Delimited parsers

[Lysxia]

A common situation is to parse an encoding prefixed by its length. So you first parse the length as an integer n, and then you would like to run a (sub)parser p :: Parser a only on the next n bytes. I could think of two solutions for users today:

• Use take to get the ByteString and apply parseOnly p. However, we lose source position information in case the subparser fails, and we have to keep the whole ByteString in memory.

• Wrap Parser (e.g., with a few monad transformers) to track things like the number of bytes read; that would allow combinators like the ones I have in mind. However, this is rather heavyweight to implement. Does an existing library already offer this? I also suspect this approach would have more overhead than necessary.

It would be nice for attoparsec to have combinators to delimit the input that a subparser gets to see, like span and splitAt in pipes-parse.

What do you think of such an addition? Is there a better solution?

[Lysxia]

#48 and #95 were in this situation before, with solutions that correspond to the first item above.

[joeyh]

I keep needing to do this kind of thing in my attoparsec parsers, and only on discovering this bug am I shaking the feeling that I'm somehow using attoparsec wrong to need to use parseOnly within a parser so frequently.

With me, it often comes up while writing something like Parser a -> Parser b, which needs to pick out the delimited data and run the sub-parser over it.

Checking endOfInput seems like one thing that can easily be gotten wrong when doing this. The example in #95 perhaps forgot to do that. I wonder if a combinator for this should require the sub-parser to consume all the input?