How to use ediff with JSON, so it abstracts from member order in objects?

Question

Currently I use ediff a lot for diffing all sorts of text files. I also use it for looking at diffs between JSON files. The problem is that a lot of differences appear which are semantically irrelevant.

e.g. This snippet:

{
  alice: "Springs",
  bob: "4 apples"
}

Is semantically the same as this one since the order of keys in a JSON object does not matter (and cannot even be assumed to be well-defined):

{
  bob: "4 apples",
  alice: "Springs"
}

Obviously, however the two files are different on a plain text basis so if I look at it in ediff, I'm likely to see noise I don't care about. Is there a way/tool/technique for filtering this out?

Answer 1

I don’t have a complete solution for you, but you can run the json through jq --sort-keys to eliminate this type of unimportant difference. You might combine this with git diff --textconv, or perhaps ediff has a similar text conversion feature. You might just wrap it up in a shell script and set ediff-diff-program to that script. Note however that a diff modified in this fashion will be suitable for viewing but not for applying, so some ediff features might not work.