Emily wrote:
Do you provide reproducible benchmarks for these claims? I'd love to see the outcome for various types of projects to make the choice for myself.
Unsubstantiated claims of "This is Faster™️" are the path to shame and treachery.
Version 0.1.2 now includes a CSV Benchmark. The benchmark generates random table layouts of up to 50 columns and decodes two named columns in 100000 lines. Speed-up on my machine is roughly 0.7x over ParsecT and 0.6x over Cassava when decoding only few columns of a broad table. In particular, a Text-based ParsecT is faster than (ByteString-based) Cassava in this scenario. This was also the initial reason why I got interested in improving performance of megaparsec parsers in the first place. Cheers, Olaf