Implement a new `Get` interface

[well-typed]

Implement a new `Get` interface

[Shimuuar]

First question is what API should be used by Get. I guess it's usual monadic parser combinators.

Second question is backtracking. No backtracking means better performance. Backtracking could be useful for complicated parsers. I'm not sure what are better tradeoffs

[dcoutts]

The main difference these days between attoparsec and binary is that binary does not do backtracking, and this means it can release the input and be a bit faster. I think that's the same design tradeoff we would want to make.

[Shimuuar]

I just checked and Alternative instance of binary's Get does backtrack. But as I said earlier I'm not sure whether it really needed. Parsec for example does just fine with optional backtracking with try combinator.

[kolmodin]

binary does backtracking since version 0.6, released Sept 2012. I doubt it's needed for most data formats, in retrospect it should probably not have been added.
I'm not a fan of optional backtracking, it's too easy to make a mistake, the parser should always backtrack.
I also think that use of many is bug prone. I've used it mostly when I want to repeat as parser until the end of input, but it also stops decoding on decoder errors, which is rarely something you want when decoding binary data.

[Shimuuar]

I also want to point out that isolate-like combinator is invaluable in parsing existing data formats. Idiom here is size of data followed by data is ubiquitous and isolate make paring it very convenient.

[axman6]

Imo, it doesn't make sense to include a generic binary parser in this package - the benefits it provides are that it meets the needs of anyone needing to serialise Haskell data structures, and has the bonus of giving people a CBOR decoder for use with other systems. To me these are orthogonal issues, a serialisation library and a binary parsing library (which we already have several good alternatives, such as the current binary and attoparsec)

[bgamari]

I tend to agree that this package should stick to what it does well; binary is already a perfectly fine tool for dealing with externally-specified binary formats.