A flexible, extensible s-expression library.
S-Cargot is a library for working with s-expressions in a modular and extensible way, opting for genericity and flexibility instead of speed. Instead of understanding one particular form of s-expression, the S-Cargot library exposes tools for parsing or emitting different kinds of s-expressions, including features not normally included in an s-expression library like reader macros or tight control over indentation in pretty-printing.
S-Cargot is a library for parsing and emitting S-expressions, designed to be flexible, customizable, and extensible. Different uses of S-expressions often understand subtly different variations on what an S-expression is. The goal of S-Cargot is to create several reusable components that can be repurposed to nearly any S-expression variant.
S-Cargot does not aim to be the fastest or most efficient s-expression library. If you need speed, then it would probably be best to roll your own AttoParsec parser. Wherever there's a choice, S-Cargot errs on the side of maximum flexibility, which means that it should be easy to plug together components to understand various existing flavors of s-expressions or to extend it in various ways to accomodate new flavors.
What Are S-Expressions?
S-expressions were originally the data representation format in Lisp implementations, but have found broad uses outside of that as a data representation and storage format. S-expressions are often understood as a representation for binary trees with optional values in the leaf nodes: an empty leaf is represented with empty parens (), a non-empty leaf is represented as the scalar value it contains (often tokens like x or other programming language literals), and an internal node is represented as (x . y) where x and y are standing in for other s-expressions. In Lisp parlance, an internal node is called a cons cell, and the first and second elements inside it are called the car and the cdr, for historical reasons. Non-empty lef nodes are referred to in the s-cargot library as atoms.
Often, s-expressions are used to represent lists, in which case the list is treated as a right-branching tree with an empty leaf as the far right child of the tree. S-expression languages have a shorthand way of representing these lists: instead of writing successsively nested pairs, as in (1 . (2 . (3 . ())), they allow the sugar (1 2 3). This is the most common way of writing s-expressions, even in languages that allow raw cons cells (or "dotted pairs") to be written.
The s-cargot library refers to expressions where every right-branching sequence ends in an empty leaf as well-formed s-expressions. Note that any s-expression which can be written without using a dotted pair is necessarily well-formed.
Unfortunately, while in common use, s-expressions do not have a single formal standard. They are often defined in an ad-hoc way, which means that s-expressions used in different contexts will, despite sharing a common parentheses-delimited structure, differ in various respects. Additionally, because s-expressions are used as the concrete syntax for languages of the Lisp family, they often have conveniences (such as comment syntaxes) and other bits of syntactic sugar (such as reader macros, which are described more fully later) that make parsing them much more complicated. Even ignoring those features, the atoms recognized by a given s-expression variation can differ widely.
The s-cargot library was designed to accomodate several different kinds of s-expression formats, so that an s-expression format can be easily expressed as a combination of existing features. It includes a few basic variations on s-expression languages as well as the tools for parsing and emitting more elaborate s-expressions variations without having to reimplement the basic plumbing yourself.
Using the Library
The central way of interacting with the S-Cargot library is by creating and modifying datatypes which represent specifications for parsing and printing s-expressions. Each of those types has two type parameters, which are often called atom and carrier:
+------ the type that represents an atom or value
|
| +- the Haskell representation of the SExpr itself
| |
parser :: SExprParser atom carrier
printer :: SExprPrinter atom carrier
Various functions will be provided that modify the carrier type (i.e. the output type of parsing or input type of serialization) or the language recognized by the parsing.
Representing S-expressions
There are three built-in representations of S-expression lists: two of them are isomorphic, as one or the other might be convenient for working with S-expression data in a particular circumstance, while the third represents only the "well-formed" subset of possible S-expressions, which is often convenient when using s-expressions for configuration or data storage.
-- cons-based representation
data SExpr atom
= SCons (SExpr atom) (SExpr atom)
| SNil
| SAtom atom
-- list-based representation
data RichSExpr atom
= RSList [RichSExpr atom]
| RSDotList [RichSExpr atom] atom
| RSAtom atom
-- well-formed representation
data WellFormedSExpr atom
= WFSList [WellFormedSExpr atom]
| WFSAtom atom
The WellFormedSExpr representation should be structurally identical to the RichSExpr representation in all cases where no improper lists appear in the source. Both of those representations are often more convenient than writing multiple nested SCons constructors, in the same way that the [1,2,3] syntax in Haskell is often less tedious than writing 1:2:3:[].
Functions for converting back and forth between representations are provided, but you can also modify a SExprSpec to parse to or serialize from a particular representation using the asRich and asWellFormed functions.
>>> decode basicParser "(a b)"
Right [SCons (SAtom "a") (SCons (SAtom "b") SNil)]
>>> decode (asRich basicParser) "(a b)"
Right [RSList [RSAtom "a",RSAtom "b"]]
>>> decode (asWellFormed basicParser) "(a b)"
Right [WFSList [WFSAtom "a",WFSAtom "b"]]
>>> decode basicParser "(a . b)"
Right [SCons (SAtom "a") (SAtom "b")]
>>> decode (asRich basicParser) "(a . b)"
Right [RSDotted [RSAtom "a"] "b"]
>>> decode (asWellFormed basicParser) "(a . b)"
Left "Found atom in cdr position"
These names and patterns can be quite long, especially when you're constructing or matching on S-expression representations in Haskell source, so S-Cargot also exports several pattern synonyms that can be used both as expressions and in pattern-matching. These are each contained in their own module, as their names conflict with each other, so it's recommended to only import the module corresponding to the type that you plan on working with:
>>> import Data.SCargot.Repr.Basic
>>> A 2 ::: A 3 ::: A 4 ::: Nil
SCons (SAtom 2) (SCons (SAtom 3) (SCons (SAtom 4) SNil))
>>> import Data.SCargot.Repr.WellFormed
>>> L [A 1,A 2,A 3]
WFSList [WFSAtom 1,WFSAtom 2,WFSAtom 3]
>>> let sexprSum (L xs) = sum (map sexprSum xs); sexprSum (A n) = n
>>> :t sexprSum
sexprSum :: Num a => WellFormedSExpr a -> a
>>> sexprSum (L [A 2, L [A 3, A 4]])
9
If you are using GHC 7.10 or later, several of these will be powerful bidirectional pattern synonyms that allow both constructing and pattern-matching on s-expressions in non-trivial ways:
>>> import Data.SCargot.Repr.Basic
>>> L [ A 2, A 3, A 4 ]
SCons (SAtom 2) (SCons (SAtom 3) (SCons (SAtom 4) SNil))
Atom Types
Any type can serve as an underlying atom type in an S-expression parser or serializer, provided that it has a Parsec parser or a serializer (i.e. a way of turning it into Text.) For these examples, I'm going to use a very simple serializer that is roughly like the one found in Data.SCargot.Basic, which parses symbolic tokens of letters, numbers, and some punctuation characters. This means that the 'serializer' here is just the identity function which returns the relevant Text value:
parser :: SExprParser Text (SExpr Text)
parser = mkParser (pack <$> many1 (alphaNum <|> oneOf "+-*/!?"))
printer :: SExprPrinter Text (SExpr Text)
printer = flatPrint id
A more elaborate atom type might distinguish between different varieties of token. A small example (that understands just alphabetic identifiers and decimal numbers) would look like this:
import Data.Text (Text, pack)
data Atom = Ident Text | Num Int deriving (Eq, Show)
pAtom :: Parser Atom
pAtom = ((Num . read) <$> many1 digit)
<|> (Ident . pack) <$> takeWhile1 isAlpha)
sAtom :: Atom -> Text
sAtom (Ident t) = t
sAtom (Num n) = pack (show n)
myParser :: SExprParser Atom (SExpr Atom)
myParser = mkParser pAtom
myPrinter :: SExprPrinter Atom (SExpr Atom)
myPrinter = flatPrint sAtom
We can then use this newly created atom type within an S-expression for both parsing and serialization:
>>> decode myParser "(foo 1)"
Right [SCons (SAtom (Ident "foo")) (SCons (SAtom (Num 1)) SNil)]
>>> encode mySpec [L [A (Num 0), A (Ident "bar")]]
"(0 bar)"
Several common atom types appear in the module Data.SCargot.Common, including various kinds of identifiers and number literals. The long-term plan for S-Cargot is to include more and more kinds of built-in atoms, in order to make putting together an S-Expression parser even easier. If you have a common syntax for an atom type that you think should be represented there, please suggest it in an issue!
To make it easier to build up parsers for atom types without having to use Parsec manually, S-Cargot also exports Data.SCargot.Atom, which provides a shorthand way of building up a SExprParser from a list of parser-constructor pairs:
import Data.SCargot.Atom (atom, mkParserFromAtoms)
import Data.SCargot.Common (parseR7RSIdent, signedDecNumber)
-- we want our atom type to understand R7RS identifiers and
-- signed decimal numbers
data Atom
= Ident Text
| Num Integer
deriving (Eq, Show)
myParser :: SExprParser Atom (SExpr Atom)
myParser = mkAtomParser
[ atom Ident parseR7RSIdent
, atom Num signedDecNumber
]
Carrier Types
As pointed out above, there are three different "carrier" types that are used to represent S-expressions by the library, but you can use any type as a carrier type for a spec. This is particularly useful when you want to parse into your own custom tree-like type. For example, if we wanted to parse a small S-expression-based arithmetic language, we could define a data type and transformations from and to an S-expression type:
import Data.Char (isDigit)
import Data.Text (Text)
import qualified Data.Text as T
data Expr = Add Expr Expr | Num Int deriving (Eq, Show)
toExpr :: RichSExpr Text -> Either String Expr
toExpr (L [A "+", l, r]) = Add <$> toExpr l <*> toExpr r
toExpr (A c)
| T.all isDigit c = pure (Num (read (T.unpack c)))
| otherwise = Left "Non-numeric token as argument"
toExpr _ = Left "Unrecognized s-expr"
fromExpr :: Expr -> RichSExpr Text
fromExpr (Add x y) = L [A "+", fromExpr x, fromExpr y]
fromExpr (Num n) = A (T.pack (show n))
then we could use the convertSpec function to add this directly to the SExprSpec:
>>> let parser' = setCarrier toExpr (asRich myParser)
>>> :t parser'
SExprParser Atom Expr
>>> decode parser' "(+ 1 2)"
Right [Add (Num 1) (Num 2)]
>>> decode parser' "(0 1 2)"
Left "Unrecognized s-expr"
Comments
By default, an S-expression parser does not include a comment syntax, but the provided withLispComments function will cause it to understand traditional Lisp line-oriented comments that begin with a semicolon:
>>> decode basicParser "(this ; has a comment\n inside)\n"
Left "(line 1, column 7):\nunexpected \";\"\nexpecting space or atom"
>>> decode (withLispComments basicParser) "(this ; has a comment\n inside)\n"
Right [SCons (SAtom "this") (SCons (SAtom "inside") SNil)]
Additionally, you can provide your own comment syntax in the form of an Parsec parser. Any Parsec parser can be used, so long as it meets the following criteria:
- it is capable of failing (as is called until SCargot believes that there are no more comments)
- it does not consume any input in the case of failure, which may involve wrapping the parser in a call to
try
For example, the following adds C++-style comments to an S-expression format:
>>> let cppComment = string "//" >> manyTill newline >> return ()
>>> decode (setComment cppComment basicParser) "(a //comment\n b)\n"
Right [SCons (SAtom "a") (SCons (SAtom "b") SNil)]
The Data.SCargot.Comments module defines some helper functions for creating comment syntaxes, so the cppComment parser above could be defined as simply
>>> let cppComment = lineComment "//"
>>> decode (setComment cppComment basicParser) "(a //comment\n b)\n"
Right [SCons (SAtom "a") (SCons (SAtom "b") SNil)]
Additionally, a handful of common comment syntaxes are defined in Data.SCargot.Comments, including C-style, Haskell-style, and generic scripting-language-style comments, so in practice, we could write the above example as
>>> decode (withCLikeLineComments basicParser) "(a //comment\n b)\n"
Right [SCons (SAtom "a") (SCons (SAtom "b") SNil)]
Reader Macros
In Lisp variants, a reader macro is a macro---a function that operates on syntactic structures---which is invoked during the scanning, or lexing, phase of a Lisp parser. This allows the lexical syntax of a Lisp to be modified. A very common reader macro in most Lisp variants is the single quote, which allows the syntax 'expr to stand as sugar for the literal s-expression (quote expr). The S-Cargot library accomodates this by keeping a map from characters to Haskell functions that can be used analogously to reader macros. This is a common enough special case that there are shorthand ways of writing this, but we could support the 'expr syntax by creating a Haskell function to turn expr into (quote expr) and adding that as a reader macro associated with the character ':
>>> let quote expr = SCons (SAtom "quote") (SCons expr SNil)
>>> :t quote
quote :: IsString atom => SExpr atom -> SExpr atom
>>> let addQuoteReader = addReader '\'' (\ parse -> fmap quote parse)
>>> addQuoteReader :: IsString atom => SExprParser atom c -> SExprParser atom c
>>> decode (addQuoteReader basicParser) "'foo"
Right [SCons (SAtom "quote") (SCons (SAtom "foo") SNil)]
A reader macro is passed the an s-expression parser so that it can perform recursive parse calls, and it can return any SExpr it would like. It may also take as much or as little of the remaining parse stream as it would like. For example, the following reader macro does not bother parsing anything else and merely returns a new token:
>>> let qmReader = addReader '?' (\ _ -> pure (SAtom "huh"))
>>> decode (qmReader basicParser) "(?1 2)"
Right [SCons (SAtom "huh") (SCons (SAtom "1") (SCons (SAtom "2") SNil))]
We can define a similar reader macro directly in Common Lisp, although it's important to note that Common Lisp converts all identifiers to uppercase, and also that the quote in line [3] is necessary so that the Common Lisp REPL doesn't attempt to evaluate (huh 1 2) as code:
[1]> (defun qm-reader (stream char) 'huh)
QM-READER
[2]> (set-macro-character #\? #'qm-reader)
T
[3]> '(?1 2)
(HUH 1 2)
Reader macros in S-Cargot can be used to define bits of Lisp syntax that are not typically considered the purview of S-expression parsers. For example, some Lisp-derived languages allow square brackets as a subsitute for proper lists, and to support this we could define a reader macro that is indicated by the [ character and repeatedly calls the parser until a ] character is reached:
>>> let vec p = (char ']' *> pure SNil) <|> (SCons <$> p <*> vec p)
>>> :t vec
vec
:: Stream s m Char =>
ParsecT s u m (SExpr atom) -> ParsecT s u m (SExpr atom)
>>> let withVecReader = addReader '[' vec
>>> decode (asRich (withVecReader basicParser)) "(1 [2 3])"
Right [RSList [RSAtom "1",RSList [RSAtom "2",RSAtom "3"]]]
Pretty-Printing and Indentation
The s-cargot library also includes a simple but often adequate pretty-printing system for S-expressions. A printer that prints a single-line s-expression is created with flatPrint:
>>> let printer = flatPrint id
>>> :t printer
SExprPrinter Text (SCargot Text)
>>> Text.putStrLn $ encode printer [L [A "foo", A "bar"]]
(foo bar)
A printer that tries to pretty-print an s-expression to fit attractively within an 80-character limit can be created with basicPrint:
>>> let printer = basicPrint id
>>> let sentence = "this stupendously preposterously supercalifragilisticexpialidociously long s-expression"
>>> let longSexpr = L [A word | word <- Text.words sentence ]
>>> Text.putStrLn $ encodeOne printer longSexpr
(this
stupendously
preposterously
supercalifragilisticexpialidociously
long
s-expression)
A printer created with basicPrint will "swing" things that are too long onto the subsequent line, indenting it a fixed number of spaces. We can modify the number of spaces with setIndentAmount:
>>> let printer = setIndentAmount 4 (basicPrint id)
>>> Text.putStrLn $ encodeOne printer longSexpr
(this
stupendously
preposterously
supercalifragilisticexpialidociously
long
s-expression)
We can also modify what counts as the 'maximum width', which for a basicPrint printer is 80 by default:
>>> let printer = setMaxWidth 8 (basicPrint id)
>>> Text.putStrLn $ encodeOne printer (L [A "one", A "two", A "three"])
(one
two
three)
Or remove the maximum, which will always put the whole s-expression onto one line, regardless of its length:
>>> let printer = removeMaxWidth (basicPrint id)
>>> Text.putStrLn $ encodeOne printer longSexpr
(this stupendously preposterously supercalifragilisticexpialidociously long s-expression)
We can also specify an indentation strategy, which decides how to indent subsequent expressions based on the head of a given expression. The default is to always "swing" subsequent expressions to the next line, but we could also specify the Align constructor, which will print the first two expressions on the same line and then any subsequent expressions horizontally aligned with the second one, like so:
>>> let printer = setIndentStrategy (\ _ -> Align) (setMaxWidth 8 (basicPrint id))
>>> Text.putStrLn $ encodeOne printer (L [A "one", A "two", A "three", A "four"])
(one two
three
four)
Or we could choose to keep some number of expressions on the same line and afterwards swing the subsequent ones:
>>> let printer = setIndentStrategy (\ _ -> SwingAfter 1) (setMaxWidth 8 (basicPrint id))
>>> Text.putStrLn $ encodeOne printer (L [A "one", A "two", A "three", A "four"])
(one two
three
four)
In many situations, we might want to choose a different indentation strategy based on the first expression within a proper list: for example, Common Lisp source code is often formatted so that, following a defun token, the function name and arguments are on the same line, and then the body of the function is indented a fixed amount. We can express an approximation of that strategy like this:
>>> let strategy (A ident) | "def" `Text.isPrefixOf` ident = SwingAfter 2; strategy _ = Align
>>> let printer = setIndentStrategy strategy (setMaxWidth 20 (basicPrint id))
>>> let fact = L [A "defun", A "fact", L [A "x"], L [A "product", L [A "range", A "1", A "x"]]]
>>> Text.putStrLn $ encodeOne printer fact
(defun fact (x)
(product (range 1 x)))
>>> let app = L [A "apply", L [A "lambda", L [A "y"], L [A "fact", A "y"]], L [A "+", A "2", A "3"]]
(apply (lambda (y) (fact y)
(+ 2 3))
Putting It All Together
Here is a final example which implements a limited arithmetic language with Haskell-style line comments and a special reader macro to understand hex literals:
{-# LANGUAGE OverloadedStrings #-}
module SCargotExample where
import Control.Applicative ((<|>))
import Data.Char (isDigit)
import Data.SCargot
import Data.SCargot.Repr.Basic
import Data.Text (Text, pack)
import Numeric (readHex)
import Text.Parsec (anyChar, char, digit, many1, manyTill, newline, satisfy, string)
import Text.Parsec.Text (Parser)
-- Our operators are going to represent addition, subtraction, or
-- multiplication
data Op = Add | Sub | Mul deriving (Eq, Show)
-- The atoms of our language are either one of the aforementioned
-- operators, or positive integers
data Atom = AOp Op | ANum Int deriving (Eq, Show)
-- Once parsed, our language will consist of the applications of
-- binary operators with literal integers at the leaves
data Expr = EOp Op Expr Expr | ENum Int deriving (Eq, Show)
-- Conversions to and from our Expr type
toExpr :: SExpr Atom -> Either String Expr
toExpr (A (AOp op) ::: l ::: r ::: Nil) = EOp op <$> toExpr l <*> toExpr r
toExpr (A (ANum n)) = pure (ENum n)
toExpr sexpr = Left ("Unable to parse expression: " ++ show sexpr)
fromExpr :: Expr -> SExpr Atom
fromExpr (EOp op l r) = A (AOp op) ::: fromExpr l ::: fromExpr r ::: Nil
fromExpr (ENum n) = A (ANum n) ::: Nil
-- Parser and serializer for our Atom type
pAtom :: Parser Atom
pAtom = ((ANum . read) <$> many1 digit)
<|> (char '+' *> pure (AOp Add))
<|> (char '-' *> pure (AOp Sub))
<|> (char '*' *> pure (AOp Mul))
sAtom :: Atom -> Text
sAtom (AOp Add) = "+"
sAtom (AOp Sub) = "-"
sAtom (AOp Mul) = "*"
sAtom (ANum n) = pack (show n)
-- Our comment syntax is going to be Haskell-like:
hsComment :: Parser ()
hsComment = string "--" >> manyTill anyChar newline >> return ()
-- Our custom reader macro: grab the parse stream and read a
-- hexadecimal number from it:
hexReader :: Reader Atom
hexReader _ = (A . ANum . rd) <$> many1 (satisfy isHexDigit)
where isHexDigit c = isDigit c || c `elem` hexChars
rd = fst . head . readHex
hexChars :: String
hexChars = "AaBbCcDdEeFf"
-- Our final s-expression parser and printer:
myLangParser :: SExprParser Atom Expr
myLangParser
= setComment hsComment -- set comment syntax to be Haskell-style
$ addReader '#' hexReader -- add hex reader
$ setCarrier toExpr -- convert final repr to Expr
$ mkParser pAtom -- create spec with Atom type
mkLangPrinter :: SExprPrinter Atom Expr
mkLangPrinter
= setFromCarrier fromExpr
$ setIndentStrategy (const Align)
$ basicPrint sAtom
>>> decode myLangParser "(+ (* 2 20) 10) (* 10 10)"
[EOp Add (EOp Mul (ENum 2) (ENum 20)) (ENum 10),EOp Mul (ENum 10) (ENum 10)]
Keep in mind that you often won't need to write all this by hand, as you can often use a variety of built-in atom types, reader macros, comment types, and representations, but it's a useful illustration of all the options that are available to you should you need them!