Ignore *all* whitespace changes with git-diff between commits

Question

I'm going through a codebase and fixing whitespace oddities and generally correcting indentation and such things, and I want to make sure I haven't inadvertently made any other changes, so I'm doing git diff -w to display differences in all changed files while ignoring whitespace differences. The problem is that this is not actually ignoring all whitespace differences—at least what I consider to be merely whitespace differences. For instance, in the following output from git diff -w,

-"Links":
-{
-
-    "Thermal":
-
-{
-
+  "Links": {
+    "Thermal": {

you can see that I've only

1. removed superfluous blank lines,
2. put curly braces on the end of the line of the key whose value they open, and
3. indented to fit the context

This question looked like it might offer an answer at first, but it deals with differences between two specific files, not between two specific commits. Everything else turned up by searching was a dead end as well. For instance, this question is about merging, not displaying differences, and this question deals with displaying word-level differences, and so forth.

Answer 1

Perhaps there is a better answer, but the best solution I've found so far is this.

First, you must control the definition of "whitespace" that Git is currently using.

git config core.whitespace '-trailing-space,-indent-with-non-tab,-tab-in-indent'

Next, you must control the definition of a word used. Instead of just using git diff -w, add --word-diff-regex='[^[:space:]]':

git diff -w --word-diff-regex='[^[:space:]]'

You'll still see the context, which (in my case, since I'm trying to ensure that there are no differences except whitespace differences) is not helpful. You can use -U0 to tell Git to give you 0 lines of context, like so,

git diff -w -U0 --word-diff-regex='[^[:space:]]'

but you'll still get output that looks pretty much like context, but it's still much better than looking through all the changes carefully and manually to make sure they are only whitespace changes.

You can also do all the above in one command. The -c flag changes git config just for one command.

git -c core.whitespace=-trailing-space,-indent-with-non-tab,-tab-in-indent diff -U0 --word-diff-regex='[^[:space:]]'

Answer 2

Ignore all whitespace changes with git-diff between commits

This question looked like it might offer an answer at first, but it deals with differences between two specific files, not between two specific commits. Everything else turned up by searching was a dead end as well....

I think you are having trouble because Git is the wrong tool for the job. It does not make the task easy, and yielding to accommodate the tool is the wrong approach. Tools are supposed to work for you and not vice versa.

Perform a second clone, and then checkout the starting revision in question. Then run regular diff on them using your current revision: diff -bur --ignore-all-space <dir1> <dir2>.

Here are some of the options for diff

-i, --ignore-case
       ignore case differences in file contents

-E, --ignore-tab-expansion
       ignore changes due to tab expansion

-Z, --ignore-trailing-space
        ignore white space at line end

-b, --ignore-space-change
       ignore changes in the amount of white space

-w, --ignore-all-space
       ignore all white space

-B, --ignore-blank-lines
       ignore changes where lines are all blank

Answer 3

The tool description of classical diff is "Compare FILES line by line."

And git diff follows that principle as well, although apparently with enhanced capabilities. But the notion of a line still seems fundamental to that tool. I haven't seen a way to work around that. And it is usually not worth the hassle to try and work against the nature of the tool.

In a situation such as described in the question, one would need to tokenize and normalize the text for both pre and post edit revisions of the files in such a way as to end up with handy units of compare for diff, which traditionally are lines, and then run your diff tool of choice on these normalized versions.