gitlab-org--gitlab-foss/spec/frontend/lib/utils/text_utility_spec.js

180 lines
6.1 KiB
JavaScript
Raw Normal View History

2017-11-13 04:11:54 -05:00
import * as textUtils from '~/lib/utils/text_utility';
2016-12-14 15:37:17 -05:00
describe('text_utility', () => {
2017-11-13 04:11:54 -05:00
describe('addDelimiter', () => {
it('should add a delimiter to the given string', () => {
expect(textUtils.addDelimiter('1234')).toEqual('1,234');
expect(textUtils.addDelimiter('222222')).toEqual('222,222');
});
2016-12-14 15:37:17 -05:00
2017-11-13 04:11:54 -05:00
it('should not add a delimiter if string contains no numbers', () => {
expect(textUtils.addDelimiter('aaaa')).toEqual('aaaa');
});
});
describe('highCountTrim', () => {
it('returns 99+ for count >= 100', () => {
2017-11-13 04:11:54 -05:00
expect(textUtils.highCountTrim(105)).toBe('99+');
expect(textUtils.highCountTrim(100)).toBe('99+');
});
it('returns exact number for count < 100', () => {
2017-11-13 04:11:54 -05:00
expect(textUtils.highCountTrim(45)).toBe(45);
2017-11-10 18:41:04 -05:00
});
});
2017-11-13 04:11:54 -05:00
describe('humanize', () => {
it('should remove underscores and uppercase the first letter', () => {
expect(textUtils.humanize('foo_bar')).toEqual('Foo bar');
});
2017-11-13 04:11:54 -05:00
});
2017-11-13 04:11:54 -05:00
describe('pluralize', () => {
it('should pluralize given string', () => {
expect(textUtils.pluralize('test', 2)).toBe('tests');
});
2017-11-13 04:11:54 -05:00
it('should pluralize when count is 0', () => {
expect(textUtils.pluralize('test', 0)).toBe('tests');
});
2017-11-13 04:11:54 -05:00
it('should not pluralize when count is 1', () => {
expect(textUtils.pluralize('test', 1)).toBe('test');
});
});
2017-11-13 04:11:54 -05:00
describe('dasherize', () => {
it('should replace underscores with dashes', () => {
expect(textUtils.dasherize('foo_bar_foo')).toEqual('foo-bar-foo');
});
});
describe('capitalizeFirstCharacter', () => {
it('returns string with first letter capitalized', () => {
expect(textUtils.capitalizeFirstCharacter('gitlab')).toEqual('Gitlab');
});
});
describe('slugifyWithHyphens', () => {
it('should replaces whitespaces with hyphens and convert to lower case', () => {
expect(textUtils.slugifyWithHyphens('My Input String')).toEqual('my-input-string');
});
});
2018-01-19 04:38:34 -05:00
describe('stripHtml', () => {
it('replaces html tag with the default replacement', () => {
expect(textUtils.stripHtml('This is a text with <p>html</p>.')).toEqual(
'This is a text with html.',
);
});
it('replaces html tags with the provided replacement', () => {
expect(textUtils.stripHtml('This is a text with <p>html</p>.', ' ')).toEqual(
'This is a text with html .',
);
});
it('passes through with null string input', () => {
expect(textUtils.stripHtml(null, ' ')).toEqual(null);
});
it('passes through with undefined string input', () => {
expect(textUtils.stripHtml(undefined, ' ')).toEqual(undefined);
});
});
2018-02-07 07:57:44 -05:00
describe('convertToCamelCase', () => {
it('converts snake_case string to camelCase string', () => {
expect(textUtils.convertToCamelCase('snake_case')).toBe('snakeCase');
});
});
describe('convertToSentenceCase', () => {
it('converts Sentence Case to Sentence case', () => {
expect(textUtils.convertToSentenceCase('Hello World')).toBe('Hello world');
});
});
2018-06-21 08:22:40 -04:00
describe('truncateSha', () => {
it('shortens SHAs to 8 characters', () => {
expect(textUtils.truncateSha('verylongsha')).toBe('verylong');
});
it('leaves short SHAs as is', () => {
expect(textUtils.truncateSha('shortsha')).toBe('shortsha');
});
});
describe('splitCamelCase', () => {
it('separates a PascalCase word to two', () => {
expect(textUtils.splitCamelCase('HelloWorld')).toBe('Hello World');
});
});
describe('getFirstCharacterCapitalized', () => {
2018-10-30 06:53:01 -04:00
it('returns the first character capitalized, if first character is alphabetic', () => {
expect(textUtils.getFirstCharacterCapitalized('loremIpsumDolar')).toEqual('L');
expect(textUtils.getFirstCharacterCapitalized('Sit amit !')).toEqual('S');
});
it('returns the first character, if first character is non-alphabetic', () => {
expect(textUtils.getFirstCharacterCapitalized(' lorem')).toEqual(' ');
expect(textUtils.getFirstCharacterCapitalized('%#!')).toEqual('%');
});
it('returns an empty string, if string is falsey', () => {
expect(textUtils.getFirstCharacterCapitalized('')).toEqual('');
expect(textUtils.getFirstCharacterCapitalized(null)).toEqual('');
});
});
describe('truncatePathMiddleToLength', () => {
it('does not truncate text', () => {
expect(textUtils.truncatePathMiddleToLength('app/test', 50)).toEqual('app/test');
});
it('truncates middle of the path', () => {
expect(textUtils.truncatePathMiddleToLength('app/test/diff', 13)).toEqual('app/…/diff');
});
it('truncates multiple times in the middle of the path', () => {
expect(textUtils.truncatePathMiddleToLength('app/test/merge_request/diff', 13)).toEqual(
'app/…/…/diff',
);
});
});
describe('slugifyWithUnderscore', () => {
it('should replaces whitespaces with underscore and convert to lower case', () => {
expect(textUtils.slugifyWithUnderscore('My Input String')).toEqual('my_input_string');
});
});
describe('truncateNamespace', () => {
it(`should return the root namespace if the namespace only includes one level`, () => {
expect(textUtils.truncateNamespace('a / b')).toBe('a');
});
it(`should return the first 2 namespaces if the namespace inlcudes exactly 2 levels`, () => {
expect(textUtils.truncateNamespace('a / b / c')).toBe('a / b');
});
it(`should return the first and last namespaces, separated by "...", if the namespace inlcudes more than 2 levels`, () => {
expect(textUtils.truncateNamespace('a / b / c / d')).toBe('a / ... / c');
expect(textUtils.truncateNamespace('a / b / c / d / e / f / g / h / i')).toBe('a / ... / h');
});
it(`should return an empty string for invalid inputs`, () => {
[undefined, null, 4, {}, true, new Date()].forEach(input => {
expect(textUtils.truncateNamespace(input)).toBe('');
});
});
it(`should not alter strings that aren't formatted as namespaces`, () => {
['', ' ', '\t', 'a', 'a \\ b'].forEach(input => {
expect(textUtils.truncateNamespace(input)).toBe(input);
});
});
});
});