Create parsers using parser combinators with helpful error messages
npm install loplop is a library to create parsers using parser combinators with helpful errors.
``javascript
function parse(tokens) {
var parser = lop.Parser();
return parser.parseTokens(expressionRule, tokens);
}
// This rule is wrapped inside lop.rule to defer evaluation until
// the rule is used -- otherwise, it would reference integerRule
// and ifRule, which don't exist yet.
var expressionRule = lop.rule(function() {
return rules.firstOf("expression",
integerRule,
ifRule
);
});
var integerRule = rules.then(
rules.tokenOfType("integer"),
function(value) {
return new IntegerNode(parseInt(value, 10));
}
);
var ifRule = rules.sequence(
rules.token("keyword", "if"),
rules.sequence.cut(),
rules.sequence.capture(expressionRule),
rules.token("keyword", "then"),
rules.sequence.capture(expressionRule),
rules.token("keyword", "else"),
rules.sequence.capture(expressionRule)
).map(function(condition, trueBranch, falseBranch) {
return new IfNode(condition, trueBranch, falseBranch);
});
`
lop tries to provide helpful errors where possible. For instance, in ifRule
as defined above, there is a cut following the keyword . Before the cut,if
if we fail to match the input, we can backtrack -- in this case, we backtrack
and see if another form of expression might match the input. However, after the
cut, we prevent backtracking. Once we've see the keyword , there's no doubt"if 1 42 else 12"
about which sort of expression this is, so if parsing fails later in this rule,
there's no point in backtracking. This allows informative error messages to be
generated: if we try to parse the string , we get the error:
Error: File: /tmp/lop-example
Line number: 1
Character number: 6:
Expected keyword "then"
but got integer "42"
When using a parser built with lop, the input is an array of tokens. A token can be any value so long as it has the property source, which must be a StringSourceRange.
The easiest way to create a tokeniser is using lop's regex tokeniser.
A regex tokeniser can be constructed by calling new lop.RegexTokeniser(rules),rules
where is a list of token rules.name
A token rule should have a property that uniquely identifies that rule,regex
and a property that is an instance of RegExp describing the token.
Calling tokenise with a string will return a list of tokens.
Each token has three properties:
* type
*
*
The tokeniser will apply the regex from each rule in order at the current position.
The current position is initially zero, the start of the string.
The first rule with a matching regex is used to produce a token,
with the token's value being the first capture of the regex,undefined
or if the regex does not define any capture groups.unrecognisedCharacter
The current position is incremented to the index of the first character unmatched by the regex.
If no rule matches at the current position,
a single character token is produced,
and the current position is incremented by one.
For instance, to create a simple tokeniser that generates a stream of words tokens separated by whitespace tokens.
`javascript
var lop = require("lop");
var rules = [
{
name: "identifier",
regex: /(\s+)/
},
{
name: "whitespace",
regex: /(\S+)/
}
];
var tokeniser = new lop.RegexTokeniser(rules);
tokeniser.tokenise(input);
`
You can also create your own tokeniser.
For instance, to create a simple tokeniser that generates a stream of words tokens separated by whitespace tokens:
`javascript
var StringSource = require("lop").StringSource;
function tokeniseString(string) {
return tokenise(new StringSource(string, "raw string"));
}
function tokenise(source) {
var string = source.asString();
var whitespaceRegex = /(\s+)/g;
var result;
var start = 0;
var parts = [];
while ((result = whitespaceRegex.exec(source)) !== null) {
parts.push({
type: "word",
value: string.substring(start, result.index),
source: source.range(start, result.index)
});
parts.push({
type: "whitespace",
value: result[1],
source: source.range(result.index, whitespaceRegex.lastIndex)
});
start = whitespaceRegex.lastIndex;
}
parts.push({
type: "word",
value: string.substring(start),
source: source.range(start, string.length)
});
parts.push({
type: "end",
source: source.range(string.length, string.length)
});
return parts.filter(function(part) {
return part.type !== "word" || part.value !== "";
});
}
`
lop also defines its own notion of a token. Each instance of lop.Token has a type, name, and source, similarly to most of the tokens that would be created by the token above. For instance, instead of:
{
type: "word",
value: value,
source: source
}
you could use:
new Token("word", value, source)
The main advantage of using lop.Token is that you can then use the rules lop.rules.token and lop.rules.tokenOfType (described later). If you don't use lop.Token, you must define your own atomic rules, but you can use the other rules without any modifications.
To parse an array of tokens, you can call the method parseTokens on lop.Parser, passing in the parsing rule and the array of tokens. For instance, assuming we already have a tokenise function (the one above would do fine):
`javascript
function parseSentence(source) {
var tokens = tokenise(source);
var parser = new lop.Parser();
var parseResult = parser.parseTokens(sentenceRule, tokens);
if (!parseResult.isSuccess()) {
throw new Error("Failed to parse: " + describeFailure(parseResult));
}
return parseResult.value();
}
function describeFailure(parseResult) {
return parseResult.errors().map(describeError).join("\n");
function describeError(error) {
return error.describe();
}
}
`
The result of parsing can be success, failure, or error. While failure indicates
that the rule didn't match the input tokens, error indicates that the input
was invalid in some way. In general, rules will backtrack when they
encounter a failure, but will completely abort when they encounter an error.
Each of these results has a number of methods:
`javascript`
result.isSuccess() // true for success, false otherwise
result.isFailure() // true for failure, false otherwise
result.isError() // true for error, false otherwise
result.value() // if success, the value that was parsed
result.remaining() // if success, the tokens that weren't consumed by parsing
result.source() // the StringSourceRange containing the consumed tokens
result.errors() // if failure or error, an array of descriptions of the failure/error
The final question is then: how do we define rules for the parser, such as the currently undefined sentenceRule?
Each rule in lop accepts an iterator over tokens, and returns a result, as
described in the previous section.
Success if the next token has type tokenType and value value, failure
otherwise. Value on success is the value of the token.
Success if the next token has type tokenType, failure otherwise. Value on
success is the value of the token.
Tries each rule in subRules on the input tokens in turn. We return the result
from the first sub-rule that returns success or error. In other words, return the
result from the first sub-rule that doesn't return failure. If all sub-rules return
failure, this rule returns failure.
Try subRule on the input tokens, and if successful, map over the result. For
instance:
`javascript`
lop.rules.then(
lop.rules.tokenOfType("integer"),
function(tokenValue) {
return parseInt(tokenValue, 10);
}
)
Try subRule on the input tokens. If the sub-rule is successful with the valuevalue, then return success with the value options.some(value). If the sub-rule fails, returnoptions.none`. If the sub-rules errors, return that error.
success with the value