79 lines
		
	
	
		
			2.8 KiB
		
	
	
	
		
			TypeScript
		
	
	
	
	
	
			
		
		
	
	
			79 lines
		
	
	
		
			2.8 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 {fakeAsync} from '@angular/core/testing/src/fake_async';
 | |
| import {SyncAsync, escapeRegExp, splitAtColon, utf8Encode} from '../src/util';
 | |
| 
 | |
| {
 | |
|   describe('util', () => {
 | |
|     describe('splitAtColon', () => {
 | |
|       it('should split when a single ":" is present', () => {
 | |
|         expect(splitAtColon('a:b', [])).toEqual(['a', 'b']);
 | |
|       });
 | |
| 
 | |
|       it('should trim parts', () => { expect(splitAtColon(' a : b ', [])).toEqual(['a', 'b']); });
 | |
| 
 | |
|       it('should support multiple ":"', () => {
 | |
|         expect(splitAtColon('a:b:c', [])).toEqual(['a', 'b:c']);
 | |
|       });
 | |
| 
 | |
|       it('should use the default value when no ":" is present', () => {
 | |
|         expect(splitAtColon('ab', ['c', 'd'])).toEqual(['c', 'd']);
 | |
|       });
 | |
|     });
 | |
| 
 | |
|     describe('RegExp', () => {
 | |
|       it('should escape regexp', () => {
 | |
|         expect(new RegExp(escapeRegExp('b')).exec('abc')).toBeTruthy();
 | |
|         expect(new RegExp(escapeRegExp('b')).exec('adc')).toBeFalsy();
 | |
|         expect(new RegExp(escapeRegExp('a.b')).exec('a.b')).toBeTruthy();
 | |
|         expect(new RegExp(escapeRegExp('a.b')).exec('axb')).toBeFalsy();
 | |
|       });
 | |
|     });
 | |
| 
 | |
|     describe('utf8encode', () => {
 | |
|       // tests from https://github.com/mathiasbynens/wtf-8
 | |
|       it('should encode to utf8', () => {
 | |
|         const tests = [
 | |
|           ['abc', 'abc'],
 | |
|           // // 1-byte
 | |
|           ['\0', '\0'],
 | |
|           // // 2-byte
 | |
|           ['\u0080', '\xc2\x80'],
 | |
|           ['\u05ca', '\xd7\x8a'],
 | |
|           ['\u07ff', '\xdf\xbf'],
 | |
|           // // 3-byte
 | |
|           ['\u0800', '\xe0\xa0\x80'],
 | |
|           ['\u2c3c', '\xe2\xb0\xbc'],
 | |
|           ['\uffff', '\xef\xbf\xbf'],
 | |
|           // //4-byte
 | |
|           ['\uD800\uDC00', '\xF0\x90\x80\x80'],
 | |
|           ['\uD834\uDF06', '\xF0\x9D\x8C\x86'],
 | |
|           ['\uDBFF\uDFFF', '\xF4\x8F\xBF\xBF'],
 | |
|           // unmatched surrogate halves
 | |
|           // high surrogates: 0xD800 to 0xDBFF
 | |
|           ['\uD800', '\xED\xA0\x80'],
 | |
|           ['\uD800\uD800', '\xED\xA0\x80\xED\xA0\x80'],
 | |
|           ['\uD800A', '\xED\xA0\x80A'],
 | |
|           ['\uD800\uD834\uDF06\uD800', '\xED\xA0\x80\xF0\x9D\x8C\x86\xED\xA0\x80'],
 | |
|           ['\uD9AF', '\xED\xA6\xAF'],
 | |
|           ['\uDBFF', '\xED\xAF\xBF'],
 | |
|           // low surrogates: 0xDC00 to 0xDFFF
 | |
|           ['\uDC00', '\xED\xB0\x80'],
 | |
|           ['\uDC00\uDC00', '\xED\xB0\x80\xED\xB0\x80'],
 | |
|           ['\uDC00A', '\xED\xB0\x80A'],
 | |
|           ['\uDC00\uD834\uDF06\uDC00', '\xED\xB0\x80\xF0\x9D\x8C\x86\xED\xB0\x80'],
 | |
|           ['\uDEEE', '\xED\xBB\xAE'],
 | |
|           ['\uDFFF', '\xED\xBF\xBF'],
 | |
|         ];
 | |
|         tests.forEach(
 | |
|             ([input, output]: [string, string]) => { expect(utf8Encode(input)).toEqual(output); });
 | |
|       });
 | |
|     });
 | |
|   });
 | |
| } |