Getting closer to the next sizable update to Mneme and wanted to share! After noticing that it can sometimes be difficult to visually parse what changed when an auto-generated value updates, I started working on semantic diffing, using the amazing Difftastic as reference and inspiration.

This has proved more of a rabbit hole than I initially thought, but is still relatively easy thanks to Elixir’s general awesomeness. Today I reached the point where my semantic diff development branch can start using semantic diffs to work on semantic diffs!

For instance, here’s what an update might look like using the latest release:

image (1)700×454 21.5 KB

Here’s the same update using semantic diffs:

image626×443 22.1 KB

There’s still a ways to go before this lands, but I’m hoping that being able to use semantic diffs in development will help expose edge-cases/etc. before I ship the feature.

v0.2 released

• Package on hex.pm
• Changelog
• Docs

Note: Mneme now requires Elixir v1.14 or later.

defp deps do
  [
    {:mneme, "~> 0.2.1", only: :test}
  ]
end

Semantic diffs

In addition to some quality-of-life changes when things go wrong, the big feature that v0.2 brings is a custom diff engine that produces dramatically more readable diffs when auto-assertions are updated. I’m really excited about this and it’s made Mneme much more pleasant to use, at least for me.

I expect there to be some formatting bugs – please report them on GitHub if you encounter anything weird! If you desire the old behavior, you can configure Mneme to continue using textual diffs as before.

Example 1. New assertions highlight only the new value

Text

image1067×453 26.2 KB

Semantic

image1001×330 24.1 KB

Example 2. Changed assertions highlight deleted/added nodes

Text

image1005×675 38.9 KB

Semantic

image959×551 33.4 KB

Example 3. Slightly changed strings call out the changed portion

Text

image983×528 26.8 KB

Semantic

image947×682 36.3 KB

Future work

• Diff quality: These diffs could still be improved. For instance, example 2 highlighted a suboptimal set of square brackets.
• Performance: For large expressions, semantic diffs will take too long and it will fall back to text diffs. I’m sure there’s still some low-hanging fruit for improving performance, however.
• Formatting: I’d like to add side-by-side formatting to decrease the vertical real-estate needed. I’d also like to explore replacing very large, unchanged nodes with ... or something in the same way that inspect does at a certain point.

I’ve released a new version (v0.2.2) that disables an optimization that was causing a poor diff in example 2 above, increasing the likelihood that subtrees would become “misaligned”.

v0.2.1

image959×551 33.5 KB

v0.2.2

image951×548 34 KB

The new diff does a much better job of reflecting the actual change: the nil is being replaced with a new subtree, and the second subtree is being modified.