If your code reads or writes some data in a streaming fashion, i.e. in chunks, you can use this library to compute a hash of the data incrementally. This is good both for constant-memory stream processing and for clean readable code using monads. My original use case is the latter, but possibly the former is significant as well (at the time of writing, I haven't done any profiling to check).

One of the common uses of hashes is ensuring integrity. Some data is published along with its hash, and you can download the data and use the hash to verify your copy is valid (not corrupt, not tampered with).

When the data is small, you can serialize it into a ByteString (if you are sending it) or read it into a bytestring (if you are receiving), and then apply a hash function. But there is a common case in which you send large data in chunks, and you want to send a hash of it afterwards. If you first compute the data to send and only then send it and compute the hash, you may end up with the entire data in memory. Instead, you can hash the data incrementally, chunk by chunk in parallel to sending the chunks, and by the time you finish sending the data, you have a ready final hash you can now send too.

A similar case exists when receiving data: Sometimes you read data (e.g. from the network) in chunks, and process it incrementally. If you applied the hash at once on the entire data, you'd have to keep it all in memory, while using an incremental approach, you could process the chunks instantly and release them, therefore using constant memory.

This library offers a simple monad transformer which fills this role.

If an all-at-once hash function is applied only after you finish receiving the data, perhaps a smart compiler can apply it incrementally, and then explicit incremental hashing won't make a performance difference. But even then, that will work only in the trivial case where the data to be hashed is streamed in one go. If you need to send it in several parts and interleave some other actions in the middle, for example build one message based on a reply for the previous one, optimizations won't help you, and you'll need incremental hashing for constant-memory stream processing.

The HashT monad transformer is useful in the same way StateT is: Instead of passing around the temporary hash context, it's kept for you in the monad transformer implicitly. You access it only when you need to, and otherwise your code doesn't have to carry it around everywhere.

This library uses the cryptonite package and supports any hash algorithm implemented there (see Crypto.Hash.Algorithms).