On Fri, 06 Sep 2013 03:07:23 +0400, Johan Tibell
I think the important issue here is whether we want to add monadic versions of functions in bytestring (where it makes sense)? It wield yield better performance than using `unpack`, but at the cost of lots of code duplication.
We have the same issue in containers where you could for example want an `updateM` function, that would let you decide whether to update a value by performing some side effect. You could simulate this using a combination of `lookup` and `insert`, but that requires twice the number of traversals of the data structure.
Right now I'm of the mind that the extra traversals (and using unpack in the case of ByteString) is better than the code duplication.
This code already is in bytestring — as part of `sort`. So, no code would actually be duplicated, just moved. (And we get sortBy practically for free!) Then, bytestring *is* code duplication. Raw loops everywhere, same raw loops with minor changes; peekByteOff-pokeByteOff; calling C for 30% speed improvements (and leaving loop-y Haskell versions commented nearby); and much more. I find myself wondering if by carefully choosing a set of basic primitives the code of bytestring could be made ten times shorter.