1148 lines
43 KiB
TypeScript
1148 lines
43 KiB
TypeScript
/**
|
|
* @license
|
|
* Copyright Google Inc. All Rights Reserved.
|
|
*
|
|
* Use of this source code is governed by an MIT-style license that can be
|
|
* found in the LICENSE file at https://angular.io/license
|
|
*/
|
|
|
|
import {getHtmlTagDefinition} from '../../src/ml_parser/html_tags';
|
|
import {InterpolationConfig} from '../../src/ml_parser/interpolation_config';
|
|
import * as lex from '../../src/ml_parser/lexer';
|
|
import {ParseLocation, ParseSourceFile, ParseSourceSpan} from '../../src/parse_util';
|
|
|
|
{
|
|
describe('HtmlLexer', () => {
|
|
describe('line/column numbers', () => {
|
|
it('should work without newlines', () => {
|
|
expect(tokenizeAndHumanizeLineColumn('<t>a</t>')).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '0:0'],
|
|
[lex.TokenType.TAG_OPEN_END, '0:2'],
|
|
[lex.TokenType.TEXT, '0:3'],
|
|
[lex.TokenType.TAG_CLOSE, '0:4'],
|
|
[lex.TokenType.EOF, '0:8'],
|
|
]);
|
|
});
|
|
|
|
it('should work with one newline', () => {
|
|
expect(tokenizeAndHumanizeLineColumn('<t>\na</t>')).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '0:0'],
|
|
[lex.TokenType.TAG_OPEN_END, '0:2'],
|
|
[lex.TokenType.TEXT, '0:3'],
|
|
[lex.TokenType.TAG_CLOSE, '1:1'],
|
|
[lex.TokenType.EOF, '1:5'],
|
|
]);
|
|
});
|
|
|
|
it('should work with multiple newlines', () => {
|
|
expect(tokenizeAndHumanizeLineColumn('<t\n>\na</t>')).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '0:0'],
|
|
[lex.TokenType.TAG_OPEN_END, '1:0'],
|
|
[lex.TokenType.TEXT, '1:1'],
|
|
[lex.TokenType.TAG_CLOSE, '2:1'],
|
|
[lex.TokenType.EOF, '2:5'],
|
|
]);
|
|
});
|
|
|
|
it('should work with CR and LF', () => {
|
|
expect(tokenizeAndHumanizeLineColumn('<t\n>\r\na\r</t>')).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '0:0'],
|
|
[lex.TokenType.TAG_OPEN_END, '1:0'],
|
|
[lex.TokenType.TEXT, '1:1'],
|
|
[lex.TokenType.TAG_CLOSE, '2:1'],
|
|
[lex.TokenType.EOF, '2:5'],
|
|
]);
|
|
});
|
|
});
|
|
|
|
describe('content ranges', () => {
|
|
it('should only process the text within the range', () => {
|
|
expect(tokenizeAndHumanizeSourceSpans(
|
|
'pre 1\npre 2\npre 3 `line 1\nline 2\nline 3` post 1\n post 2\n post 3',
|
|
{range: {startPos: 19, startLine: 2, startCol: 7, endPos: 39}}))
|
|
.toEqual([
|
|
[lex.TokenType.TEXT, 'line 1\nline 2\nline 3'],
|
|
[lex.TokenType.EOF, ''],
|
|
]);
|
|
});
|
|
|
|
it('should take into account preceding (non-processed) lines and columns', () => {
|
|
expect(tokenizeAndHumanizeLineColumn(
|
|
'pre 1\npre 2\npre 3 `line 1\nline 2\nline 3` post 1\n post 2\n post 3',
|
|
{range: {startPos: 19, startLine: 2, startCol: 7, endPos: 39}}))
|
|
.toEqual([
|
|
[lex.TokenType.TEXT, '2:7'],
|
|
[lex.TokenType.EOF, '4:6'],
|
|
]);
|
|
});
|
|
});
|
|
|
|
describe('comments', () => {
|
|
it('should parse comments', () => {
|
|
expect(tokenizeAndHumanizeParts('<!--t\ne\rs\r\nt-->')).toEqual([
|
|
[lex.TokenType.COMMENT_START],
|
|
[lex.TokenType.RAW_TEXT, 't\ne\ns\nt'],
|
|
[lex.TokenType.COMMENT_END],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should store the locations', () => {
|
|
expect(tokenizeAndHumanizeSourceSpans('<!--t\ne\rs\r\nt-->')).toEqual([
|
|
[lex.TokenType.COMMENT_START, '<!--'],
|
|
[lex.TokenType.RAW_TEXT, 't\ne\rs\r\nt'],
|
|
[lex.TokenType.COMMENT_END, '-->'],
|
|
[lex.TokenType.EOF, ''],
|
|
]);
|
|
});
|
|
|
|
it('should report <!- without -', () => {
|
|
expect(tokenizeAndHumanizeErrors('<!-a')).toEqual([
|
|
[lex.TokenType.COMMENT_START, 'Unexpected character "a"', '0:3']
|
|
]);
|
|
});
|
|
|
|
it('should report missing end comment', () => {
|
|
expect(tokenizeAndHumanizeErrors('<!--')).toEqual([
|
|
[lex.TokenType.RAW_TEXT, 'Unexpected character "EOF"', '0:4']
|
|
]);
|
|
});
|
|
|
|
it('should accept comments finishing by too many dashes (even number)', () => {
|
|
expect(tokenizeAndHumanizeSourceSpans('<!-- test ---->')).toEqual([
|
|
[lex.TokenType.COMMENT_START, '<!--'],
|
|
[lex.TokenType.RAW_TEXT, ' test --'],
|
|
[lex.TokenType.COMMENT_END, '-->'],
|
|
[lex.TokenType.EOF, ''],
|
|
]);
|
|
});
|
|
|
|
it('should accept comments finishing by too many dashes (odd number)', () => {
|
|
expect(tokenizeAndHumanizeSourceSpans('<!-- test --->')).toEqual([
|
|
[lex.TokenType.COMMENT_START, '<!--'],
|
|
[lex.TokenType.RAW_TEXT, ' test -'],
|
|
[lex.TokenType.COMMENT_END, '-->'],
|
|
[lex.TokenType.EOF, ''],
|
|
]);
|
|
});
|
|
});
|
|
|
|
describe('doctype', () => {
|
|
it('should parse doctypes', () => {
|
|
expect(tokenizeAndHumanizeParts('<!doctype html>')).toEqual([
|
|
[lex.TokenType.DOC_TYPE, 'doctype html'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should store the locations', () => {
|
|
expect(tokenizeAndHumanizeSourceSpans('<!doctype html>')).toEqual([
|
|
[lex.TokenType.DOC_TYPE, '<!doctype html>'],
|
|
[lex.TokenType.EOF, ''],
|
|
]);
|
|
});
|
|
|
|
it('should report missing end doctype', () => {
|
|
expect(tokenizeAndHumanizeErrors('<!')).toEqual([
|
|
[lex.TokenType.DOC_TYPE, 'Unexpected character "EOF"', '0:2']
|
|
]);
|
|
});
|
|
});
|
|
|
|
describe('CDATA', () => {
|
|
it('should parse CDATA', () => {
|
|
expect(tokenizeAndHumanizeParts('<![CDATA[t\ne\rs\r\nt]]>')).toEqual([
|
|
[lex.TokenType.CDATA_START],
|
|
[lex.TokenType.RAW_TEXT, 't\ne\ns\nt'],
|
|
[lex.TokenType.CDATA_END],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should store the locations', () => {
|
|
expect(tokenizeAndHumanizeSourceSpans('<![CDATA[t\ne\rs\r\nt]]>')).toEqual([
|
|
[lex.TokenType.CDATA_START, '<![CDATA['],
|
|
[lex.TokenType.RAW_TEXT, 't\ne\rs\r\nt'],
|
|
[lex.TokenType.CDATA_END, ']]>'],
|
|
[lex.TokenType.EOF, ''],
|
|
]);
|
|
});
|
|
|
|
it('should report <![ without CDATA[', () => {
|
|
expect(tokenizeAndHumanizeErrors('<![a')).toEqual([
|
|
[lex.TokenType.CDATA_START, 'Unexpected character "a"', '0:3']
|
|
]);
|
|
});
|
|
|
|
it('should report missing end cdata', () => {
|
|
expect(tokenizeAndHumanizeErrors('<![CDATA[')).toEqual([
|
|
[lex.TokenType.RAW_TEXT, 'Unexpected character "EOF"', '0:9']
|
|
]);
|
|
});
|
|
});
|
|
|
|
describe('open tags', () => {
|
|
it('should parse open tags without prefix', () => {
|
|
expect(tokenizeAndHumanizeParts('<test>')).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 'test'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse namespace prefix', () => {
|
|
expect(tokenizeAndHumanizeParts('<ns1:test>')).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, 'ns1', 'test'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse void tags', () => {
|
|
expect(tokenizeAndHumanizeParts('<test/>')).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 'test'],
|
|
[lex.TokenType.TAG_OPEN_END_VOID],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should allow whitespace after the tag name', () => {
|
|
expect(tokenizeAndHumanizeParts('<test >')).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 'test'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should store the locations', () => {
|
|
expect(tokenizeAndHumanizeSourceSpans('<test>')).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '<test'],
|
|
[lex.TokenType.TAG_OPEN_END, '>'],
|
|
[lex.TokenType.EOF, ''],
|
|
]);
|
|
});
|
|
|
|
});
|
|
|
|
describe('attributes', () => {
|
|
it('should parse attributes without prefix', () => {
|
|
expect(tokenizeAndHumanizeParts('<t a>')).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 't'],
|
|
[lex.TokenType.ATTR_NAME, '', 'a'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse attributes with interpolation', () => {
|
|
expect(tokenizeAndHumanizeParts('<t a="{{v}}" b="s{{m}}e" c="s{{m//c}}e">')).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 't'],
|
|
[lex.TokenType.ATTR_NAME, '', 'a'],
|
|
[lex.TokenType.ATTR_QUOTE, '"'],
|
|
[lex.TokenType.ATTR_VALUE, '{{v}}'],
|
|
[lex.TokenType.ATTR_QUOTE, '"'],
|
|
[lex.TokenType.ATTR_NAME, '', 'b'],
|
|
[lex.TokenType.ATTR_QUOTE, '"'],
|
|
[lex.TokenType.ATTR_VALUE, 's{{m}}e'],
|
|
[lex.TokenType.ATTR_QUOTE, '"'],
|
|
[lex.TokenType.ATTR_NAME, '', 'c'],
|
|
[lex.TokenType.ATTR_QUOTE, '"'],
|
|
[lex.TokenType.ATTR_VALUE, 's{{m//c}}e'],
|
|
[lex.TokenType.ATTR_QUOTE, '"'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse attributes with prefix', () => {
|
|
expect(tokenizeAndHumanizeParts('<t ns1:a>')).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 't'],
|
|
[lex.TokenType.ATTR_NAME, 'ns1', 'a'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse attributes whose prefix is not valid', () => {
|
|
expect(tokenizeAndHumanizeParts('<t (ns1:a)>')).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 't'],
|
|
[lex.TokenType.ATTR_NAME, '', '(ns1:a)'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse attributes with single quote value', () => {
|
|
expect(tokenizeAndHumanizeParts('<t a=\'b\'>')).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 't'],
|
|
[lex.TokenType.ATTR_NAME, '', 'a'],
|
|
[lex.TokenType.ATTR_QUOTE, '\''],
|
|
[lex.TokenType.ATTR_VALUE, 'b'],
|
|
[lex.TokenType.ATTR_QUOTE, '\''],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse attributes with double quote value', () => {
|
|
expect(tokenizeAndHumanizeParts('<t a="b">')).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 't'],
|
|
[lex.TokenType.ATTR_NAME, '', 'a'],
|
|
[lex.TokenType.ATTR_QUOTE, '"'],
|
|
[lex.TokenType.ATTR_VALUE, 'b'],
|
|
[lex.TokenType.ATTR_QUOTE, '"'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse attributes with unquoted value', () => {
|
|
expect(tokenizeAndHumanizeParts('<t a=b>')).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 't'],
|
|
[lex.TokenType.ATTR_NAME, '', 'a'],
|
|
[lex.TokenType.ATTR_VALUE, 'b'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should allow whitespace', () => {
|
|
expect(tokenizeAndHumanizeParts('<t a = b >')).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 't'],
|
|
[lex.TokenType.ATTR_NAME, '', 'a'],
|
|
[lex.TokenType.ATTR_VALUE, 'b'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse attributes with entities in values', () => {
|
|
expect(tokenizeAndHumanizeParts('<t a="AA">')).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 't'],
|
|
[lex.TokenType.ATTR_NAME, '', 'a'],
|
|
[lex.TokenType.ATTR_QUOTE, '"'],
|
|
[lex.TokenType.ATTR_VALUE, 'AA'],
|
|
[lex.TokenType.ATTR_QUOTE, '"'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should not decode entities without trailing ";"', () => {
|
|
expect(tokenizeAndHumanizeParts('<t a="&" b="c&&d">')).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 't'],
|
|
[lex.TokenType.ATTR_NAME, '', 'a'],
|
|
[lex.TokenType.ATTR_QUOTE, '"'],
|
|
[lex.TokenType.ATTR_VALUE, '&'],
|
|
[lex.TokenType.ATTR_QUOTE, '"'],
|
|
[lex.TokenType.ATTR_NAME, '', 'b'],
|
|
[lex.TokenType.ATTR_QUOTE, '"'],
|
|
[lex.TokenType.ATTR_VALUE, 'c&&d'],
|
|
[lex.TokenType.ATTR_QUOTE, '"'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse attributes with "&" in values', () => {
|
|
expect(tokenizeAndHumanizeParts('<t a="b && c &">')).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 't'],
|
|
[lex.TokenType.ATTR_NAME, '', 'a'],
|
|
[lex.TokenType.ATTR_QUOTE, '"'],
|
|
[lex.TokenType.ATTR_VALUE, 'b && c &'],
|
|
[lex.TokenType.ATTR_QUOTE, '"'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse values with CR and LF', () => {
|
|
expect(tokenizeAndHumanizeParts('<t a=\'t\ne\rs\r\nt\'>')).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 't'],
|
|
[lex.TokenType.ATTR_NAME, '', 'a'],
|
|
[lex.TokenType.ATTR_QUOTE, '\''],
|
|
[lex.TokenType.ATTR_VALUE, 't\ne\ns\nt'],
|
|
[lex.TokenType.ATTR_QUOTE, '\''],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should store the locations', () => {
|
|
expect(tokenizeAndHumanizeSourceSpans('<t a=b>')).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '<t'],
|
|
[lex.TokenType.ATTR_NAME, 'a'],
|
|
[lex.TokenType.ATTR_VALUE, 'b'],
|
|
[lex.TokenType.TAG_OPEN_END, '>'],
|
|
[lex.TokenType.EOF, ''],
|
|
]);
|
|
});
|
|
|
|
});
|
|
|
|
describe('closing tags', () => {
|
|
it('should parse closing tags without prefix', () => {
|
|
expect(tokenizeAndHumanizeParts('</test>')).toEqual([
|
|
[lex.TokenType.TAG_CLOSE, '', 'test'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse closing tags with prefix', () => {
|
|
expect(tokenizeAndHumanizeParts('</ns1:test>')).toEqual([
|
|
[lex.TokenType.TAG_CLOSE, 'ns1', 'test'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should allow whitespace', () => {
|
|
expect(tokenizeAndHumanizeParts('</ test >')).toEqual([
|
|
[lex.TokenType.TAG_CLOSE, '', 'test'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should store the locations', () => {
|
|
expect(tokenizeAndHumanizeSourceSpans('</test>')).toEqual([
|
|
[lex.TokenType.TAG_CLOSE, '</test>'],
|
|
[lex.TokenType.EOF, ''],
|
|
]);
|
|
});
|
|
|
|
it('should report missing name after </', () => {
|
|
expect(tokenizeAndHumanizeErrors('</')).toEqual([
|
|
[lex.TokenType.TAG_CLOSE, 'Unexpected character "EOF"', '0:2']
|
|
]);
|
|
});
|
|
|
|
it('should report missing >', () => {
|
|
expect(tokenizeAndHumanizeErrors('</test')).toEqual([
|
|
[lex.TokenType.TAG_CLOSE, 'Unexpected character "EOF"', '0:6']
|
|
]);
|
|
});
|
|
});
|
|
|
|
describe('entities', () => {
|
|
it('should parse named entities', () => {
|
|
expect(tokenizeAndHumanizeParts('a&b')).toEqual([
|
|
[lex.TokenType.TEXT, 'a&b'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse hexadecimal entities', () => {
|
|
expect(tokenizeAndHumanizeParts('AA')).toEqual([
|
|
[lex.TokenType.TEXT, 'AA'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse decimal entities', () => {
|
|
expect(tokenizeAndHumanizeParts('A')).toEqual([
|
|
[lex.TokenType.TEXT, 'A'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should store the locations', () => {
|
|
expect(tokenizeAndHumanizeSourceSpans('a&b')).toEqual([
|
|
[lex.TokenType.TEXT, 'a&b'],
|
|
[lex.TokenType.EOF, ''],
|
|
]);
|
|
});
|
|
|
|
it('should report malformed/unknown entities', () => {
|
|
expect(tokenizeAndHumanizeErrors('&tbo;')).toEqual([[
|
|
lex.TokenType.TEXT,
|
|
'Unknown entity "tbo" - use the "&#<decimal>;" or "&#x<hex>;" syntax', '0:0'
|
|
]]);
|
|
expect(tokenizeAndHumanizeErrors('&#asdf;')).toEqual([
|
|
[lex.TokenType.TEXT, 'Unexpected character "s"', '0:3']
|
|
]);
|
|
expect(tokenizeAndHumanizeErrors('
sdf;')).toEqual([
|
|
[lex.TokenType.TEXT, 'Unexpected character "s"', '0:4']
|
|
]);
|
|
|
|
expect(tokenizeAndHumanizeErrors('઼')).toEqual([
|
|
[lex.TokenType.TEXT, 'Unexpected character "EOF"', '0:6']
|
|
]);
|
|
});
|
|
});
|
|
|
|
describe('regular text', () => {
|
|
it('should parse text', () => {
|
|
expect(tokenizeAndHumanizeParts('a')).toEqual([
|
|
[lex.TokenType.TEXT, 'a'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse interpolation', () => {
|
|
expect(tokenizeAndHumanizeParts('{{ a }}b{{ c // comment }}')).toEqual([
|
|
[lex.TokenType.TEXT, '{{ a }}b{{ c // comment }}'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse interpolation with custom markers', () => {
|
|
expect(tokenizeAndHumanizeParts('{% a %}', {interpolationConfig: {start: '{%', end: '%}'}}))
|
|
.toEqual([
|
|
[lex.TokenType.TEXT, '{% a %}'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should handle CR & LF', () => {
|
|
expect(tokenizeAndHumanizeParts('t\ne\rs\r\nt')).toEqual([
|
|
[lex.TokenType.TEXT, 't\ne\ns\nt'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse entities', () => {
|
|
expect(tokenizeAndHumanizeParts('a&b')).toEqual([
|
|
[lex.TokenType.TEXT, 'a&b'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse text starting with "&"', () => {
|
|
expect(tokenizeAndHumanizeParts('a && b &')).toEqual([
|
|
[lex.TokenType.TEXT, 'a && b &'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should store the locations', () => {
|
|
expect(tokenizeAndHumanizeSourceSpans('a')).toEqual([
|
|
[lex.TokenType.TEXT, 'a'],
|
|
[lex.TokenType.EOF, ''],
|
|
]);
|
|
});
|
|
|
|
it('should allow "<" in text nodes', () => {
|
|
expect(tokenizeAndHumanizeParts('{{ a < b ? c : d }}')).toEqual([
|
|
[lex.TokenType.TEXT, '{{ a < b ? c : d }}'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
|
|
expect(tokenizeAndHumanizeSourceSpans('<p>a<b</p>')).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '<p'],
|
|
[lex.TokenType.TAG_OPEN_END, '>'],
|
|
[lex.TokenType.TEXT, 'a<b'],
|
|
[lex.TokenType.TAG_CLOSE, '</p>'],
|
|
[lex.TokenType.EOF, ''],
|
|
]);
|
|
|
|
expect(tokenizeAndHumanizeParts('< a>')).toEqual([
|
|
[lex.TokenType.TEXT, '< a>'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse valid start tag in interpolation', () => {
|
|
expect(tokenizeAndHumanizeParts('{{ a <b && c > d }}')).toEqual([
|
|
[lex.TokenType.TEXT, '{{ a '],
|
|
[lex.TokenType.TAG_OPEN_START, '', 'b'],
|
|
[lex.TokenType.ATTR_NAME, '', '&&'],
|
|
[lex.TokenType.ATTR_NAME, '', 'c'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.TEXT, ' d }}'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should be able to escape {', () => {
|
|
expect(tokenizeAndHumanizeParts('{{ "{" }}')).toEqual([
|
|
[lex.TokenType.TEXT, '{{ "{" }}'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should be able to escape {{', () => {
|
|
expect(tokenizeAndHumanizeParts('{{ "{{" }}')).toEqual([
|
|
[lex.TokenType.TEXT, '{{ "{{" }}'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should treat expansion form as text when they are not parsed', () => {
|
|
expect(tokenizeAndHumanizeParts(
|
|
'<span>{a, b, =4 {c}}</span>', {tokenizeExpansionForms: false}))
|
|
.toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 'span'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.TEXT, '{a, b, =4 {c}}'],
|
|
[lex.TokenType.TAG_CLOSE, '', 'span'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
});
|
|
|
|
describe('raw text', () => {
|
|
it('should parse text', () => {
|
|
expect(tokenizeAndHumanizeParts(`<script>t\ne\rs\r\nt</script>`)).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 'script'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.RAW_TEXT, 't\ne\ns\nt'],
|
|
[lex.TokenType.TAG_CLOSE, '', 'script'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should not detect entities', () => {
|
|
expect(tokenizeAndHumanizeParts(`<script>&</SCRIPT>`)).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 'script'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.RAW_TEXT, '&'],
|
|
[lex.TokenType.TAG_CLOSE, '', 'script'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should ignore other opening tags', () => {
|
|
expect(tokenizeAndHumanizeParts(`<script>a<div></script>`)).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 'script'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.RAW_TEXT, 'a<div>'],
|
|
[lex.TokenType.TAG_CLOSE, '', 'script'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should ignore other closing tags', () => {
|
|
expect(tokenizeAndHumanizeParts(`<script>a</test></script>`)).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 'script'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.RAW_TEXT, 'a</test>'],
|
|
[lex.TokenType.TAG_CLOSE, '', 'script'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should store the locations', () => {
|
|
expect(tokenizeAndHumanizeSourceSpans(`<script>a</script>`)).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '<script'],
|
|
[lex.TokenType.TAG_OPEN_END, '>'],
|
|
[lex.TokenType.RAW_TEXT, 'a'],
|
|
[lex.TokenType.TAG_CLOSE, '</script>'],
|
|
[lex.TokenType.EOF, ''],
|
|
]);
|
|
});
|
|
});
|
|
|
|
describe('escapable raw text', () => {
|
|
it('should parse text', () => {
|
|
expect(tokenizeAndHumanizeParts(`<title>t\ne\rs\r\nt</title>`)).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 'title'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.ESCAPABLE_RAW_TEXT, 't\ne\ns\nt'],
|
|
[lex.TokenType.TAG_CLOSE, '', 'title'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should detect entities', () => {
|
|
expect(tokenizeAndHumanizeParts(`<title>&</title>`)).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 'title'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.ESCAPABLE_RAW_TEXT, '&'],
|
|
[lex.TokenType.TAG_CLOSE, '', 'title'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should ignore other opening tags', () => {
|
|
expect(tokenizeAndHumanizeParts(`<title>a<div></title>`)).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 'title'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.ESCAPABLE_RAW_TEXT, 'a<div>'],
|
|
[lex.TokenType.TAG_CLOSE, '', 'title'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should ignore other closing tags', () => {
|
|
expect(tokenizeAndHumanizeParts(`<title>a</test></title>`)).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 'title'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.ESCAPABLE_RAW_TEXT, 'a</test>'],
|
|
[lex.TokenType.TAG_CLOSE, '', 'title'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should store the locations', () => {
|
|
expect(tokenizeAndHumanizeSourceSpans(`<title>a</title>`)).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '<title'],
|
|
[lex.TokenType.TAG_OPEN_END, '>'],
|
|
[lex.TokenType.ESCAPABLE_RAW_TEXT, 'a'],
|
|
[lex.TokenType.TAG_CLOSE, '</title>'],
|
|
[lex.TokenType.EOF, ''],
|
|
]);
|
|
});
|
|
|
|
});
|
|
|
|
describe('expansion forms', () => {
|
|
it('should parse an expansion form', () => {
|
|
expect(
|
|
tokenizeAndHumanizeParts(
|
|
'{one.two, three, =4 {four} =5 {five} foo {bar} }', {tokenizeExpansionForms: true}))
|
|
.toEqual([
|
|
[lex.TokenType.EXPANSION_FORM_START],
|
|
[lex.TokenType.RAW_TEXT, 'one.two'],
|
|
[lex.TokenType.RAW_TEXT, 'three'],
|
|
[lex.TokenType.EXPANSION_CASE_VALUE, '=4'],
|
|
[lex.TokenType.EXPANSION_CASE_EXP_START],
|
|
[lex.TokenType.TEXT, 'four'],
|
|
[lex.TokenType.EXPANSION_CASE_EXP_END],
|
|
[lex.TokenType.EXPANSION_CASE_VALUE, '=5'],
|
|
[lex.TokenType.EXPANSION_CASE_EXP_START],
|
|
[lex.TokenType.TEXT, 'five'],
|
|
[lex.TokenType.EXPANSION_CASE_EXP_END],
|
|
[lex.TokenType.EXPANSION_CASE_VALUE, 'foo'],
|
|
[lex.TokenType.EXPANSION_CASE_EXP_START],
|
|
[lex.TokenType.TEXT, 'bar'],
|
|
[lex.TokenType.EXPANSION_CASE_EXP_END],
|
|
[lex.TokenType.EXPANSION_FORM_END],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse an expansion form with text elements surrounding it', () => {
|
|
expect(tokenizeAndHumanizeParts(
|
|
'before{one.two, three, =4 {four}}after', {tokenizeExpansionForms: true}))
|
|
.toEqual([
|
|
[lex.TokenType.TEXT, 'before'],
|
|
[lex.TokenType.EXPANSION_FORM_START],
|
|
[lex.TokenType.RAW_TEXT, 'one.two'],
|
|
[lex.TokenType.RAW_TEXT, 'three'],
|
|
[lex.TokenType.EXPANSION_CASE_VALUE, '=4'],
|
|
[lex.TokenType.EXPANSION_CASE_EXP_START],
|
|
[lex.TokenType.TEXT, 'four'],
|
|
[lex.TokenType.EXPANSION_CASE_EXP_END],
|
|
[lex.TokenType.EXPANSION_FORM_END],
|
|
[lex.TokenType.TEXT, 'after'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse an expansion form as a tag single child', () => {
|
|
expect(tokenizeAndHumanizeParts(
|
|
'<div><span>{a, b, =4 {c}}</span></div>', {tokenizeExpansionForms: true}))
|
|
.toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 'div'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.TAG_OPEN_START, '', 'span'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.EXPANSION_FORM_START],
|
|
[lex.TokenType.RAW_TEXT, 'a'],
|
|
[lex.TokenType.RAW_TEXT, 'b'],
|
|
[lex.TokenType.EXPANSION_CASE_VALUE, '=4'],
|
|
[lex.TokenType.EXPANSION_CASE_EXP_START],
|
|
[lex.TokenType.TEXT, 'c'],
|
|
[lex.TokenType.EXPANSION_CASE_EXP_END],
|
|
[lex.TokenType.EXPANSION_FORM_END],
|
|
[lex.TokenType.TAG_CLOSE, '', 'span'],
|
|
[lex.TokenType.TAG_CLOSE, '', 'div'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse an expansion forms with elements in it', () => {
|
|
expect(tokenizeAndHumanizeParts(
|
|
'{one.two, three, =4 {four <b>a</b>}}', {tokenizeExpansionForms: true}))
|
|
.toEqual([
|
|
[lex.TokenType.EXPANSION_FORM_START],
|
|
[lex.TokenType.RAW_TEXT, 'one.two'],
|
|
[lex.TokenType.RAW_TEXT, 'three'],
|
|
[lex.TokenType.EXPANSION_CASE_VALUE, '=4'],
|
|
[lex.TokenType.EXPANSION_CASE_EXP_START],
|
|
[lex.TokenType.TEXT, 'four '],
|
|
[lex.TokenType.TAG_OPEN_START, '', 'b'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.TEXT, 'a'],
|
|
[lex.TokenType.TAG_CLOSE, '', 'b'],
|
|
[lex.TokenType.EXPANSION_CASE_EXP_END],
|
|
[lex.TokenType.EXPANSION_FORM_END],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse an expansion forms containing an interpolation', () => {
|
|
expect(tokenizeAndHumanizeParts(
|
|
'{one.two, three, =4 {four {{a}}}}', {tokenizeExpansionForms: true}))
|
|
.toEqual([
|
|
[lex.TokenType.EXPANSION_FORM_START],
|
|
[lex.TokenType.RAW_TEXT, 'one.two'],
|
|
[lex.TokenType.RAW_TEXT, 'three'],
|
|
[lex.TokenType.EXPANSION_CASE_VALUE, '=4'],
|
|
[lex.TokenType.EXPANSION_CASE_EXP_START],
|
|
[lex.TokenType.TEXT, 'four {{a}}'],
|
|
[lex.TokenType.EXPANSION_CASE_EXP_END],
|
|
[lex.TokenType.EXPANSION_FORM_END],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse nested expansion forms', () => {
|
|
expect(tokenizeAndHumanizeParts(
|
|
`{one.two, three, =4 { {xx, yy, =x {one}} }}`, {tokenizeExpansionForms: true}))
|
|
.toEqual([
|
|
[lex.TokenType.EXPANSION_FORM_START],
|
|
[lex.TokenType.RAW_TEXT, 'one.two'],
|
|
[lex.TokenType.RAW_TEXT, 'three'],
|
|
[lex.TokenType.EXPANSION_CASE_VALUE, '=4'],
|
|
[lex.TokenType.EXPANSION_CASE_EXP_START],
|
|
[lex.TokenType.EXPANSION_FORM_START],
|
|
[lex.TokenType.RAW_TEXT, 'xx'],
|
|
[lex.TokenType.RAW_TEXT, 'yy'],
|
|
[lex.TokenType.EXPANSION_CASE_VALUE, '=x'],
|
|
[lex.TokenType.EXPANSION_CASE_EXP_START],
|
|
[lex.TokenType.TEXT, 'one'],
|
|
[lex.TokenType.EXPANSION_CASE_EXP_END],
|
|
[lex.TokenType.EXPANSION_FORM_END],
|
|
[lex.TokenType.TEXT, ' '],
|
|
[lex.TokenType.EXPANSION_CASE_EXP_END],
|
|
[lex.TokenType.EXPANSION_FORM_END],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
});
|
|
|
|
describe('errors', () => {
|
|
it('should report unescaped "{" on error', () => {
|
|
expect(tokenizeAndHumanizeErrors(`<p>before { after</p>`, {tokenizeExpansionForms: true}))
|
|
.toEqual([[
|
|
lex.TokenType.RAW_TEXT,
|
|
`Unexpected character "EOF" (Do you have an unescaped "{" in your template? Use "{{ '{' }}") to escape it.)`,
|
|
'0:21',
|
|
]]);
|
|
});
|
|
|
|
it('should include 2 lines of context in message', () => {
|
|
const src = '111\n222\n333\nE\n444\n555\n666\n';
|
|
const file = new ParseSourceFile(src, 'file://');
|
|
const location = new ParseLocation(file, 12, 123, 456);
|
|
const span = new ParseSourceSpan(location, location);
|
|
const error = new lex.TokenError('**ERROR**', null !, span);
|
|
expect(error.toString())
|
|
.toEqual(`**ERROR** ("\n222\n333\n[ERROR ->]E\n444\n555\n"): file://@123:456`);
|
|
});
|
|
});
|
|
|
|
describe('unicode characters', () => {
|
|
it('should support unicode characters', () => {
|
|
expect(tokenizeAndHumanizeSourceSpans(`<p>İ</p>`)).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '<p'],
|
|
[lex.TokenType.TAG_OPEN_END, '>'],
|
|
[lex.TokenType.TEXT, 'İ'],
|
|
[lex.TokenType.TAG_CLOSE, '</p>'],
|
|
[lex.TokenType.EOF, ''],
|
|
]);
|
|
});
|
|
});
|
|
|
|
describe('(processing escaped strings)', () => {
|
|
it('should unescape standard escape sequences', () => {
|
|
expect(tokenizeAndHumanizeParts('\\\' \\\' \\\'', {escapedString: true})).toEqual([
|
|
[lex.TokenType.TEXT, '\' \' \''],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
expect(tokenizeAndHumanizeParts('\\" \\" \\"', {escapedString: true})).toEqual([
|
|
[lex.TokenType.TEXT, '\" \" \"'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
expect(tokenizeAndHumanizeParts('\\` \\` \\`', {escapedString: true})).toEqual([
|
|
[lex.TokenType.TEXT, '\` \` \`'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
expect(tokenizeAndHumanizeParts('\\\\ \\\\ \\\\', {escapedString: true})).toEqual([
|
|
[lex.TokenType.TEXT, '\\ \\ \\'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
expect(tokenizeAndHumanizeParts('\\n \\n \\n', {escapedString: true})).toEqual([
|
|
[lex.TokenType.TEXT, '\n \n \n'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
expect(tokenizeAndHumanizeParts('\\r \\r \\r', {escapedString: true})).toEqual([
|
|
[lex.TokenType.TEXT, '\n \n \n'], // post processing converts `\r` to `\n`
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
expect(tokenizeAndHumanizeParts('\\v \\v \\v', {escapedString: true})).toEqual([
|
|
[lex.TokenType.TEXT, '\v \v \v'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
expect(tokenizeAndHumanizeParts('\\t \\t \\t', {escapedString: true})).toEqual([
|
|
[lex.TokenType.TEXT, '\t \t \t'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
expect(tokenizeAndHumanizeParts('\\b \\b \\b', {escapedString: true})).toEqual([
|
|
[lex.TokenType.TEXT, '\b \b \b'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
expect(tokenizeAndHumanizeParts('\\f \\f \\f', {escapedString: true})).toEqual([
|
|
[lex.TokenType.TEXT, '\f \f \f'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
expect(tokenizeAndHumanizeParts(
|
|
'\\\' \\" \\` \\\\ \\n \\r \\v \\t \\b \\f', {escapedString: true}))
|
|
.toEqual([
|
|
[lex.TokenType.TEXT, '\' \" \` \\ \n \n \v \t \b \f'],
|
|
[lex.TokenType.EOF],
|
|
|
|
]);
|
|
});
|
|
|
|
it('should unescape null sequences', () => {
|
|
expect(tokenizeAndHumanizeParts('\\0', {escapedString: true})).toEqual([
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
// \09 is not an octal number so the \0 is taken as EOF
|
|
expect(tokenizeAndHumanizeParts('\\09', {escapedString: true})).toEqual([
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should unescape octal sequences', () => {
|
|
// \19 is read as an octal `\1` followed by a normal char `9`
|
|
// \1234 is read as an octal `\123` followed by a normal char `4`
|
|
// \999 is not an octal number so its backslash just gets removed.
|
|
expect(tokenizeAndHumanizeParts(
|
|
'\\001 \\01 \\1 \\12 \\223 \\19 \\2234 \\999', {escapedString: true}))
|
|
.toEqual([
|
|
[lex.TokenType.TEXT, '\x01 \x01 \x01 \x0A \x93 \x019 \x934 999'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should unescape hex sequences', () => {
|
|
expect(tokenizeAndHumanizeParts('\\x12 \\x4F \\xDC', {escapedString: true})).toEqual([
|
|
[lex.TokenType.TEXT, '\x12 \x4F \xDC'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should report an error on an invalid hex sequence', () => {
|
|
expect(tokenizeAndHumanizeErrors('\\xGG', {escapedString: true})).toEqual([
|
|
[null, 'Invalid hexadecimal escape sequence', '0:2']
|
|
]);
|
|
|
|
expect(tokenizeAndHumanizeErrors('abc \\x xyz', {escapedString: true})).toEqual([
|
|
[lex.TokenType.TEXT, 'Invalid hexadecimal escape sequence', '0:6']
|
|
]);
|
|
|
|
expect(tokenizeAndHumanizeErrors('abc\\x', {escapedString: true})).toEqual([
|
|
[lex.TokenType.TEXT, 'Unexpected character "EOF"', '0:5']
|
|
]);
|
|
});
|
|
|
|
it('should unescape fixed length Unicode sequences', () => {
|
|
expect(tokenizeAndHumanizeParts('\\u0123 \\uABCD', {escapedString: true})).toEqual([
|
|
[lex.TokenType.TEXT, '\u0123 \uABCD'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should error on an invalid fixed length Unicode sequence', () => {
|
|
expect(tokenizeAndHumanizeErrors('\\uGGGG', {escapedString: true})).toEqual([
|
|
[null, 'Invalid hexadecimal escape sequence', '0:2']
|
|
]);
|
|
});
|
|
|
|
it('should unescape variable length Unicode sequences', () => {
|
|
expect(tokenizeAndHumanizeParts(
|
|
'\\u{01} \\u{ABC} \\u{1234} \\u{123AB}', {escapedString: true}))
|
|
.toEqual([
|
|
[lex.TokenType.TEXT, '\u{01} \u{ABC} \u{1234} \u{123AB}'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should error on an invalid variable length Unicode sequence', () => {
|
|
expect(tokenizeAndHumanizeErrors('\\u{GG}', {escapedString: true})).toEqual([
|
|
[null, 'Invalid hexadecimal escape sequence', '0:3']
|
|
]);
|
|
});
|
|
|
|
it('should unescape line continuations', () => {
|
|
expect(tokenizeAndHumanizeParts('abc\\\ndef', {escapedString: true})).toEqual([
|
|
[lex.TokenType.TEXT, 'abcdef'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
expect(tokenizeAndHumanizeParts('\\\nx\\\ny\\\n', {escapedString: true})).toEqual([
|
|
[lex.TokenType.TEXT, 'xy'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should remove backslash from "non-escape" sequences', () => {
|
|
expect(tokenizeAndHumanizeParts('\a \g \~', {escapedString: true})).toEqual([
|
|
[lex.TokenType.TEXT, 'a g ~'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should unescape sequences in plain text', () => {
|
|
expect(tokenizeAndHumanizeParts('abc\ndef\\nghi\\tjkl\\`\\\'\\"mno', {escapedString: true}))
|
|
.toEqual([
|
|
[lex.TokenType.TEXT, 'abc\ndef\nghi\tjkl`\'"mno'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should unescape sequences in raw text', () => {
|
|
expect(tokenizeAndHumanizeParts(
|
|
'<script>abc\ndef\\nghi\\tjkl\\`\\\'\\"mno</script>', {escapedString: true}))
|
|
.toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 'script'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.RAW_TEXT, 'abc\ndef\nghi\tjkl`\'"mno'],
|
|
[lex.TokenType.TAG_CLOSE, '', 'script'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should unescape sequences in escapable raw text', () => {
|
|
expect(tokenizeAndHumanizeParts(
|
|
'<title>abc\ndef\\nghi\\tjkl\\`\\\'\\"mno</title>', {escapedString: true}))
|
|
.toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 'title'],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.ESCAPABLE_RAW_TEXT, 'abc\ndef\nghi\tjkl`\'"mno'],
|
|
[lex.TokenType.TAG_CLOSE, '', 'title'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should parse over escape sequences in tag definitions', () => {
|
|
expect(tokenizeAndHumanizeParts('<t a=\\"b\\" \\n c=\\\'d\\\'>', {escapedString: true}))
|
|
.toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 't'],
|
|
[lex.TokenType.ATTR_NAME, '', 'a'],
|
|
[lex.TokenType.ATTR_QUOTE, '"'],
|
|
[lex.TokenType.ATTR_VALUE, 'b'],
|
|
[lex.TokenType.ATTR_QUOTE, '"'],
|
|
[lex.TokenType.ATTR_NAME, '', 'c'],
|
|
[lex.TokenType.ATTR_QUOTE, '\''],
|
|
[lex.TokenType.ATTR_VALUE, 'd'],
|
|
[lex.TokenType.ATTR_QUOTE, '\''],
|
|
[lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
});
|
|
|
|
it('should tokenize the correct span when there are escape sequences', () => {
|
|
const text =
|
|
'selector: "app-root",\ntemplate: "line 1\\n\\"line 2\\"\\nline 3",\ninputs: []';
|
|
const range = {
|
|
startPos: 33,
|
|
startLine: 1,
|
|
startCol: 10,
|
|
endPos: 59,
|
|
};
|
|
expect(tokenizeAndHumanizeParts(text, {range, escapedString: true})).toEqual([
|
|
[lex.TokenType.TEXT, 'line 1\n"line 2"\nline 3'],
|
|
[lex.TokenType.EOF],
|
|
]);
|
|
expect(tokenizeAndHumanizeSourceSpans(text, {range, escapedString: true})).toEqual([
|
|
[lex.TokenType.TEXT, 'line 1\\n\\"line 2\\"\\nline 3'],
|
|
[lex.TokenType.EOF, ''],
|
|
]);
|
|
});
|
|
|
|
it('should account for escape sequences when computing source spans ', () => {
|
|
const text = '<t>line 1</t>\n' + // <- unescaped line break
|
|
'<t>line 2</t>\\n' + // <- escaped line break
|
|
'<t>line 3\\\n' + // <- line continuation
|
|
'</t>';
|
|
|
|
expect(tokenizeAndHumanizeParts(text, {escapedString: true})).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '', 't'], [lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.TEXT, 'line 1'], [lex.TokenType.TAG_CLOSE, '', 't'],
|
|
[lex.TokenType.TEXT, '\n'],
|
|
|
|
[lex.TokenType.TAG_OPEN_START, '', 't'], [lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.TEXT, 'line 2'], [lex.TokenType.TAG_CLOSE, '', 't'],
|
|
[lex.TokenType.TEXT, '\n'],
|
|
|
|
[lex.TokenType.TAG_OPEN_START, '', 't'], [lex.TokenType.TAG_OPEN_END],
|
|
[lex.TokenType.TEXT, 'line 3'], // <- line continuation does not appear in token
|
|
[lex.TokenType.TAG_CLOSE, '', 't'],
|
|
|
|
[lex.TokenType.EOF]
|
|
]);
|
|
expect(tokenizeAndHumanizeLineColumn(text, {escapedString: true})).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '0:0'],
|
|
[lex.TokenType.TAG_OPEN_END, '0:2'],
|
|
[lex.TokenType.TEXT, '0:3'],
|
|
[lex.TokenType.TAG_CLOSE, '0:9'],
|
|
[lex.TokenType.TEXT, '0:13'], // <- real newline increments the row
|
|
|
|
[lex.TokenType.TAG_OPEN_START, '1:0'],
|
|
[lex.TokenType.TAG_OPEN_END, '1:2'],
|
|
[lex.TokenType.TEXT, '1:3'],
|
|
[lex.TokenType.TAG_CLOSE, '1:9'],
|
|
[lex.TokenType.TEXT, '1:14'], // <- escaped newline does not increment the row
|
|
|
|
[lex.TokenType.TAG_OPEN_START, '1:15'],
|
|
[lex.TokenType.TAG_OPEN_END, '1:17'],
|
|
[lex.TokenType.TEXT, '1:18'], // <- the line continuation increments the row
|
|
[lex.TokenType.TAG_CLOSE, '2:0'],
|
|
|
|
[lex.TokenType.EOF, '2:4'],
|
|
]);
|
|
expect(tokenizeAndHumanizeSourceSpans(text, {escapedString: true})).toEqual([
|
|
[lex.TokenType.TAG_OPEN_START, '<t'], [lex.TokenType.TAG_OPEN_END, '>'],
|
|
[lex.TokenType.TEXT, 'line 1'], [lex.TokenType.TAG_CLOSE, '</t>'],
|
|
[lex.TokenType.TEXT, '\n'],
|
|
|
|
[lex.TokenType.TAG_OPEN_START, '<t'], [lex.TokenType.TAG_OPEN_END, '>'],
|
|
[lex.TokenType.TEXT, 'line 2'], [lex.TokenType.TAG_CLOSE, '</t>\\'],
|
|
[lex.TokenType.TEXT, 'n'],
|
|
|
|
[lex.TokenType.TAG_OPEN_START, '<t'], [lex.TokenType.TAG_OPEN_END, '>'],
|
|
[lex.TokenType.TEXT, 'line 3\\\n'], [lex.TokenType.TAG_CLOSE, '</t>'],
|
|
|
|
[lex.TokenType.EOF, '']
|
|
]);
|
|
});
|
|
});
|
|
});
|
|
}
|
|
|
|
function tokenizeWithoutErrors(input: string, options?: lex.TokenizeOptions): lex.Token[] {
|
|
const tokenizeResult = lex.tokenize(input, 'someUrl', getHtmlTagDefinition, options);
|
|
|
|
if (tokenizeResult.errors.length > 0) {
|
|
const errorString = tokenizeResult.errors.join('\n');
|
|
throw new Error(`Unexpected parse errors:\n${errorString}`);
|
|
}
|
|
|
|
return tokenizeResult.tokens;
|
|
}
|
|
|
|
function tokenizeAndHumanizeParts(input: string, options?: lex.TokenizeOptions): any[] {
|
|
return tokenizeWithoutErrors(input, options).map(token => [<any>token.type].concat(token.parts));
|
|
}
|
|
|
|
function tokenizeAndHumanizeSourceSpans(input: string, options?: lex.TokenizeOptions): any[] {
|
|
return tokenizeWithoutErrors(input, options)
|
|
.map(token => [<any>token.type, token.sourceSpan.toString()]);
|
|
}
|
|
|
|
function humanizeLineColumn(location: ParseLocation): string {
|
|
return `${location.line}:${location.col}`;
|
|
}
|
|
|
|
function tokenizeAndHumanizeLineColumn(input: string, options?: lex.TokenizeOptions): any[] {
|
|
return tokenizeWithoutErrors(input, options)
|
|
.map(token => [<any>token.type, humanizeLineColumn(token.sourceSpan.start)]);
|
|
}
|
|
|
|
function tokenizeAndHumanizeErrors(input: string, options?: lex.TokenizeOptions): any[] {
|
|
return lex.tokenize(input, 'someUrl', getHtmlTagDefinition, options)
|
|
.errors.map(e => [<any>e.tokenType, e.msg, humanizeLineColumn(e.span.start)]);
|
|
}
|