Principled Setoid and Ord with improved performance

[Avaq]

Opened a draft PR for discussion.
See #151 (comment).

Remember we did that thing where we made Z.equals more useful by having it fall back to structural equality for garbage inputs (#154)? We ended up with a useful, but unprincipled and slow, equality function. That's fine and all, but I don't like that:

• Sanctuary is now forced to use this slow function, despite already having ensured that inputs are Setoids of the same type.
• Many Setoid implementations build on Z.equals, making them slow for no reason.
• Some Ord implementations, and Z.lte itself, builds on Z.equals.

[Avaq]

The new code is problematic: Z.equals doesn't check that both inputs are the same Setoid all the way down; it checks only x before dispatching to setoidEquals, meaning it will dispatch with garbage inputs such as equals ([null], [42]) for which setoidEquals will happily return true because of the Setoid implementation for null.

[Avaq]

This probably means that we'd have to have two completely separate implementations for equals and setoidEquals, with two separate type classes. This is something I hoped to avoid initially.

[Avaq]

Right, so I removed the whole structural deep equality function, which I believe we said (face to face) we should reimplement in a separate package if the need is great enough.

[Avaq]

The effect of these changes on performance:

$ npm run bench -- --benchmark old-vs-new --match '*{equals,lte}*'

> sanctuary-type-classes@13.0.0 bench
> sanctuary-benchmark "--benchmark" "old-vs-new" "--match" "*{equals,lte}*"

Running benchmarks: old-vs-new...

┌─────────────────────────┬──────────────────────────┬──────────────────────────┬────────┬──────────┬───┐                                                                                                                                                      
│ suite                   │ old                      │ new                      │ diff   │ change   │ α │
├─────────────────────────┼──────────────────────────┼──────────────────────────┼────────┼──────────┼───┤
│ methods.equals.Identity │ 202,779 Hz ±0.22% (n 94) │ 209,578 Hz ±0.59% (n 95) │ 001.6% │ +003.4%  │   │
├─────────────────────────┼──────────────────────────┼──────────────────────────┼────────┼──────────┼───┤
│ methods.equals.Object   │ 57,084 Hz ±1.84% (n 87)  │ 395,393 Hz ±2.48% (n 83) │ 074.8% │ +592.7%  │ ✓ │
├─────────────────────────┼──────────────────────────┼──────────────────────────┼────────┼──────────┼───┤
│ methods.lte.Identity    │ 145,642 Hz ±0.68% (n 93) │ 139,259 Hz ±4.06% (n 78) │ 002.2% │ -004.4%  │   │
├─────────────────────────┼──────────────────────────┼──────────────────────────┼────────┼──────────┼───┤
│ methods.lte.Object      │ 37,530 Hz ±2.05% (n 89)  │ 444,342 Hz ±2.20% (n 89) │ 084.4% │ +1084.0% │ ✓ │
└─────────────────────────┴──────────────────────────┴──────────────────────────┴────────┴──────────┴───┘

[Avaq]

If we go this route, this will naturally have to be released as a breaking change. I don't expect many people were using lte and friends on anything other than real Ords, but I expect a few users to be impacted by the change to equals.

[Avaq]

Although.. it hasn't been that long ago that equals was producing real useless garbage output for non-Setoids. Maybe it's not too bad.

[davidchambers]

🎉 🎉 🎉