userdiff_word_regex(two);
> >> 
> >> "Something like" comes from special cases like /dev/null for new/deleted
> >> files, etc.
> >
> > You mean to avoid the cost of initializing the regex in case one and the 
> > same file is diffed against itself?
> 
> No.
> 
> What I meant is much simpler than that.
> 
> If one and two are the same filename, and earlier gitattributes lookup for
> the path already failed to produce any when you checked one, isn't it very
> likely that the gitattributes lookup for two would fail the same way to
> produce any result?