https://github.com/VincentToups/parseltongue.git
git clone 'https://github.com/VincentToups/parseltongue.git'
(ql:quickload :parseltongue)
Copyright 2012, Vincent Toups This program is distributed under the terms of the GNU Lesser General Public License (see license.txt).
This is a parser combinator library for Common Lisp inspired by SMUG, by Drew Crampsie and augmented with improvements inspired by my own usage of this library and other monadic computations in various Lisps.
This library probably has the same functionality as SMUG but migth be more familiar to those used to Haskell's do notation.
Fear not, you needn't understand how monads work in order to make good
use of this library. For those uninterested in how the library works,
there are only two things you need to know to use it: what parsers
are, how to use combinators on them, and how to use the special syntax
in the parser and defparser forms.
Parsers, in this library, are functions which accept a single
parameter, the input, and return either nil, if they cannot parse
anything from that input, or a list of parser-pair structs (defined
by the library), each of which represents a possible parsing of the
input, along with the remainder of the input which was not parsed.
For instance, a parser that parses “a” from a string looks like:
(defun =parse-a (input)
(if (empty? input)
nil
(let ((first-char (next input)))
(if (string= first-char "a")
(list (parser-pair first-char
(rest-of input)))
nil))))
The function next is a method which fetches the next item from an
input. next is defined for lists and strings by default, but can be
extended by the user with defmethod. rest-of is its partner, it
returns the rest of the input. parser-pair constructs a
parser-pair struct instance whose first value is the parsed result,
and whose second value is the rest of the input that was not parsed.
As indicated above, returning nil means nothing was parsed at all.
Any function which conforms to this type can be treated as a parser by
the library. If you want to write you own parsers without ever
touching the special syntax in the library, you do so just as we did
above. By convention, parsers in the library start with the =
character, to distinguish them visually from other functions.
We could have written the above parser with a combinator from the
library, =>string, like so:
(defun =parse-a (input) (funcall (=>string "a") input))
=>string is a function which takes a string and returns a parser,
which we use to parse the input. Any function which produces a parser
begins with => in this library. The simplest such function is the
function referred to as the parser return function, =>.
(=> 'some-value)
=> returns a parser which does nothing to the input, and returns a
single parser-pair (in a list), whose return value is some-value.
The whole idea of this library is to construct parsers from simpler parsers.
The library provides syntax to make writing parsers easier. We could
have defined =parse-a above like so, for instance:
(defparser =parse-a
(x <- =item)
(if (string= x "a") (=> x) =nil))
Since we often treat parsers as both functions and regular
variables, defparser establishes both a variable =parse-a and a
function by the same name. This function, as defined above, is
equivalent to the previous definitions, but how do we read it?
Within the body of a defparser (or parser form, which is the
anonymous version), each expression must be either a parser itself
or a “binding expression” of the form
`(variable <- parser-expression)`
When an expression is just a parser, that parser must succeed or the entire parser being defined will return nil. Subsequent parsers executed in the body will then see only the input not parsed by previous parsers.
When a binding expression is encountered, the parser on the right
hand side is to the current input, which may have been parsed down by
previous expressions in the body, and, if the parse succeeds, then in
the rest of the body, the variable in the left-hand-side is bound to
the value in the value-slot of the parser-pairs returned by the
parser. Hence, in the above expression, the first form:
(x <- =item)
Sees =item applied to the input of the parser. =item always pulls
one item off of the input and only fails when the input is empty.
If the input is empty, then the parse fails, and no further forms are
executed. If =item succeeds, then we look to the next line. The
next line is not a binding form. It is an if statement, which is
fine, subject to the constraint that each branch must return a
parser. If the item we've parsed from the input is “a”, then we use
=> to create a parser which inserts “a” as its parser-pair's value.
If not, we return the =nil parser, which always fails.
If we wanted to parse “a” and then “b”, we'd write:
(defparser =parse-ab
(=>string "a")
(=>string "b"))
The =>or combinator produces a parser if any of its input parsers
succeed, returning the value of the first success from left to
right. For instance, to parser “a” or “b”:
(defparser =a-or-b
(=>or (=>string "a")
(=>string "b")))
Then:
(=a-or-b "abc") -> (list (parser-pair "a" "bc"))
(=a-or-b "bbc") -> (list (parser-pair "b" "bc"))
The combinator =>and succeeds only when all of its input parsers
succeed in turn, finally returning the result of the last parser:
(funcall (=>and (=>string "a") (=>string "b")) "ab") ->
(list (parser-pair "b" ""))
The combinator =>items parses n or fewer items from the input,
regardless of what they are:
(funcall (=>items 3) "abcd") ->
(list (parser-pair (list "a" "b" "c") "d"))
The combinator =>zero-plus-more parsers as many of its input parser
as possible and returns them in a list, possibly an empty one:
(funcall (=>zero-plus-more
(=>string "a"))
"aaaab") ->
(list (parser-pair (list "a" "a" "a" "a")
"b"))
The combinator =>one-plus-more does the same except it fails if
there is not at least one parsable object.
Parseltongue, like SMUG, is actually a non-deterministic library -
parsers can parse in multiple ways simultaneously. I'll write some
documentation about that later, but if you are using it for regular
deterministic parser, you are usually interested in only the first
parser-result. Hence, the function parse/first-result is a handy
thing:
(parser/first-result (=>string "a") "abc") -> "a"
It returns the first parse result and leaves off the leftover input.
I'd like to thank Drew for writing up SMUG, which was critical in developing an understanding of monads in Lisp and obviously in inspiring this library. He also provided some correspondence when I didn't understand aspects of his code.
If you like this library, it is almost a line for line port of an Elisp parser combinator library I also wrote, available in my emacs-utils repository here on github.