internal: replace difference with similar

[tommilligan]

See #44

Major rewrite to remove the dependency on difference, which is unmaintained.

[codecov-io]

Codecov Report

Merging #51 (bbeb51f) into main (f02609f) will increase coverage by 2.87%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main      #51      +/-   ##
==========================================
+ Coverage   96.55%   99.42%   +2.87%     
==========================================
  Files           5        5              
  Lines         145      174      +29     
==========================================
+ Hits          140      173      +33     
+ Misses          5        1       -4     

Impacted Files	Coverage Δ
tests/assert_ne.rs	96.96% <ø> (ø)
tests/pretty_string.rs	100.00% <ø> (ø)
src/lib.rs	100.00% <100.00%> (ø)
src/printer.rs	100.00% <100.00%> (ø)
tests/assert_eq.rs	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update f02609f...bbeb51f. Read the comment docs.

[tommilligan]

@mitsuhiko I've benchmarked this branch vs main using the benchmarks here, unfortunately there appear to be some performance regressions for writing the diff:

I'm going to continue investigating, but this means I'm unlikely to merge this implementation as is.

[tommilligan]

Closing for now as obsolete from #52, happy to revisit if the performance can be improved.

[mitsuhiko]

Thanks for the heads up. I will add some benchmarks to figure out what’s going on.

[mitsuhiko]

@tommilligan I'm not sure yet why myers performs so bad but I added a simple LCS implementation to see how it performs. I noticed two things from this:

• iter_all_changes internally needs to call Box::new for stupid lifetime reasons I can't figure out. You get a small performance benefit by calling ops + iter_changes yourself.
• next version of similar comes with Lcs you can use like this:

    let diff = TextDiff::configure()
        .algorithm(Algorithm::Lcs)
        .diff_lines(left, right);

Both of those together are similar in performance to the current main version of rust-pretty-assertions. Sadly I'm on a laptop so I can't do stable benchmarks here. It does say it outperforms diff but I can't be sure.

I also swapped out Myer's for a better implementation now that performs significantly better.

[mitsuhiko]

Pushed out a 1.2.0 with the changes. And thanks for the box fix.