Currently every WKT number parses via T::from_str (tokenizer.rs:85). fast-float (MIT/Apache-2.0) is the same Eisel-Lemire algorithm specialized to decimals.
Swapping it for my internal WKT reader on larger files results in roughly 10-15% faster reads with identical rounding.
This would be an addition to #118 as it requires concrete numeric types instead of the current T: FromStr. Smaller improvement than ryu but still worthwhile in my opinion.
Currently every WKT number parses via
T::from_str(tokenizer.rs:85). fast-float (MIT/Apache-2.0) is the same Eisel-Lemire algorithm specialized to decimals.Swapping it for my internal WKT reader on larger files results in roughly 10-15% faster reads with identical rounding.
This would be an addition to #118 as it requires concrete numeric types instead of the current
T: FromStr. Smaller improvement than ryu but still worthwhile in my opinion.