D3212: patch: implement a new worddiff algorithm

[quark (Jun Wu)]

quark added a comment.


  Git first had a `contrib/diff-highlight/diff-highlight` script which inverts foreground/background for hunks with len(deleted_lines) = len(inserted_lines).
  
  Then the latest version shows diff inline. That is:
  
    common words [+inserted words with green color][-deleted words with red color] common words
  
  I dislike that, since it could be ambiguous ("[+ x]" could be part of the original text).
  
  For colors, it's really a hard question. I think if we can detect "dim" is unsupported and make it a no-op, then it'd be fine to use. `infocmp` can report "dim" correctly on my Linux terminal. Internally, we patched color.py to use 16 and 256 colors even if terminfo reports 8 colors. But I guess that's not an acceptable solution here.

INLINE COMMENTS

> yuja wrote in patch.py:2536
> Nit: maybe we can sort out tokens here instead of re-parsing tabs, newlines, trailing whitespaces later.
> 
> But I'm not sure if that will make things simpler.

For a split list `['\n', '\t', ' ']`, mdiff might return a hunk that joins them. So `''.join(al[a1:a2])` will become more complex.

I think having mdiff step free from EOL/tab handling makes the code easier to read.

REPOSITORY
  rHG Mercurial

REVISION DETAIL
  https://phab.mercurial-scm.org/D3212

To: quark, #hg-reviewers, durin42, yuja
Cc: yuja, spectral, mercurial-devel