On Jan 6, 2008 9:13 PM, Bryan O'Sullivan
Ooh, nice. We could really do with an incremental version, too, which could be spoonfed chunks of bytes, and dole out values as deserialisation completes.
Passing back a Left String is in some sense not much of an improvement over calling error, as if I merely doesn't have enough bytes accumulated yet, I can't restart parsing from the point the bytes ran out. Any chance you'd like to hand back a continuation instead?
The Left String was just a way to return an error string, otherwise a parse failure can be a little oblique. I think an incremental parser would be sufficiently different that it would be a different monad. Of course, the code duplication (again) makes me a sad panda, but avoiding that involves a lot of work on the binary package and the maintainers were uninterested last time I tried that. It would seem that there would be three possible outcomes from an incremental Get: - Failure: some bitstreams are just invalid and no amount of extra data will ever fix that - Complete [Result]: the last chunk of data has been processed. Maybe this should also include the remainder of the data? - Partial Cont [Result]: needs more data, but here's a (possibly empty) list of results so far. Applying a ByteString to the Cont from the Partial result would result in one of the same three outcomes etc. What do you think of that? It wouldn't be too hard to make up something like that using ContT if it would be useful to you. AGL -- Adam Langley agl@imperialviolet.org http://www.imperialviolet.org 650-283-9641