Files
openlayers/old/test/spec/ol/expr/lexer.test.js
2013-11-20 11:39:21 +01:00

510 lines
16 KiB
JavaScript

goog.provide('ol.test.expression.Lexer');
describe('ol.expr.Lexer', function() {
describe('constructor', function() {
it('creates a new lexer', function() {
var lexer = new ol.expr.Lexer('foo');
expect(lexer).to.be.a(ol.expr.Lexer);
});
});
describe('#next()', function() {
it('returns one token at a time', function() {
var source = 'foo === "bar"';
var lexer = new ol.expr.Lexer(source);
// scan first token
var token = lexer.next();
expect(token.type).to.be(ol.expr.TokenType.IDENTIFIER);
expect(token.value).to.be('foo');
// scan second token
token = lexer.next();
expect(token.type).to.be(ol.expr.TokenType.PUNCTUATOR);
expect(token.value).to.be('===');
// scan third token
token = lexer.next();
expect(token.type).to.be(ol.expr.TokenType.STRING_LITERAL);
expect(token.value).to.be('bar');
// scan again
token = lexer.next();
expect(token.type).to.be(ol.expr.TokenType.EOF);
// and again
token = lexer.next();
expect(token.type).to.be(ol.expr.TokenType.EOF);
});
});
describe('#peek()', function() {
var lexer;
beforeEach(function() {
lexer = new ol.expr.Lexer('foo > 42 && bar == "chicken"');
});
it('looks ahead without consuming token', function() {
var token = lexer.peek();
expect(token.type).to.be(ol.expr.TokenType.IDENTIFIER);
expect(token.value).to.be('foo');
token = lexer.next();
expect(token.type).to.be(ol.expr.TokenType.IDENTIFIER);
expect(token.value).to.be('foo');
});
it('works after a couple scans', function() {
var token = lexer.next();
expect(token.type).to.be(ol.expr.TokenType.IDENTIFIER);
expect(token.value).to.be('foo');
token = lexer.next();
expect(token.type).to.be(ol.expr.TokenType.PUNCTUATOR);
expect(token.value).to.be('>');
token = lexer.peek();
expect(token.type).to.be(ol.expr.TokenType.NUMERIC_LITERAL);
expect(token.value).to.be(42);
token = lexer.next();
expect(token.type).to.be(ol.expr.TokenType.NUMERIC_LITERAL);
expect(token.value).to.be(42);
});
it('returns the same thing when called multiple times', function() {
var token = lexer.peek();
expect(token.type).to.be(ol.expr.TokenType.IDENTIFIER);
expect(token.value).to.be('foo');
for (var i = 0; i < 10; ++i) {
token = lexer.peek();
expect(token.type).to.be(ol.expr.TokenType.IDENTIFIER);
expect(token.value).to.be('foo');
}
});
});
describe('#scanIdentifier_()', function() {
function scan(source, part) {
var lexer = new ol.expr.Lexer(source);
var token = lexer.scanIdentifier_(lexer.getCurrentCharCode_());
if (!part) {
expect(token.index).to.be(0);
expect(lexer.peek().type).to.be(ol.expr.TokenType.EOF);
}
return token;
}
it('works for short identifiers', function() {
var token = scan('a');
expect(token.value).to.be('a');
expect(token.type).to.be(ol.expr.TokenType.IDENTIFIER);
});
it('works for longer identifiers', function() {
var token = scan('foo');
expect(token.value).to.be('foo');
expect(token.type).to.be(ol.expr.TokenType.IDENTIFIER);
});
it('works for $ anywhere', function() {
var token = scan('$foo$bar$');
expect(token.value).to.be('$foo$bar$');
expect(token.type).to.be(ol.expr.TokenType.IDENTIFIER);
});
it('works for _ anywhere', function() {
var token = scan('_foo_bar_');
expect(token.value).to.be('_foo_bar_');
expect(token.type).to.be(ol.expr.TokenType.IDENTIFIER);
});
it('works for keywords', function() {
var token = scan('delete');
expect(token.value).to.be('delete');
expect(token.type).to.be(ol.expr.TokenType.KEYWORD);
});
it('works for null', function() {
var token = scan('null');
expect(token.value).to.be('null');
expect(token.type).to.be(ol.expr.TokenType.NULL_LITERAL);
});
it('works for boolean true', function() {
var token = scan('true');
expect(token.value).to.be('true');
expect(token.type).to.be(ol.expr.TokenType.BOOLEAN_LITERAL);
});
it('works for boolean false', function() {
var token = scan('false');
expect(token.value).to.be('false');
expect(token.type).to.be(ol.expr.TokenType.BOOLEAN_LITERAL);
});
it('works with unicode escape sequences', function() {
var token = scan('\u006f\u006c\u0033');
expect(token.value).to.be('ol3');
expect(token.type).to.be(ol.expr.TokenType.IDENTIFIER);
});
it('works with hex escape sequences', function() {
var token = scan('\x6f\x6c\x33');
expect(token.value).to.be('ol3');
expect(token.type).to.be(ol.expr.TokenType.IDENTIFIER);
});
it('throws for identifiers starting with a number', function() {
expect(function() {
scan('4foo');
}).throwException();
});
it('throws for identifiers starting with a punctuation char', function() {
expect(function() {
scan('!foo');
}).throwException();
});
it('only scans valid identifier part', function() {
var token = scan('foo>bar', true);
expect(token.value).to.be('foo');
expect(token.type).to.be(ol.expr.TokenType.IDENTIFIER);
});
});
describe('#scanNumericLiteral_()', function() {
function scan(source) {
var lexer = new ol.expr.Lexer(source);
var token = lexer.scanNumericLiteral_(lexer.getCurrentCharCode_());
expect(token.index).to.be(0);
expect(lexer.peek().type).to.be(ol.expr.TokenType.EOF);
return token;
}
it('works for integers', function() {
var token = scan('123');
expect(token.value).to.be(123);
expect(token.type).to.be(ol.expr.TokenType.NUMERIC_LITERAL);
});
it('throws for bogus integer', function() {
expect(function() {
scan('123z');
}).throwException(function(err) {
expect(err).to.be.an(ol.expr.UnexpectedToken);
var token = err.token;
expect(token.value).to.be('z');
expect(token.index).to.be(3);
});
});
it('works for float', function() {
var token = scan('123.456');
expect(token.value).to.be(123.456);
expect(token.type).to.be(ol.expr.TokenType.NUMERIC_LITERAL);
});
it('throws for bogus float', function() {
expect(function() {
scan('123.4x4');
}).throwException(function(err) {
expect(err).to.be.an(ol.expr.UnexpectedToken);
var token = err.token;
expect(token.value).to.be('x');
expect(token.index).to.be(5);
});
});
it('works with exponent', function() {
var token = scan('1.234e5');
expect(token.value).to.be(1.234e5);
expect(token.type).to.be(ol.expr.TokenType.NUMERIC_LITERAL);
});
it('works with explicit positive exponent', function() {
var token = scan('1.234e+5');
expect(token.value).to.be(1.234e5);
expect(token.type).to.be(ol.expr.TokenType.NUMERIC_LITERAL);
});
it('works with negative exponent', function() {
var token = scan('1.234e-5');
expect(token.value).to.be(1.234e-5);
expect(token.type).to.be(ol.expr.TokenType.NUMERIC_LITERAL);
});
it('throws for bogus float', function() {
expect(function() {
scan('1.234eo4');
}).throwException(function(err) {
expect(err).to.be.an(ol.expr.UnexpectedToken);
var token = err.token;
expect(token.value).to.be('o');
expect(token.index).to.be(6);
});
});
it('works with octals', function() {
var token = scan('02322');
expect(token.value).to.be(1234);
expect(token.type).to.be(ol.expr.TokenType.NUMERIC_LITERAL);
});
it('throws for bogus octal', function() {
// note that this is more strict than most es5 engines
expect(function() {
scan('02392');
}).throwException(function(err) {
expect(err).to.be.an(ol.expr.UnexpectedToken);
var token = err.token;
expect(token.value).to.be('9');
expect(token.index).to.be(3);
});
});
it('works with hex', function() {
var token = scan('0x4d2');
expect(token.value).to.be(1234);
expect(token.type).to.be(ol.expr.TokenType.NUMERIC_LITERAL);
});
it('throws for bogus hex', function() {
// note that this is more strict than most es5 engines
expect(function() {
scan('0x4G');
}).throwException(function(err) {
expect(err).to.be.an(ol.expr.UnexpectedToken);
var token = err.token;
expect(token.value).to.be('G');
expect(token.index).to.be(3);
});
});
});
describe('#scanPunctuator_()', function() {
function scan(source) {
var lexer = new ol.expr.Lexer(source);
var token = lexer.scanPunctuator_(lexer.getCurrentCharCode_());
expect(token.index).to.be(0);
expect(lexer.peek().type).to.be(ol.expr.TokenType.EOF);
return token;
}
it('works for dot', function() {
var token = scan('.');
expect(token.value).to.be('.');
expect(token.type).to.be(ol.expr.TokenType.PUNCTUATOR);
});
it('works for bang', function() {
var token = scan('!');
expect(token.value).to.be('!');
expect(token.type).to.be(ol.expr.TokenType.PUNCTUATOR);
});
it('works for double equal', function() {
var token = scan('==');
expect(token.value).to.be('==');
expect(token.type).to.be(ol.expr.TokenType.PUNCTUATOR);
});
it('works for triple equal', function() {
var token = scan('===');
expect(token.value).to.be('===');
expect(token.type).to.be(ol.expr.TokenType.PUNCTUATOR);
});
it('works for not double equal', function() {
var token = scan('!=');
expect(token.value).to.be('!=');
expect(token.type).to.be(ol.expr.TokenType.PUNCTUATOR);
});
it('works for not triple equal', function() {
var token = scan('!==');
expect(token.value).to.be('!==');
expect(token.type).to.be(ol.expr.TokenType.PUNCTUATOR);
});
it('works for logical or', function() {
var token = scan('||');
expect(token.value).to.be('||');
expect(token.type).to.be(ol.expr.TokenType.PUNCTUATOR);
});
it('works for logical and', function() {
var token = scan('&&');
expect(token.value).to.be('&&');
expect(token.type).to.be(ol.expr.TokenType.PUNCTUATOR);
});
it('works for plus', function() {
var token = scan('+');
expect(token.value).to.be('+');
expect(token.type).to.be(ol.expr.TokenType.PUNCTUATOR);
});
it('works for minus', function() {
var token = scan('-');
expect(token.value).to.be('-');
expect(token.type).to.be(ol.expr.TokenType.PUNCTUATOR);
});
it('works for star', function() {
var token = scan('*');
expect(token.value).to.be('*');
expect(token.type).to.be(ol.expr.TokenType.PUNCTUATOR);
});
it('works for slash', function() {
var token = scan('/');
expect(token.value).to.be('/');
expect(token.type).to.be(ol.expr.TokenType.PUNCTUATOR);
});
it('works for percent', function() {
var token = scan('%');
expect(token.value).to.be('%');
expect(token.type).to.be(ol.expr.TokenType.PUNCTUATOR);
});
});
describe('#scanStringLiteral_()', function() {
function scan(source) {
var lexer = new ol.expr.Lexer(source);
var token = lexer.scanStringLiteral_(lexer.getCurrentCharCode_());
expect(token.index).to.be(0);
expect(lexer.peek().type).to.be(ol.expr.TokenType.EOF);
return token;
}
it('parses double quoted string', function() {
var token = scan('"my string"');
expect(token.value).to.be('my string');
expect(token.type).to.be(ol.expr.TokenType.STRING_LITERAL);
});
it('parses double quoted string with internal single quotes', function() {
var token = scan('"my \'quoted\' string"');
expect(token.value).to.be('my \'quoted\' string');
expect(token.type).to.be(ol.expr.TokenType.STRING_LITERAL);
});
it('parses double quoted string with escaped double quotes', function() {
var token = scan('"my \\"quoted\\" string"');
expect(token.value).to.be('my "quoted" string');
expect(token.type).to.be(ol.expr.TokenType.STRING_LITERAL);
});
it('parses double quoted string with escaped backslash', function() {
var token = scan('"my \\\ string"');
expect(token.value).to.be('my \ string');
expect(token.type).to.be(ol.expr.TokenType.STRING_LITERAL);
});
it('parses double quoted string with unicode escape sequences', function() {
var token = scan('"\u006f\u006c\u0033"');
expect(token.value).to.be('ol3');
expect(token.type).to.be(ol.expr.TokenType.STRING_LITERAL);
});
it('parses double quoted string with hex escape sequences', function() {
var token = scan('"\x6f\x6c\x33"');
expect(token.value).to.be('ol3');
expect(token.type).to.be(ol.expr.TokenType.STRING_LITERAL);
});
it('parses double quoted string with tab', function() {
var token = scan('"a\ttab"');
expect(token.value).to.be('a\ttab');
expect(token.type).to.be(ol.expr.TokenType.STRING_LITERAL);
});
it('throws on unterminated double quote', function() {
expect(function() {
scan('"never \'ending\' string');
}).to.throwException(function(err) {
expect(err).to.be.an(ol.expr.UnexpectedToken);
var token = err.token;
expect(token.type).to.be(ol.expr.TokenType.EOF);
expect(token.index).to.be(22);
});
});
it('parses single quoted string', function() {
var token = scan('\'my string\'');
expect(token.value).to.be('my string');
expect(token.type).to.be(ol.expr.TokenType.STRING_LITERAL);
});
it('parses single quoted string with internal double quotes', function() {
var token = scan('\'my "quoted" string\'');
expect(token.value).to.be('my "quoted" string');
expect(token.type).to.be(ol.expr.TokenType.STRING_LITERAL);
});
it('parses single quoted string with escaped single quotes', function() {
var token = scan('\'my \\\'quoted\\\' string\'');
expect(token.value).to.be('my \'quoted\' string');
expect(token.type).to.be(ol.expr.TokenType.STRING_LITERAL);
});
it('parses single quoted string with escaped backslash', function() {
var token = scan('\'my \\\ string\'');
expect(token.value).to.be('my \ string');
expect(token.type).to.be(ol.expr.TokenType.STRING_LITERAL);
});
it('parses single quoted string with unicode escape sequences', function() {
var token = scan('\'\u006f\u006c\u0033\'');
expect(token.value).to.be('ol3');
expect(token.type).to.be(ol.expr.TokenType.STRING_LITERAL);
});
it('parses single quoted string with hex escape sequences', function() {
var token = scan('\'\x6f\x6c\x33\'');
expect(token.value).to.be('ol3');
expect(token.type).to.be(ol.expr.TokenType.STRING_LITERAL);
});
it('parses single quoted string with tab', function() {
var token = scan('\'a\ttab\'');
expect(token.value).to.be('a\ttab');
expect(token.type).to.be(ol.expr.TokenType.STRING_LITERAL);
});
it('throws on unterminated single quote', function() {
expect(function() {
scan('\'never "ending" string');
}).to.throwException(function(err) {
expect(err).to.be.an(ol.expr.UnexpectedToken);
var token = err.token;
expect(token.type).to.be(ol.expr.TokenType.EOF);
expect(token.index).to.be(22);
});
});
});
});
goog.require('ol.expr.Lexer');
goog.require('ol.expr.TokenType');
goog.require('ol.expr.UnexpectedToken');