175 lines
6.3 KiB
JavaScript
175 lines
6.3 KiB
JavaScript
import { cloneDeep, isScalar, parseRules, regexForFormat, snakeToCamel } from '@/libs/utils'
|
|
import rules from '@/validation/rules.ts'
|
|
|
|
describe('parseRules', () => {
|
|
it('parses single string rules, returning empty arguments array', () => {
|
|
expect(parseRules('required', rules)).toEqual([
|
|
[rules.required, [], 'required', null]
|
|
])
|
|
})
|
|
|
|
it('throws errors for invalid validation rules', () => {
|
|
expect(() => {
|
|
parseRules('required|notarule', rules, null)
|
|
}).toThrow()
|
|
})
|
|
|
|
it('parses arguments for a rule', () => {
|
|
expect(parseRules('in:foo,bar', rules)).toEqual([
|
|
[rules.in, ['foo', 'bar'], 'in', null]
|
|
])
|
|
})
|
|
|
|
it('parses multiple string rules and arguments', () => {
|
|
expect(parseRules('required|in:foo,bar', rules)).toEqual([
|
|
[rules.required, [], 'required', null],
|
|
[rules.in, ['foo', 'bar'], 'in', null]
|
|
])
|
|
})
|
|
|
|
it('parses multiple array rules and arguments', () => {
|
|
expect(parseRules(['required', 'in:foo,bar'], rules)).toEqual([
|
|
[rules.required, [], 'required', null],
|
|
[rules.in, ['foo', 'bar'], 'in', null]
|
|
])
|
|
})
|
|
|
|
it('parses array rules with expression arguments', () => {
|
|
expect(parseRules([
|
|
['matches', /^abc/, '1234']
|
|
], rules)).toEqual([
|
|
[rules.matches, [/^abc/, '1234'], 'matches', null]
|
|
])
|
|
})
|
|
|
|
it('parses string rules with caret modifier', () => {
|
|
expect(parseRules('^required|min:10', rules)).toEqual([
|
|
[rules.required, [], 'required', '^'],
|
|
[rules.min, ['10'], 'min', null],
|
|
])
|
|
})
|
|
|
|
it('parses array rule with caret modifier', () => {
|
|
expect(parseRules([['required'], ['^max', '10']], rules)).toEqual([
|
|
[rules.required, [], 'required', null],
|
|
[rules.max, ['10'], 'max', '^'],
|
|
])
|
|
})
|
|
})
|
|
|
|
|
|
describe('regexForFormat', () => {
|
|
it('allows MM format with other characters', () => expect(regexForFormat('abc/MM').test('abc/01')).toBe(true))
|
|
|
|
it('fails MM format with single digit', () => expect(regexForFormat('abc/MM').test('abc/1')).toBe(false))
|
|
|
|
it('allows M format with single digit', () => expect(regexForFormat('M/abc').test('1/abc')).toBe(true))
|
|
|
|
it('fails MM format when out of range', () => expect(regexForFormat('M/abc').test('13/abc')).toBe(false))
|
|
|
|
it('fails M format when out of range', () => expect(regexForFormat('M/abc').test('55/abc')).toBe(false))
|
|
|
|
it('Replaces double digits before singles', () => expect(regexForFormat('MMM').test('313131')).toBe(false))
|
|
|
|
it('allows DD format with zero digit', () => expect(regexForFormat('xyz/DD').test('xyz/01')).toBe(true))
|
|
|
|
it('fails DD format with single digit', () => expect(regexForFormat('xyz/DD').test('xyz/9')).toBe(false))
|
|
|
|
it('allows D format with single digit', () => expect(regexForFormat('xyz/D').test('xyz/9')).toBe(true))
|
|
|
|
it('fails D format with out of range digit', () => expect(regexForFormat('xyz/D').test('xyz/92')).toBe(false))
|
|
|
|
it('fails DD format with out of range digit', () => expect(regexForFormat('xyz/D').test('xyz/32')).toBe(false))
|
|
|
|
it('allows YY format with double zeros', () => expect(regexForFormat('YY').test('00')).toBe(true))
|
|
|
|
it('fails YY format with four zeros', () => expect(regexForFormat('YY').test('0000')).toBe(false))
|
|
|
|
it('allows YYYY format with four zeros', () => expect(regexForFormat('YYYY').test('0000')).toBe(true))
|
|
|
|
it('allows MD-YY', () => expect(regexForFormat('MD-YY').test('12-00')).toBe(true))
|
|
|
|
it('allows DM-YY', () => expect(regexForFormat('DM-YY').test('12-00')).toBe(true))
|
|
|
|
it('allows date like MM/DD/YYYY', () => expect(regexForFormat('MM/DD/YYYY').test('12/18/1987')).toBe(true))
|
|
|
|
it('allows date like YYYY-MM-DD', () => expect(regexForFormat('YYYY-MM-DD').test('1987-01-31')).toBe(true))
|
|
|
|
it('fails date like YYYY-MM-DD with out of bounds day', () => expect(regexForFormat('YYYY-MM-DD').test('1987-01-32')).toBe(false))
|
|
})
|
|
|
|
describe('isScalar', () => {
|
|
it('passes on strings', () => expect(isScalar('hello')).toBe(true))
|
|
|
|
it('passes on numbers', () => expect(isScalar(123)).toBe(true))
|
|
|
|
it('passes on booleans', () => expect(isScalar(false)).toBe(true))
|
|
|
|
it('passes on symbols', () => expect(isScalar(Symbol(123))).toBe(true))
|
|
|
|
it('passes on null', () => expect(isScalar(null)).toBe(true))
|
|
|
|
it('passes on undefined', () => expect(isScalar(undefined)).toBe(true))
|
|
|
|
it('fails on pojo', () => expect(isScalar({})).toBe(false))
|
|
})
|
|
|
|
describe('cloneDeep', () => {
|
|
it('basic objects stay the same', () => expect(cloneDeep({ a: 123, b: 'hello' })).toEqual({ a: 123, b: 'hello' }))
|
|
|
|
it('basic nested objects stay the same', () => {
|
|
expect(cloneDeep({ a: 123, b: { c: 'hello-world' } }))
|
|
.toEqual({ a: 123, b: { c: 'hello-world' } })
|
|
})
|
|
|
|
it('simple pojo reference types are re-created', () => {
|
|
const c = { c: 'hello-world' }
|
|
const clone = cloneDeep({ a: 123, b: c })
|
|
expect(clone.b === c).toBe(false)
|
|
})
|
|
|
|
it('retains array structures inside of a pojo', () => {
|
|
const obj = { a: 'abcd', d: ['first', 'second'] }
|
|
const clone = cloneDeep(obj)
|
|
expect(Array.isArray(clone.d)).toBe(true)
|
|
})
|
|
|
|
it('removes references inside array structures', () => {
|
|
const deepObj = {foo: 'bar'}
|
|
const obj = { a: 'abcd', d: ['first', deepObj] }
|
|
const clone = cloneDeep(obj)
|
|
expect(clone.d[1] === deepObj).toBe(false)
|
|
})
|
|
})
|
|
|
|
describe('snakeToCamel', () => {
|
|
it('converts underscore separated words to camelCase', () => {
|
|
expect(snakeToCamel('this_is_snake_case')).toBe('thisIsSnakeCase')
|
|
})
|
|
|
|
it('converts underscore separated words to camelCase even if they start with a number', () => {
|
|
expect(snakeToCamel('this_is_snake_case_2nd_example')).toBe('thisIsSnakeCase2ndExample')
|
|
})
|
|
|
|
it('has no effect on already camelCase words', () => {
|
|
expect(snakeToCamel('thisIsCamelCase')).toBe('thisIsCamelCase')
|
|
})
|
|
|
|
it('does not capitalize the first word or strip first underscore if a phrase starts with an underscore', () => {
|
|
expect(snakeToCamel('_this_starts_with_an_underscore')).toBe('_thisStartsWithAnUnderscore')
|
|
})
|
|
|
|
it('ignores double underscores anywhere in a word', () => {
|
|
expect(snakeToCamel('__unlikely__thing__')).toBe('__unlikely__thing__')
|
|
})
|
|
|
|
it('has no effect hyphenated words', () => {
|
|
expect(snakeToCamel('not-a-good-name')).toBe('not-a-good-name')
|
|
})
|
|
|
|
it('returns the same function if passed', () => {
|
|
const fn = () => {}
|
|
expect(snakeToCamel(fn)).toBe(fn)
|
|
})
|
|
})
|