What is Parsley?
Parsley is a very fast parser combinator library that outperforms the other libraries in both the parsec family, as well as Happy. To make this possible, it makes use of Typed Template Haskell to generate the code for the parsers.
Parsley is distributed on Hackage, and can be added by depending on the package
The version policy for
parsley-core adheres to the regular Haskell PVP, but the two major versions are distinguished: the first is the Public API major version, which represents backwards incompatible changes
in the regular PVP sense that could affect
parsley itself (note
parsley only imports from
Parsley.Internal itself); the second version is the
Internal API major version, which would only effect users who use part of the internal parsley
modules. As such, for people that are not explicitly importing anything from
Parsley.Internal.*, the second major version does not matter:
0.3.0.0 would be compatible, for instance.
To use it, you'll need to write you parsers in another file from where they will be used: this is due to Template Haskell.
How does Parsley being a Staged Selective library change its use?
By being a Selective Parser Combinator library, Parsley does not support monadic operations such
return. Instead, the most powerful operations are
branch. Most monadic
power can be recovered using the functionality provided by
Parsley.Register, as well as the
The reason monads are not supported is because of the Staging: Parsley parsers are compiled ahead
of time to produce fast code, but this means the entirety of the parser must be known before any
input is provided, ruling out dynamic monadic operations. The use of staging (in this instance provided
by Typed Template Haskell) means that the signatures of the combinators do not correspond to their
counterparts in other libraries: they don't use raw values, they use code of values. A consequence
of this is that Parsley does not implement instances of
do-notation also cannot be used even with
ApplicativeDo, except perhaps
RebindableSyntax is used.
Code is provided to the combinators by way of the datatype
Defunc if you're feeling fancy),
which pairs a normal value with its Haskell code representation:
data WQ a = WQ a (Code a)
This gives us combinators like:
pure :: WQ a -> Parser a satisfy :: WQ a -> Parser a char :: Char -> Parser a char c = satisfy (WQ (== c) [||(== c)||])
WQ explicitly like this can get annoying, which is what the
parsley-garnish package is for!
Currently, the garnish provides one plugin called
OverloadedQuotes, which replaces the behaviour of
the default Untyped Template Haskell quotes in a file so that they produce one of
Parsley.OverloadedQuotesPlugin module in the
parsley-garnish package for more information.
How does it work?
In short, Parsley represents all parsers as Abstract Syntax Trees (ASTs). The representation of the parsers the users write is called the Combinator Tree, which is analysed and optimised by Parsley. This representation is then transformed into an Abstract Machine in CPS form, this is analysed further before being partially evaluated at compile-time to generate high quality Haskell code. For the long version, I'd recommend checking out the paper!
If you encounter a bug when using Parsley, try and minimise the example of the parser (and the input)
that triggers the bug. If possible, make a self contained example: this will help me to identify the
issue without too much issue. It might be helpful to import
parsley-core:Parsley.Internal.Verbose to provide a
debug dump that I can check out.
- This work spawned a paper at ICFP 2020: Staged Selective Parser Combinators
For talks on how writing parsers changes when using Parsley see either of these:
- Garnishing Parsec with Parsley - Berlin Functional Programming Group, January 2021
- Exploring Parsley: Working with Staged Selective Parsers - MuniHac 2020
For the technical overview of how Parsley works:
- Staged Selective Parser Combinators - ICFP 2020