This is the common wc command, implemented in Haskell.
Turns out that it's about 4-6 times faster than GNU coreutils' wc,
depending on the exact subset of statistics being computed.
hwc supports counting bytes, words and lines, as well as UTF-8 characters and maximum line length.
It does so in a composable and efficient manner, so that only those statistics requested by user
are actually calculated.
It all started from this blog post. The author of that post manages to beat the C version by resorting to parallel input file handling (and that's a good demonstration, no critique here — go figure how to parallelize the C version correctly!), but I was curious if it's possible to beat the (single-threaded) C version by a single-threaded Haskell version.
Turns out the answer is yes, and with some very minor modifications to the original author's implementation.
For more details about the implementation, please refer to this post.