Replaces uses of SmallVec withTinyVec, which is the same idea implemented in 100% safe code.
Performance is generally in line with SmallVec in this case, with some benchmarks being slightly slower and some slightly faster:
SmallVec 1.1.0
test bench_is_nfc_ascii ... bench: 15 ns/iter (+/- 0)test bench_is_nfc_normalized ... bench: 26 ns/iter (+/- 0)test bench_is_nfc_not_normalized ... bench: 455 ns/iter (+/- 7)test bench_is_nfc_stream_safe_ascii ... bench: 14 ns/iter (+/- 0)test bench_is_nfc_stream_safe_normalized ... bench: 39 ns/iter (+/- 0)test bench_is_nfc_stream_safe_not_normalized ... bench: 526 ns/iter (+/- 9)test bench_is_nfd_ascii ... bench: 17 ns/iter (+/- 3)test bench_is_nfd_normalized ... bench: 43 ns/iter (+/- 0)test bench_is_nfd_not_normalized ... bench: 15 ns/iter (+/- 0)test bench_is_nfd_stream_safe_ascii ... bench: 17 ns/iter (+/- 3)test bench_is_nfd_stream_safe_normalized ... bench: 72 ns/iter (+/- 1)test bench_is_nfd_stream_safe_not_normalized ... bench: 15 ns/iter (+/- 0)test bench_nfc_ascii ... bench: 557 ns/iter (+/- 6)test bench_nfc_long ... bench: 235,422 ns/iter (+/- 2,928)test bench_nfd_ascii ... bench: 325 ns/iter (+/- 6)test bench_nfd_long ... bench: 133,196 ns/iter (+/- 1,167)test bench_nfkc_ascii ... bench: 557 ns/iter (+/- 19)test bench_nfkc_long ... bench: 252,428 ns/iter (+/- 2,076)test bench_nfkd_ascii ... bench: 300 ns/iter (+/- 3)test bench_nfkd_long ... bench: 137,857 ns/iter (+/- 1,435)test bench_streamsafe_adversarial ... bench: 471 ns/iter (+/- 7)test bench_streamsafe_ascii ... bench: 81 ns/iter (+/- 1)
TinyVec 0.3.2
test bench_is_nfc_ascii ... bench: 14 ns/iter (+/- 0)test bench_is_nfc_normalized ... bench: 26 ns/iter (+/- 0)test bench_is_nfc_not_normalized ... bench: 440 ns/iter (+/- 3)test bench_is_nfc_stream_safe_ascii ... bench: 14 ns/iter (+/- 0)test bench_is_nfc_stream_safe_normalized ... bench: 38 ns/iter (+/- 0)test bench_is_nfc_stream_safe_not_normalized ... bench: 524 ns/iter (+/- 5)test bench_is_nfd_ascii ... bench: 16 ns/iter (+/- 2)test bench_is_nfd_normalized ... bench: 42 ns/iter (+/- 0)test bench_is_nfd_not_normalized ... bench: 14 ns/iter (+/- 0)test bench_is_nfd_stream_safe_ascii ... bench: 16 ns/iter (+/- 0)test bench_is_nfd_stream_safe_normalized ... bench: 53 ns/iter (+/- 0)test bench_is_nfd_stream_safe_not_normalized ... bench: 15 ns/iter (+/- 0)test bench_nfc_ascii ... bench: 595 ns/iter (+/- 8)test bench_nfc_long ... bench: 216,145 ns/iter (+/- 3,176)test bench_nfd_ascii ... bench: 361 ns/iter (+/- 5)test bench_nfd_long ... bench: 141,053 ns/iter (+/- 552)test bench_nfkc_ascii ... bench: 587 ns/iter (+/- 8)test bench_nfkc_long ... bench: 225,678 ns/iter (+/- 2,640)test bench_nfkd_ascii ... bench: 333 ns/iter (+/- 7)test bench_nfkd_long ... bench: 149,023 ns/iter (+/- 613)test bench_streamsafe_adversarial ... bench: 485 ns/iter (+/- 2)test bench_streamsafe_ascii ... bench: 68 ns/iter (+/- 0)
As you can see, e.g.bench_is_nfd_stream_safe_normalized andbench_nfkc_long have improved whilebench_nfkd_long andbench_nfc_ascii have regressed, but none of the changes are dramatic.
cargo test passes, I haven't run any parametric or fuzz testing (not sure if this crate has any). TinyVec itself has been fuzz-tested for equivalence tostd::Vec usingarbitrary-model-tests framework.
Replaces uses of SmallVec withTinyVec, which is the same idea implemented in 100% safe code.
Performance is generally in line with SmallVec in this case, with some benchmarks being slightly slower and some slightly faster:
SmallVec 1.1.0
TinyVec 0.3.2
As you can see, e.g.
bench_is_nfd_stream_safe_normalizedandbench_nfkc_longhave improved whilebench_nfkd_longandbench_nfc_asciihave regressed, but none of the changes are dramatic.cargo testpasses, I haven't run any parametric or fuzz testing (not sure if this crate has any). TinyVec itself has been fuzz-tested for equivalence tostd::Vecusingarbitrary-model-tests framework.