123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133 |
- if (process.env.OBJECT_IMPL) global.TYPED_ARRAY_SUPPORT = false
- var B = require('../').Buffer
- var test = require('tape')
- test('detect utf16 surrogate pairs', function (t) {
- var text = '\uD83D\uDE38' + '\uD83D\uDCAD' + '\uD83D\uDC4D'
- var buf = new B(text)
- t.equal(text, buf.toString())
- t.end()
- })
- test('detect utf16 surrogate pairs over U+20000 until U+10FFFF', function (t) {
- var text = '\uD842\uDFB7' + '\uD93D\uDCAD' + '\uDBFF\uDFFF'
- var buf = new B(text)
- t.equal(text, buf.toString())
- t.end()
- })
- test('replace orphaned utf16 surrogate lead code point', function (t) {
- var text = '\uD83D\uDE38' + '\uD83D' + '\uD83D\uDC4D'
- var buf = new B(text)
- t.deepEqual(buf, new B([ 0xf0, 0x9f, 0x98, 0xb8, 0xef, 0xbf, 0xbd, 0xf0, 0x9f, 0x91, 0x8d ]))
- t.end()
- })
- test('replace orphaned utf16 surrogate trail code point', function (t) {
- var text = '\uD83D\uDE38' + '\uDCAD' + '\uD83D\uDC4D'
- var buf = new B(text)
- t.deepEqual(buf, new B([ 0xf0, 0x9f, 0x98, 0xb8, 0xef, 0xbf, 0xbd, 0xf0, 0x9f, 0x91, 0x8d ]))
- t.end()
- })
- test('do not write partial utf16 code units', function (t) {
- var f = new B([0, 0, 0, 0, 0])
- t.equal(f.length, 5)
- var size = f.write('あいうえお', 'utf16le')
- t.equal(size, 4)
- t.deepEqual(f, new B([0x42, 0x30, 0x44, 0x30, 0x00]))
- t.end()
- })
- test('handle partial utf16 code points when encoding to utf8 the way node does', function (t) {
- var text = '\uD83D\uDE38' + '\uD83D\uDC4D'
- var buf = new B(8)
- buf.fill(0)
- buf.write(text)
- t.deepEqual(buf, new B([ 0xf0, 0x9f, 0x98, 0xb8, 0xf0, 0x9f, 0x91, 0x8d ]))
- buf = new B(7)
- buf.fill(0)
- buf.write(text)
- t.deepEqual(buf, new B([ 0xf0, 0x9f, 0x98, 0xb8, 0x00, 0x00, 0x00 ]))
- buf = new B(6)
- buf.fill(0)
- buf.write(text)
- t.deepEqual(buf, new B([ 0xf0, 0x9f, 0x98, 0xb8, 0x00, 0x00 ]))
- buf = new B(5)
- buf.fill(0)
- buf.write(text)
- t.deepEqual(buf, new B([ 0xf0, 0x9f, 0x98, 0xb8, 0x00 ]))
- buf = new B(4)
- buf.fill(0)
- buf.write(text)
- t.deepEqual(buf, new B([ 0xf0, 0x9f, 0x98, 0xb8 ]))
- buf = new B(3)
- buf.fill(0)
- buf.write(text)
- t.deepEqual(buf, new B([ 0x00, 0x00, 0x00 ]))
- buf = new B(2)
- buf.fill(0)
- buf.write(text)
- t.deepEqual(buf, new B([ 0x00, 0x00 ]))
- buf = new B(1)
- buf.fill(0)
- buf.write(text)
- t.deepEqual(buf, new B([ 0x00 ]))
- t.end()
- })
- test('handle invalid utf16 code points when encoding to utf8 the way node does', function (t) {
- var text = 'a' + '\uDE38\uD83D' + 'b'
- var buf = new B(8)
- buf.fill(0)
- buf.write(text)
- t.deepEqual(buf, new B([ 0x61, 0xef, 0xbf, 0xbd, 0xef, 0xbf, 0xbd, 0x62 ]))
- buf = new B(7)
- buf.fill(0)
- buf.write(text)
- t.deepEqual(buf, new B([ 0x61, 0xef, 0xbf, 0xbd, 0xef, 0xbf, 0xbd ]))
- buf = new B(6)
- buf.fill(0)
- buf.write(text)
- t.deepEqual(buf, new B([ 0x61, 0xef, 0xbf, 0xbd, 0x00, 0x00 ]))
- buf = new B(5)
- buf.fill(0)
- buf.write(text)
- t.deepEqual(buf, new B([ 0x61, 0xef, 0xbf, 0xbd, 0x00 ]))
- buf = new B(4)
- buf.fill(0)
- buf.write(text)
- t.deepEqual(buf, new B([ 0x61, 0xef, 0xbf, 0xbd ]))
- buf = new B(3)
- buf.fill(0)
- buf.write(text)
- t.deepEqual(buf, new B([ 0x61, 0x00, 0x00 ]))
- buf = new B(2)
- buf.fill(0)
- buf.write(text)
- t.deepEqual(buf, new B([ 0x61, 0x00 ]))
- buf = new B(1)
- buf.fill(0)
- buf.write(text)
- t.deepEqual(buf, new B([ 0x61 ]))
- t.end()
- })
|