The inlinable_string crate provides the InlinableString type — an
owned, grow-able UTF-8 string that stores small strings inline and avoids
heap-allocation — and the StringExt trait which abstracts string
operations over both std::string::String and InlinableString (or even your
own custom string type).
StringExt's API is mostly identical to std::string::String; unstable and
deprecated methods are not included. A StringExt implementation is provided
for both std::string::String and InlinableString. This enables
InlinableString to generally work as a drop-in replacement for
std::string::String and &StringExt to work with references to either type.
Here are some current (micro)benchmark results. I encourage you to verify them
yourself by running cargo bench --feature nightly with a nightly Rust! I am
also very open to adding more realistic and representative benchmarks! Share
some ideas with me!
Constructing from a large &str:
test benches::bench_inlinable_string_from_large ... bench: 32 ns/iter (+/- 6)
test benches::bench_std_string_from_large ... bench: 31 ns/iter (+/- 10)
Constructing from a small &str:
test benches::bench_inlinable_string_from_small ... bench: 1 ns/iter (+/- 0)
test benches::bench_std_string_from_small ... bench: 26 ns/iter (+/- 14)
Pushing a large &str onto an empty string:
test benches::bench_inlinable_string_push_str_large_onto_empty ... bench: 37 ns/iter (+/- 12)
test benches::bench_std_string_push_str_large_onto_empty ... bench: 30 ns/iter (+/- 9)
Pushing a small &str onto an empty string:
test benches::bench_inlinable_string_push_str_small_onto_empty ... bench: 11 ns/iter (+/- 4)
test benches::bench_std_string_push_str_small_onto_empty ... bench: 23 ns/iter (+/- 10)
Pushing a large &str onto a large string:
test benches::bench_inlinable_string_push_str_large_onto_large ... bench: 80 ns/iter (+/- 24)
test benches::bench_std_string_push_str_large_onto_large ... bench: 78 ns/iter (+/- 23)
Pushing a small &str onto a small string:
test benches::bench_inlinable_string_push_str_small_onto_small ... bench: 17 ns/iter (+/- 6)
test benches::bench_std_string_push_str_small_onto_small ... bench: 60 ns/iter (+/- 15)
TLDR: If your string's size tends to stay within INLINE_STRING_CAPACITY, then
InlinableString is much faster. Crossing the threshold and forcing a promotion
from inline storage to heap allocation will slow it down more than
std::string::String and you can see the expected drop off in such cases, but
that is generally a one time cost. Once the strings are already larger than
INLINE_STRING_CAPACITY, then the performance difference is
negligible. However, take all this with a grain of salt! These are very micro
benchmarks and your (hashtag) Real World workload may differ greatly!
Either
$ cargo add inlinable_string
or add this to your Cargo.toml:
[dependencies]
inlinable_string = "0.1.0"