Currently when wrangler needs to compute the diff of two json objects uses its own diffing logic, this however is not ideal for a few reasons.

The main issue is that the implementation is completely text-based and very lossy. Before objects can be compared they need to be stringified (example) and then a line-by-line diffing is applied. This results in a result that doesn't contain much information regarding what changed, it mostly includes how many lines have been added, removed or modified (this for example can include lines that imply add a semicolon due to a new properties being added).

Ideally we should have a diffing implementation that provides information on what specific fields have changed and how.

Additionally the current diffing logic presents quirks that are difficult to deal with (example).