Re: Suggestion for ediff

[Stefan Monnier]

>> It'd probably be best to layer the diffs: first do a word-granularity diff
>> and then do a char-granularity diff within words.

> That's my opinion too.

>> But maybe that'd be too slow.

> This is why I was proposing just looking at the beginning and end of
> each word, which is two comparison if the words are actually
> different, and linear in the number of common chars otherwise.

I see what you meant, then.  Sorry for being a bit slow.  Now that I think
about it, it may be possible to make char-level diffs on words reasonably
fast, by running char-level diff only once (rather than once per word),
basically putting all the words-to-diff into two files, separated by
sufficiently large and unique text to make sure that diff won't get
out-of-sync.


        Stefan