How can I visualize per-character differences in a unified diff file?

Question

Say I get a patch created with git format-patch. The file is basically a unified diff with some metadata. If I open the file in Vim, I can see which lines have been modified, but I cannot see which characters in the changed lines differ. Does anyone know a way (in Vim, or some other free software that runs on Ubuntu) to visualize per-character differences?

A counter example where per-character diff is visualized is when executing vimdiff a b.

update Fri Nov 12 22:36:23 UTC 2010

diffpatch is helpful for the scenario where you're working with a single file.

update Thu Jun 16 17:56:10 UTC 2016

Check out diff-highlight in git 2.9. This script does exactly what I was originally seeking.

Solution 1

Given your references to Vim in the question, I'm not sure if this is the answer you want :) but Emacs can do this. Open the file containing the diff, make sure that you're in diff-mode (if the file is named foo.diff or foo.patch this happens automatically; otherwise type M-x diff-mode RET), go to the hunk you are interested in and hit C-c C-b for refine-hunk. Or step through the file one hunk at a time with M-n; that will do the refining automatically.

Solution 2

In git, you can merge without committing. Merge your patch first, then do:

git diff --word-diff-regex=.

Note the dot after the equals sign.

Solution 3

Here are some versions with less noisy output than git diff --word-diff-regex=<re> and that require less typing than, but are equivalent to, git diff --color-words --word-diff-regex=<re>.

Simple (does highlight space changes):

git diff --color-words

Simple (highlights individual character changes; does not highlight space changes):

git diff --color-words=.

More complex (does highlight space changes):

git diff --color-words='[^[:space:]]|([[:alnum:]]|UTF_8_GUARD)+'

In general:

git diff --color-words=<re>

where <re> is a regexp defining "words" for the purpose of identifying changes.

These are less noisy in that they color the changed "words", whereas using just --word-diff-regex=<re> surrounds matched "words" with colored -/+ markers.

Solution 4

git diff --color-words="[^[:space:]]|([[:alnum:]]|UTF_8_GUARD)+"

The above regex (from Thomas Rast) does a decent job of separating diff fragments at the punctuation/character level (while not being as noisy as --word-diff-regex=.).

I posted a screenshot of the resulting output here.

---

Update:

This article has some great suggestions. Specifically, the contrib/ tree of the git repo has a diff-highlight perl script that shows fine-grained highlights.

Quick start to use it:

$ curl https://git.kernel.org/cgit/git/git.git/plain/contrib/diff-highlight/diff-highlight > diff-highlight
$ chmod u+x diff-highlight
$ git diff --color=always HEAD~10 | diff-highlight | less -R

Solution 5

If you have nothing against installing NodeJS, there's a package called "diff-so-fancy" (https://github.com/so-fancy/diff-so-fancy), which is very easy to install and works perfectly:

npm install -g diff-so-fancy
git diff --color | diff-so-fancy | less -R

Edit: Just found out it's actually a wrapper for the official diff-highlight... At least it's easier to install for perlophobes like me and the GitHub page is nicely documented :)

View more solutions

Comments

[Daenyth]

This might be better on superuser.com

[Adam Monsen]

Perhaps. I chose stackoverflow.com since the FAQ mentions this is the place for questions about "software tools commonly used by programmers"

[Mark Longair]

I'm not sure that this directly answers your question, but git diff --color-words is very useful for just seeing what words have change within lines, rather than the usual unified diff output. It is word-based rather than character-based, though, so if there's not much whitespace in the content you're diffing then the output may be less neat. (Edited: Oops, I see that I misunderstood what you're asking for - nevertheless maybe this comment would be useful to someone.)

[ks1322]

Related stackoverflow.com/q/49278577/72178

[Adam Monsen]

wdiff is interesting, thanks! To clarify my original question, I'm looking for something that provides enhanced syntax highlighting for a single file that happens to be in unified diff format.

[Adam Monsen]

Works for me! Heh, I've used Vim for 10 years, but I just installed emacs. :)

[Aleksander Adamowski]

Slightly offtopic (about word-for-word diffs, not enhancing a preexisting diff output), but I've found the following combinations best for word-for-word visualizations: * wdiff old_file new_file | cdiff * vimdiff , then inside vim :windo wincmd K in order to switch to vertical window layout (one below the other) from the side by side one. That layout is much better for files with long lines.

[Aleksander Adamowski]

BTW, Some other tools worth checking out, not mentioned in the linked article: wdiff2, mdiff, and the Google's online tool.

[yingted]

@Sairam Thanks, it's fixed. +1 for noticing it.

[Eddified]

You can shorten it to --color-words=[^[:space:]]|([[:alnum:]]|UTF_8_GUARD)+'

[gcb]

i had to add ' to the beginning of the value there. otherwise i got an error. Also, i simply using --color-words i get the exact same behaviour as using that regexp.

[Justin M. Keyes]

@gcb what text did you test it on to get the "exact same behavior"?

[gcb]

@JustinM.Keyes a 4 line changes in a source file. nothing weird. javascript i think. Ascii.

[Justin M. Keyes]

@gcb The text content matters. If your changes are separated by whitespace, there's no difference. But if you change if you change something like foo.bar to foo.qux you will see the difference.

[ntc2]

Better: git diff --color-words=..

[ntc2]

Simpler: git diff --color-words='[^[:space:]]|([[:alnum:]]|UTF_8_GUARD)+'.

[Justin M. Keyes]

@ntc2 Cool, thanks. Any idea if that is backwards compatible with say git 1.7.x?

[ntc2]

No idea when this feature was introduced; saw it when reading man git diff to understand what your answer was doing. Also, looks like my point is the same as @Eddfied's in the first comment :P

[alxndr]

I wanna give a shoutout to @ntc2, that is exactly what I was looking for when googling "git diff by character"!

[Tyler Collier]

@ntc2 You should make your comment an answer.

[Tyler Collier]

I myself like --color-words, without the =. part.

[Adam Monsen]

Upvoters please note, my original use case assumes you only have a patch file, no git repo or even base/modified versions. That's why I accepted @legoscia's answer... it describes exactly what was requested.

[n.r.]

git diff --color-words='\w' would work better with diacritics (git v1.7.10.4)

[abhisekp]

@ntc2 git diff --color-words=. and git diff --color-words . works differently. Better is git diff --color-words ..

[ntc2]

@abhisekp: what's the difference?

[abhisekp]

@ntc2 Here you go i.imgur.com/Fa8vCtO.png But you're right about Better: git diff --color-words=.

[ntc2]

@abhisekp: thanks for the pic. I think I figured it out: the git diff --color-words . is really the same as git diff --color-words -- .! I.e., the . is interpreted as a path. You can verify with mkdir x y; echo foo > x/test; git add x/test; git commit -m test; echo boo > x/test; cd y; git diff --color-words=.; git diff --color-words .; git diff --color-words -- ..

[abhisekp]

@ntc2 yes. That's a path. I got it. But I liked the way you presented the whole thing in a single like. Nice. +1

[Tobias Kienzler]

Your more complex version works great. I appended --word-diff=plain to additionally have [- and -] surround deletions and {+ and +} surround additions. As the manual warns, though, actual occurrences of these delimiters in the source are not escaped in any way

[Tobias Kienzler]

Your more complex version unfortunately doesn't seem to highlight e.g. indentation changes, I've opened a question on this

[WoLfPwNeR]

This answer is great! However is there a way to actually change the background of those changes to green/red?

[ntc2]

@WoLfPwNeR: if you're asking how to make git diff change the background colors instead of the foreground colors, this might help: github.com/git/git/tree/master/contrib/diff-highlight; I have not tried it myself.

[WoLfPwNeR]

@ntc2 Looks like it doesn't work with --color-words, but otherwise it looks nice. I'll stick with line diff then.

[Aaron Tribou]

I'm getting a strange newline issue using this vs just --color-words=.. On some lines using this regex, the diff-ed character (in my case, a # comment character that was removed in a YAML file) jumps to the beginning of the line and is no longer indented correctly. Using --color-words=. appears to show the correct indentation.

[Hatshepsut]

Thanks for this great answer! Is it possible to hide lines without differences from the output?

[ntc2]

@Hatshepsut: try adding -U0?

[Eugen Konkov]

More complex re does not highlight space changes

[rofrol]

diff-hightligh needs to be made with make atm. github.com/Homebrew/homebrew-core/issues/…

[Vitaly Zdanevich]

No solution for git commit -p?

[Vitaly Zdanevich]

URL is not available.

[Hi-Angel]

Works with git log too, but geez this is ugly. It surrounds differences with square brackets, and confuses whether these are part of the code or not.

[Hi-Angel]

But emacs doesn't support reading from stdin, I can't do e.g. git log master.. -p | emacs -

[legoscia]

@Hi-Angel You could open Emacs and type M-! to run the command and capture the output in a buffer.

[Ashutosh Jindal]

I had installed git with Homebrew and already had that script at /usr/local/share/git-core/contrib/diff-highlight/diff-highli‌​ght . This seems to suggest that Homebrew's git does install the entire contrib in /usr/local/share/git-core/contrib/. So finally, the following worked for me git diff --color=always | /usr/local/share/git-core/contrib/diff-highlight/diff-highli‌​ght

[Gabriel Staples]

@ntc2, git diff --color-words=., or similar, must be the git basis by which meld does it's char-by-char highlighting magic (see a screenshot of meld in my answer here). I've been wondering abou that for years.

[Hritik]

If you wanted to see changes in last commit, git show --color-words=.