Re: [PATCH 3/4] diff-highlight: match up lines before highlighting

[Jonathan Lebon <jonathan.lebon@gmail.com>]

On Tue, Nov 3, 2015 at 5:03 PM, Jeff King <peff@peff.net> wrote:
> Your is _much_ slower. I get:
>
>   real    0m25.538s
>   user    0m25.420s
>   sys     0m0.120s
>
> for the old versus:
>
>   real    2m3.580s
>   user    2m3.548s
>   sys     0m0.156s

Thanks for investigating and trying it out. I got the same results
here as well.

> for your series. In an interactive setting, the latency may not be that
> noticeable, but if you are digging far into history (e.g., "git log -p",
> then using "/" in less to search for a commit or some test), I suspect
> it would be very noticeable.

Agreed.

> I was thinking there was some low-hanging fruit in memoizing the
> calculations, but even the prefix/suffix computation is pairwise. I'm
> not really sure how to make this much faster.

I gave memoization a try to see if it could improve the situation. I also
lowered maxhunksize to 10. Doing `git log -p` on git.git went from 2m31
to 2m11. So I think it would require a whole other approach overall.

> As for the output itself, the diff between the two looks promising. The
> first several cases I looked at ar strict improvements. Some of them are
> kind of weird, especially in English text.

Yes, I'm very happy with the improvements and run with these patches all
the time for now.

> In the other thread I mentioned earlier, the solution I cooked up was
> dropping highlighting entirely for hunks over a certain percentage of
> highlighting. I wonder if we could do something similar here (e.g.,
> don't match lines where more than 50% of the line would be highlighted).

I looked over but haven't tested the patches in the other thread yet. But
overall, the LCS definitely looks promising. I'm hoping to find some time
to have a more serious go at it and maybe pick it up where you left off.

>
> -Peff

Thanks again for reviewing these patches and apologies for the delayed
reply.

Jonathan