Lightweight parser combinator library for Idris, inspired by Parsec.

Build Status

Module overview:


This package is used (almost) the same way as Parsec, except for one difference: backtracking.


In other words, the following two pieces of code are equivalent (using illustrative combinator names):


elem :: Parser Int
elem = (try (string "0x") >> hexNumber) <|> decNumber


elem : Parser Int
elem = (string "0x" $> commitTo hexNumber) <|> decNumber
-- which may be abbreviated as:
--   = (string "0x" >! hexNumber) <|> decNumber

After reading the prefix 0x, both parsers commit to reading a hexadecimal number or nothing at all — Parsec does this automatically, Lightyear uses the commitTo combinator for this purpose. On the other hand, Parsec requires the string "0x" to be wrapped in try because if we are reading 0123, we definitely don't want to commit to the left branch upon seeing the leading 0.

For convenience, commitTo is merged in monadic and applicative operators, yielding the operators >!=, >!, <$!>, <$!, and $!>. The ! in the names is inspired by the notation used for cuts in Prolog.

A parser that uses commitment might look like this (notice the leading char '@' that leads to commitment):

entry : Parser Entry
entry = char '@' >! do
  typ <- pack <@> some (satisfy (/= '{'))
  token "{"
  ident <- pack <@> some (satisfy (/= ','))
  token ","
  items <- item `sepBy` comma
  token "}"
  return $ En typ ident items

Lazy Branching with <|>|

It is worth noting that Idris itself is a strict language, and thus the <|> operator will evaluate both its arguments eagerly by default. In order to lazily evaluate different parsing branches we are required to use a special operator: <|>|. In general, all recursive calls of combinators have to occur in a lazy context. (With mutual recursion, this generalises to the rule that each call cycle has to be broken by laziness in at least one place.)

In the wild, it might look like this:

partial parseExpr : Parser SExpr
parseExpr = parseName <|>| ( MkSExpr <$> parens (many parseExpr) )

In the above example, the whole RHS of <|>| is lazy, and so the recursive occurrence of parseExpr in it will be evaluated only if the LHS of <|>| fails. Using <|> would cause infinite recursion.

For convenience, a version of <*> that lazily evaluates its second argument is included as <*>|. Conversely to <|>|, the RHS of <*>| will be evaluated only if the LHS of <*>| succeeds.


Lightyear is used to parse BibTeX in <a href="https://github.com/ziman/bibdris/blob/master/Bibtex.idr">bibdris</a>.


$ make clean
$ make test
$ make install