Surprising over-large diff

[tomwhoiscontrary]

(1) A description of the issue. A screenshot is often helpful too.

I have a Java source file with various changes, including a small change in the middle of a large method, including extracting some code into a new method which comes immediately after the large method. When i diffed my change, i was surprised to see that the whole large method was shown as deleted and re-added, so that the diff could line up some common parts in the old version of the large method and the new extracted method.

See this diff from difft:

• JSONFetcher.java.difft.txt

And this one from diff -u, which is much smaller:

• JSONFetcher.java.diffu.txt

(2) A copy of what you're diffing.

See attached (you will need to rename these to remove the .txt):

• JSONFetcher.baseline.java.txt
• JSONFetcher.changed.java.txt

(3) The version of difftastic you're using (see difft --version) and your operating system.

$ difft --version
Difftastic 0.52.0
$ grep PRETTY_NAME /etc/os-release
PRETTY_NAME="Ubuntu 18.04.6 LTS"

[Wilfred]

Thanks for the repro. That does look wrong, the result is far from the most minimal diff, which is what difftastic should be producing.

[Wilfred]

The expected diff should be more like this:

[Wilfred]

This looks like a bug with the 'split unchanged' logic, as the main differ is seeing three separate things to diff. The problem disappears with DFT_DBG_KEEP_UNCHANGED=y, although I don't recommend this debug setting in general (it's crucial for performance).