Thesis
The accepted paradigm for doing set contains (or set operations in general) pre Go 1.18 (generics) is to use a map[type]struct{}. I believe that using this for small datasets is going to be slower than using slices. I also just don't like the way map[string]struct{}{} looks. Disgusting.
String Results
Summary
Slices only behaved better on small N, and only slightly. When we were at N = 100, we already are at a 30% faster using MapSet. Looks like for most string cases, using a map as the underlying storage makes sense.
N = 100
BenchmarkSliceSet-10 51308 22884 ns/op
BenchmarkSliceSetExp-10 54638 22027 ns/op
BenchmarkMapSet-10 79706 14912 ns/op
N = 1000
BenchmarkSliceSet-10 741 1504577 ns/op
BenchmarkSliceSetExp-10 793 1535511 ns/op
BenchmarkMapSet-10 6694 180629 ns/op
N = 10
BenchmarkSliceSet-10 1110586 1066 ns/op
BenchmarkSliceSetExp-10 1000000 1052 ns/op
BenchmarkMapSet-10 987474 1222 ns/op
Int results
Summary
Ints perform better as slices for longer. At N = 100, we still have it at about 54% faster. However, the gap really widens once we get to N = 1000.
N = 100
BenchmarkIntMapSet-10 197214 5917 ns/op
BenchmarkIntSliceSet-10 435561 2709 ns/op
N = 10
BenchmarkIntMapSet-10 3872775 311.5 ns/op
BenchmarkIntSliceSet-10 8202166 144.7 ns/op
N = 1000
BenchmarkIntMapSet-10 20090 60114 ns/op
BenchmarkIntSliceSet-10 6854 175596 ns/op