Cassava is quite new, but has the same goals as lazy-csv.
Its about a year old now - http://blog.johantibell.com/2012/08/a-new-fast-and-easy-to-use-csv-library.html
I know Johan has been working on the benchmarks of late - it would be very good to know how the two compare in features
To run, check out the cassava repo on GitHub and run: cabal configure --enable-benchmarks && cabal build && cabal bench
Here are the results (all the normal caveats for benchmarking applies):
benchmarking positional/decode/presidents/without conversion
mean: 62.85965 us, lb 62.56705 us, ub 63.26101 us, ci 0.950
std dev: 1.751446 us, lb 1.371323 us, ub 2.295576 us, ci 0.950
benchmarking positional/decode/streaming/presidents/without conversion
mean: 93.81925 us, lb 91.14701 us, ub 98.19217 us, ci 0.950
std dev: 17.20842 us, lb 11.58690 us, ub 23.41786 us, ci 0.950
mean: 133.2609 us, lb 132.4415 us, ub 135.3085 us, ci 0.950
std dev: 6.193178 us, lb 3.123661 us, ub 12.83148 us, ci 0.950
The two first set of numbers are for cassava (in the all-at-once vs streaming mode). The last set is for lazy-csv.
The feature sets of the two libraries are quite different. Both do basic CSV parsing (with some extensions).
* lazy-csv parses CSV data to something akin to [[ByteString]], but with a heavy focus on error recovery and precise error messages.
* cassava parses CSV data to [a], where a is a user-defined type that represents a CSV record. There are options to recover from *type conversion* errors, but not from malformed CSV. cassava has several parsing modes: incremental for parsing interleaved with I/O, streaming for lazy parsing (with or without I/O), and all-at-once parsing for when you want to hold all the data in memory.