binrep: Encode precise binary representations directly in types

[ data, library, mit, serialization ] [ Propose Tags ]

Flags

Manual Flags

NameDescriptionDefault
icu

use text-icu package (requires ICU library)

Enabled

Use -f <flag> to enable a flag, or -f -<flag> to disable that flag. More info

Note: This package has metadata revisions in the cabal description newer than included in the tarball. To unpack the package including the revisions, use 'cabal get'.

Maintainer's Corner

Package maintainers

For package maintainers and hackage trustees

Candidates

• No Candidates
Versions [RSS] 0.1.0, 0.2.0, 0.3.0, 0.3.1 CHANGELOG.md aeson (>=2.0 && <2.1), base (>=4.14 && <5), bytestring (>=0.11 && <0.12), either (>=5.0.1.1 && <5.1), flatparse (>=0.3.5.0 && <0.4), mason (>=0.2.5 && <0.3), megaparsec (>=9.2.0 && <9.3), refined (>=0.7 && <0.9), strongweak (>=0.3.1 && <0.4), text (>=1.2 && <2.1), text-icu (>=0.7.0.0 && <0.9), vector (>=0.12.3.1 && <0.13), vector-sized (>=1.5.0 && <1.6) [details] MIT Ben Orchard Ben Orchard Revision 1 made by raehik at 2023-01-11T12:02:48Z Data, Serialization https://github.com/raehik/binrep#readme https://github.com/raehik/binrep/issues head: git clone https://github.com/raehik/binrep by raehik at 2022-08-30T17:50:08Z NixOS:0.3.1 1 direct, 0 indirect [details] 119 total (19 in the last 30 days) (no votes yet) [estimated by Bayesian average] λ λ λ Docs uploaded by userBuild status unknown

[back to package description]

binrep

binrep is a library for precisely modelling binary schemas and working with them effectively and efficiently in Haskell. Here's why it's useful:

• Explicit: Binary representation primitives such as C-style bytestrings (null-terminated), sized explicit-endian machine integers, and null-padded data enable defining Haskell data types with the binary schema "baked in".
• Low boilerplate: Generic parsers and serializers further reduce boilerplate for straightforward schemas. (See Generic binary representation for details.)
• Easy validation: Goes hand in hand with my strongweak library to allow working with unwrapped data internally, and enforcing all the binary representation invariants before serializing - no extra definitions required.
• Performant: Parsing and serialization is low-level and extremely fast, using flatparse and mason respectively.

Usage

Dependencies

You need the ICU library. For running, you just need the runtime. For building, you need development files as well (headers etc). Alternatively, you may turn off the ICU features with a Cabal flag.

Philosophy

Modelling, not serializing

binrep is good at modelling binary data formats. It is not a plain "serialization" library, where the actual binary representation is hidden from the user (intentionally, with good reason). The binary and cereal libraries are great choices for that. They are interested in defining efficient binary codecs for Haskell data. However, their codec typeclasses hide representation decisions from the user. In cereal,

These are fine decisions. But they aren't accurate to the types. Endianness is an implementation decision.

binrep refuses to work with a machine integer unless it knows the endianness. Bytestrings are split into C-style (null-terminated) and Pascal-style (length-prefixed). This enforces careful consideration for the binary data being modelled.

Validation without boilerplate

A C-style bytestring must not contain any 0x00 null bytes. A Pascal-style bytestring must be short enough to be able to encode its length in the length prefix machine integer. But checking such invariants is tedious work. Am I really going to wrap everything in a bunch of newtypes and force users to call a bunch of checker functions every time?

Yes and no. Yes, binrep uses newtypes extensively, though most are type synonyms over the Refined newtype from Nikita Volkov's wonderful refined library. No, binrep doesn't want you to wrangle with these day-to-day. One solution is to define a simplified "weak" type, and convert between it and the binary-safe "strong" type. My strongweak library provides supporting definitions for this pattern, and generic derivers which will work with binrep's binary representation primitives.

Performant primitives

Parsing uses András Kovács' flatparse library. Serializing is via Fumiaki Kinoshita's mason library. These are about as fast as you can get in 2022.

We only define serializers for validated types, meaning we can potentially skip safety checks, that other serializers would do. Except we still do them, but validation is an explicitly required step before serialization.

This might change if we start to support weirder binary representations, specifically offset-based data.

Generic binary representation

binrep's generic deriving makes very few decisions:

• Constructors are encoded by sequentially encoding every enclosed field.
• Empty constructors thus serialize to 0 bytes.
• Sum types are encoded via a tag obtained from the constructor names.
• It's the same approach as aeson, with a bit more flexibility: see below.

Sum types (data types with multiple constructors) are handled by first encoding a "tag field", the value of which then indicates which constructor to use. You must provide a function to convert from a constructor name to a (unique) tag. You could encode them as a null-terminated ASCII bytestring (this is the default), or as a single byte. To ease this, you may consider putting the tag value in constructor names:

data BinarySumType = B1 | B2

getConstructorTag :: String -> Word8
getConstructorTag = read . drop 1

-- >>> getConstructorTag "B1"
-- 1

-- Or use our generic helper, which takes hex values:
--
-- >>> cSumTagHex @Word8 (drop . 1) "BFF"
-- 255


Similar projects

Kaitai Struct

Kaitai Struct is a wonderful declarative parser generator project. They bolt an expression language and a whole lot of binary cleverness on top of a nice YAML schema. It comes with an IDE, a visualizer, and you can compile schemas down to parsers for various different languages (no Haskell...).

Design principles like their fancy absolute offset handling and language neutrality have stunted serialization support. Though it's more like they have such powerful parsing that they can parse formats that can't be edited and re-serialized naively, like archives with file indexes. For proper handling, one should store a file table, and serialization generates the index. So in reverse, you would want to combine them. But it's a bit program-y. In binrep, you are in a programming language, so it's less of a problem... but I'm not sure if we can be very efficient at absolute offset stuff.

Realistically, Kaitai Struct is the best decision for fast iteration on reversing unknown data. binrep is useful for loading data straight into Haskell for further processing, especially converting between simpler formats.

Wuffs

Wuffs is a crazy exploration into safe low-level code via strong typing. You have to annotate every possibly dangerous statement with a proof of safety. It's a tedious, explicit, very safe and very fast imperative language for defining parsers and serializers.

Wuffs is more a codec engineer's tool than a reverse engineer's one. binrep isn't really interested in speed, and being a Haskell library we get to focus on defining types and their composition in a declarative & functional manner. As such, we get to define more useful things quicker using binrep. Though we share many core ideas, such as refinement types.

Check out Wuffs if you need to write a bunch of codecs and they really, really need to be both fast and safe. The trade-off is, of course, your time.