Clojure bindings for the ANTLR 4 parser library, an adaptive LL(*) parser. Looks a lot like Instaparse, only much faster, with richer grammar definitions, and none of Instaparse's delightful features.
Just add clj-antlr to your project.clj, and load a grammar file at runtime.
No ANTLR installation is required; clj-antlr will load the grammar for you, no compilation needed. No macros, either! Running the parser in interpreted mode is a tad slower than the compiled parsers that Antlr can emit, but means a lot less hassle for folks to get started.
user=> (require ['clj-antlr.core :as 'antlr])
nil
user=> (def json (antlr/parser "grammars/Json.g4"))
#'user/json
user=> (pprint (json "[1,2,3]"))
(:jsonText
(:jsonArray
"["
(:jsonValue (:jsonNumber "1"))
","
(:jsonValue (:jsonNumber "2"))
","
(:jsonValue (:jsonNumber "3"))
"]"))
Parsers act like functions, and can take strings, InputStreams, and Readers as their arguments. They emit trees of lists: each list begins with the keyword node name, and is followed by the nodes' children. Terminal nodes are represented as strings.
You can define parsers directly with strings, too. ANTLR 4.2 will complain but compile successfully; 4.2.1 will include a patch to fix that.
user=> (def aaa (antlr/parser "grammar Aaa;
#_=> aaa : AA+;
#_=> AA : [Aa]+ ;
#_=> WS : ' ' -> channel(HIDDEN) ;"))
#'user/aaa
user=> (aaa "aAAaa A aAA AAAAaAA")
(:aaa "aAAaa" "A" "aAA" "AAAAaAA")
Works with split (lexer and parser) grammars:
clj::clj-antlr.core-test=> (let [g (parser "grammars/L.g4" "grammars/T.g4" {})] (g "abbc"))
(:s "a" "b" "b" "c")
ANTLR can recover from errors in mid-parse by performing single-token insertion and single-token deletion on mismatched error tokens, where possible. This means an ANTLR parse may throw an error, but still produce useful parse information; or produce multiple errors. Parsing an invalid string will throw an exception with a textual explanation of the errors encountered:
user=> (json "[1,2,,3,]")
ParseError extraneous input ',' expecting {'null', '{', '[', 'false', 'true', NUMBER, STRING}
mismatched input ']' expecting {'null', '{', '[', 'false', 'true', NUMBER, STRING} clj-antlr.common/parse-error (common.clj:106)
But wait, there's more! ParseErrors are deref-able, yielding detailed debugging information:
user=> (try (json "[1,2,,3,]") (catch clj_antlr.ParseError e (pprint @e)))
({:symbol #<CommonToken [@5,5:5=',',<4>,1:5]>,
:line 1,
:char 5,
:message
"extraneous input ',' expecting {'null', '{', '[', 'false', 'true', NUMBER, STRING}"}
{:token #<CommonToken [@8,8:8=']',<1>,1:8]>,
:expected #<IntervalSet {2..3, 5, 7, 9..10, 12}>,
:state 25,
:rule #<InterpreterRuleContext [51 15]>,
:stack ("jsonText" "jsonArray" "jsonValue"),
:symbol #<CommonToken [@8,8:8=']',<1>,1:8]>,
:line 1,
:char 8,
:message
"mismatched input ']' expecting {'null', '{', '[', 'false', 'true', NUMBER, STRING}"})
You can use the line and char numbers, in addition to the messages, to guide the user in generating correct syntax. Clj-antlr handles both lexer and parser errors; though the debugging information available at different passes may vary.
user=> (try (json "⊂") (catch clj_antlr.ParseError e (pprint @e)))
({:token nil,
:expected nil,
:state -1,
:rule nil,
:symbol nil,
:line 1,
:char 0,
:message "token recognition error at: '⊂'"}
{:token #<CommonToken [@0,1:0='<EOF>',<-1>,1:1]>,
:expected #<IntervalSet {3, 5}>,
:state 16,
:rule #<InterpreterRuleContext []>,
:symbol #<CommonToken [@0,1:0='<EOF>',<-1>,1:1]>,
:line 1,
:char 1,
:message "no viable alternative at input '<EOF>'"})
clj-antlr will still produce parse trees from invalid input. Use the {:throw?
false} option, either when constructing the parser, or as an argument to the
parse
function.
user=> (->> "[1,2" (antlr/parse json {:throw? false}) pprint)
(:jsonText
(:jsonArray
"["
(:jsonValue (:jsonNumber "1"))
","
(:jsonValue (:jsonNumber "2"))))
Any parse errors will be available as metadata on the returned tree:
user=> (->> "[1,2" (antlr/parse json {:throw? false}) meta :errors pprint)
({:token #<CommonToken [@4,4:3='<EOF>',<-1>,1:4]>,
:expected #<IntervalSet {1, 4}>,
:state 54,
:rule #<InterpreterRuleContext [15]>,
:symbol #<CommonToken [@4,4:3='<EOF>',<-1>,1:4]>,
:line 1,
:char 4,
:message "no viable alternative at input '<EOF>'"})
Sometimes, clj-antlr is able to identify invalid rules in the parse tree, and
wrap them with a :clj-antlr/error
node.
user=> (->> "[1, {\"foo\"::}, 3]" (antlr/parse json {:throw? false}) pprint)
(:jsonText
(:jsonArray
"["
(:jsonValue (:jsonNumber "1"))
","
(:jsonValue
(:jsonObject
"{"
(:member "\"foo\"" ":" (:clj-antlr/error (:jsonValue ":")))
"}"))
","
(:jsonValue (:jsonNumber "3"))
"]"))
But not always. This input generates errors in the top-level :errors metadata map, but creates an invalid parse tree without any error nodes. I think this is a bug in clj-antlr or ANTLR itself; if you have suggestions, I'd like to hear them.
user=> (->> "[1,,3]" (antlr/parse json {:throw? false}) pprint)
(:jsonText
(:jsonArray
"["
(:jsonValue (:jsonNumber "1"))
","
(:jsonValue "," (:jsonNumber "3"))
"]"))
All options may be passed at parser construction time:
user=> (antlr/parse (antlr/parser "grammars/Cadr.g4" {:case-sensitive? false})
"CdDr")
(:cadr "C" "d" "D" "r")
... and also overridden at parse time via antlr.core/parse
:
user=> (antlr/parse (antlr/parser "grammars/Cadr.g4")
{:case-sensitive? false}
"CdDr")
(:cadr "C" "d" "D" "r")
user=> (doc antlr/parser)
-------------------------
clj-antlr.core/parser
([filename] [filename opts])
Constructs a new parser. Takes a filename for an Antlr v4 grammar. Options:
:format The parse tree to generate. One of
:sexp (default) Nested lists, node names first
:raw Equivalent to identity
<any function> Takes a map of {:tree, :parser, etc}
:root The string name of the rule to begin parsing. Defaults to
the first rule in the grammar.
:throw? If truthy, parse errors will be thrown. Defaults true.
:case-sensitive? Whether the lexer must match the exact case of characters.
Defaults true. If false, the tokenizer will only receive
lowercase characters. The generated parse tree will still
retain the case of the original text.
:use-alternates? If truthy, uses the alternate name for a node, rather than
the rule name. Defaults false.
Here's a ton of ANTLR 4 parsers for various languages!
On a real-world 3.5KB JSON object, clj-antlr with a typical JSON grammar is about 100 times faster than an identical AST built by an Instaparse grammar. Since Instaparse doesn't really have a separation between grammar and lexer rules, I'm using regular expressions for strings, ints, etc; but the transformation between grammars is pretty straightforward.
kingsbury@hackbook:~/clj-antlr master$ lein test :perf
Benchmarking instaparse
WARNING: Final GC required 1.645508727049022 % of runtime
Evaluation count : 660 in 60 samples of 11 calls.
Execution time mean : 97.557634 ms
Execution time std-deviation : 5.132833 ms
Execution time lower quantile : 91.651987 ms ( 2.5%)
Execution time upper quantile : 108.289375 ms (97.5%)
Overhead used : 10.328888 ns
Found 4 outliers in 60 samples (6.6667 %)
low-severe 3 (5.0000 %)
low-mild 1 (1.6667 %)
Variance from outliers : 38.4948 % Variance is moderately inflated by outliers
Benchmarking clj-antlr
Evaluation count : 64440 in 60 samples of 1074 calls.
Execution time mean : 958.366202 µs
Execution time std-deviation : 36.434070 µs
Execution time lower quantile : 901.210266 µs ( 2.5%)
Execution time upper quantile : 1.032678 ms (97.5%)
Overhead used : 10.328888 ns
Ran 1 tests containing 1 assertions.
0 failures, 0 errors.
Copyright © 2014 Kyle Kingsbury [email protected], and Factual, Inc. Includes ANTLR code under the BSD 3-clause license, written by Terence Parr and Sam Harwell. My sincerest appreciation to all ANTLR contributors as well. :)
Distributed under the Eclipse Public License, the same as Clojure.