node/test/parallel/test-buffer-alloc.js
Zach Bjornson 7420835390 src: fix build for older clang
Removes use of builtins that are unavailable for older clang. Per
benchmarks, only uses builtins on Windows, where speedup is
significant.

Also adds test for unaligned ucs2 buffer write. Between #3410
and #7645, bytes were swapped twice on bigendian platforms if buffer
was not two-byte aligned. See comment in #7645.

PR-URL: https://github.com/nodejs/node/pull/7645
Fixes: https://github.com/nodejs/node/issues/7618
Reviewed-By: Anna Henningsen <anna@addaleax.net>
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Reviewed-By: James M Snell <jasnell@gmail.com>
2016-10-04 09:28:18 +01:00

1096 lines
36 KiB
JavaScript

'use strict';
const common = require('../common');
const assert = require('assert');
const vm = require('vm');
const Buffer = require('buffer').Buffer;
const SlowBuffer = require('buffer').SlowBuffer;
const b = Buffer.allocUnsafe(1024);
assert.strictEqual(1024, b.length);
b[0] = -1;
assert.strictEqual(b[0], 255);
for (let i = 0; i < 1024; i++) {
b[i] = i % 256;
}
for (let i = 0; i < 1024; i++) {
assert.strictEqual(i % 256, b[i]);
}
const c = Buffer.allocUnsafe(512);
assert.strictEqual(512, c.length);
const d = Buffer.from([]);
assert.strictEqual(0, d.length);
// Test offset properties
{
const b = Buffer.alloc(128);
assert.strictEqual(128, b.length);
assert.strictEqual(0, b.byteOffset);
assert.strictEqual(0, b.offset);
}
// Test creating a Buffer from a Uint32Array
{
const ui32 = new Uint32Array(4).fill(42);
const e = Buffer.from(ui32);
for (const [index, value] of e.entries()) {
assert.strictEqual(value, ui32[index]);
}
}
// Test creating a Buffer from a Uint32Array (old constructor)
{
const ui32 = new Uint32Array(4).fill(42);
const e = Buffer(ui32);
for (const [key, value] of e.entries()) {
assert.deepStrictEqual(value, ui32[key]);
}
}
// Test invalid encoding for Buffer.toString
assert.throws(() => b.toString('invalid'),
/Unknown encoding: invalid/);
// invalid encoding for Buffer.write
assert.throws(() => b.write('test string', 0, 5, 'invalid'),
/Unknown encoding: invalid/);
// unsupported arguments for Buffer.write
assert.throws(() => b.write('test', 'utf8', 0),
/is no longer supported/);
// try to create 0-length buffers
assert.doesNotThrow(() => Buffer.from(''));
assert.doesNotThrow(() => Buffer.from('', 'ascii'));
assert.doesNotThrow(() => Buffer.from('', 'latin1'));
assert.doesNotThrow(() => Buffer.alloc(0));
assert.doesNotThrow(() => Buffer.allocUnsafe(0));
assert.doesNotThrow(() => new Buffer(''));
assert.doesNotThrow(() => new Buffer('', 'ascii'));
assert.doesNotThrow(() => new Buffer('', 'latin1'));
assert.doesNotThrow(() => new Buffer('', 'binary'));
assert.doesNotThrow(() => Buffer(0));
// try to write a 0-length string beyond the end of b
assert.throws(() => b.write('', 2048), RangeError);
// throw when writing to negative offset
assert.throws(() => b.write('a', -1), RangeError);
// throw when writing past bounds from the pool
assert.throws(() => b.write('a', 2048), RangeError);
// throw when writing to negative offset
assert.throws(() => b.write('a', -1), RangeError);
// try to copy 0 bytes worth of data into an empty buffer
b.copy(Buffer.alloc(0), 0, 0, 0);
// try to copy 0 bytes past the end of the target buffer
b.copy(Buffer.alloc(0), 1, 1, 1);
b.copy(Buffer.alloc(1), 1, 1, 1);
// try to copy 0 bytes from past the end of the source buffer
b.copy(Buffer.alloc(1), 0, 2048, 2048);
// testing for smart defaults and ability to pass string values as offset
{
const writeTest = Buffer.from('abcdes');
writeTest.write('n', 'ascii');
writeTest.write('o', '1', 'ascii');
writeTest.write('d', '2', 'ascii');
writeTest.write('e', 3, 'ascii');
writeTest.write('j', 4, 'ascii');
assert.strictEqual(writeTest.toString(), 'nodejs');
}
// Offset points to the end of the buffer
// (see https://github.com/nodejs/node/issues/8127).
assert.doesNotThrow(() => Buffer.alloc(1).write('', 1, 0));
// ASCII slice test
{
const asciiString = 'hello world';
for (let i = 0; i < asciiString.length; i++) {
b[i] = asciiString.charCodeAt(i);
}
const asciiSlice = b.toString('ascii', 0, asciiString.length);
assert.strictEqual(asciiString, asciiSlice);
}
{
const asciiString = 'hello world';
const offset = 100;
assert.strictEqual(asciiString.length, b.write(asciiString, offset, 'ascii'));
const asciiSlice = b.toString('ascii', offset, offset + asciiString.length);
assert.strictEqual(asciiString, asciiSlice);
}
{
const asciiString = 'hello world';
const offset = 100;
const sliceA = b.slice(offset, offset + asciiString.length);
const sliceB = b.slice(offset, offset + asciiString.length);
for (let i = 0; i < asciiString.length; i++) {
assert.equal(sliceA[i], sliceB[i]);
}
}
// UTF-8 slice test
{
const utf8String = '¡hέlló wôrld!';
const offset = 100;
b.write(utf8String, 0, Buffer.byteLength(utf8String), 'utf8');
let utf8Slice = b.toString('utf8', 0, Buffer.byteLength(utf8String));
assert.equal(utf8String, utf8Slice);
assert.strictEqual(Buffer.byteLength(utf8String),
b.write(utf8String, offset, 'utf8'));
utf8Slice = b.toString('utf8', offset,
offset + Buffer.byteLength(utf8String));
assert.strictEqual(utf8String, utf8Slice);
const sliceA = b.slice(offset, offset + Buffer.byteLength(utf8String));
const sliceB = b.slice(offset, offset + Buffer.byteLength(utf8String));
for (let i = 0; i < Buffer.byteLength(utf8String); i++) {
assert.strictEqual(sliceA[i], sliceB[i]);
}
}
{
const slice = b.slice(100, 150);
assert.strictEqual(50, slice.length);
for (let i = 0; i < 50; i++) {
assert.strictEqual(b[100 + i], slice[i]);
}
}
{
// make sure only top level parent propagates from allocPool
const b = Buffer.allocUnsafe(5);
const c = b.slice(0, 4);
const d = c.slice(0, 2);
assert.strictEqual(b.parent, c.parent);
assert.strictEqual(b.parent, d.parent);
}
{
// also from a non-pooled instance
const b = Buffer.allocUnsafeSlow(5);
const c = b.slice(0, 4);
const d = c.slice(0, 2);
assert.strictEqual(c.parent, d.parent);
}
{
// Bug regression test
const testValue = '\u00F6\u65E5\u672C\u8A9E'; // ö日本語
const buffer = Buffer.allocUnsafe(32);
const size = buffer.write(testValue, 0, 'utf8');
const slice = buffer.toString('utf8', 0, size);
assert.strictEqual(slice, testValue);
}
{
// Test triple slice
const a = Buffer.allocUnsafe(8);
for (let i = 0; i < 8; i++) a[i] = i;
const b = a.slice(4, 8);
assert.strictEqual(4, b[0]);
assert.strictEqual(5, b[1]);
assert.strictEqual(6, b[2]);
assert.strictEqual(7, b[3]);
const c = b.slice(2, 4);
assert.strictEqual(6, c[0]);
assert.strictEqual(7, c[1]);
}
{
const d = Buffer.from([23, 42, 255]);
assert.strictEqual(d.length, 3);
assert.strictEqual(d[0], 23);
assert.strictEqual(d[1], 42);
assert.strictEqual(d[2], 255);
assert.deepStrictEqual(d, Buffer.from(d));
}
{
// Test for proper UTF-8 Encoding
const e = Buffer.from('über');
assert.deepStrictEqual(e, Buffer.from([195, 188, 98, 101, 114]));
}
{
// Test for proper ascii Encoding, length should be 4
const f = Buffer.from('über', 'ascii');
assert.deepStrictEqual(f, Buffer.from([252, 98, 101, 114]));
}
['ucs2', 'ucs-2', 'utf16le', 'utf-16le'].forEach((encoding) => {
{
// Test for proper UTF16LE encoding, length should be 8
const f = Buffer.from('über', encoding);
assert.deepStrictEqual(f, Buffer.from([252, 0, 98, 0, 101, 0, 114, 0]));
}
{
// Length should be 12
const f = Buffer.from('привет', encoding);
assert.deepStrictEqual(
f, Buffer.from([63, 4, 64, 4, 56, 4, 50, 4, 53, 4, 66, 4])
);
assert.strictEqual(f.toString(encoding), 'привет');
}
{
const f = Buffer.from([0, 0, 0, 0, 0]);
assert.strictEqual(f.length, 5);
const size = f.write('あいうえお', encoding);
assert.strictEqual(size, 4);
assert.deepStrictEqual(f, Buffer.from([0x42, 0x30, 0x44, 0x30, 0x00]));
}
});
{
const f = Buffer.from('\uD83D\uDC4D', 'utf-16le'); // THUMBS UP SIGN (U+1F44D)
assert.strictEqual(f.length, 4);
assert.deepStrictEqual(f, Buffer.from('3DD84DDC', 'hex'));
}
// Test construction from arrayish object
{
const arrayIsh = {0: 0, 1: 1, 2: 2, 3: 3, length: 4};
let g = Buffer.from(arrayIsh);
assert.deepStrictEqual(g, Buffer.from([0, 1, 2, 3]));
const strArrayIsh = {0: '0', 1: '1', 2: '2', 3: '3', length: 4};
g = Buffer.from(strArrayIsh);
assert.deepStrictEqual(g, Buffer.from([0, 1, 2, 3]));
}
//
// Test toString('base64')
//
assert.strictEqual('TWFu', (Buffer.from('Man')).toString('base64'));
{
// test that regular and URL-safe base64 both work
const expected = [0xff, 0xff, 0xbe, 0xff, 0xef, 0xbf, 0xfb, 0xef, 0xff];
assert.deepStrictEqual(Buffer.from('//++/++/++//', 'base64'),
Buffer.from(expected));
assert.deepStrictEqual(Buffer.from('__--_--_--__', 'base64'),
Buffer.from(expected));
}
{
// big example
const quote = 'Man is distinguished, not only by his reason, but by this ' +
'singular passion from other animals, which is a lust ' +
'of the mind, that by a perseverance of delight in the ' +
'continued and indefatigable generation of knowledge, ' +
'exceeds the short vehemence of any carnal pleasure.';
const expected = 'TWFuIGlzIGRpc3Rpbmd1aXNoZWQsIG5vdCBvbmx5IGJ5IGhpcyByZWFzb' +
'24sIGJ1dCBieSB0aGlzIHNpbmd1bGFyIHBhc3Npb24gZnJvbSBvdGhlci' +
'BhbmltYWxzLCB3aGljaCBpcyBhIGx1c3Qgb2YgdGhlIG1pbmQsIHRoYXQ' +
'gYnkgYSBwZXJzZXZlcmFuY2Ugb2YgZGVsaWdodCBpbiB0aGUgY29udGlu' +
'dWVkIGFuZCBpbmRlZmF0aWdhYmxlIGdlbmVyYXRpb24gb2Yga25vd2xlZ' +
'GdlLCBleGNlZWRzIHRoZSBzaG9ydCB2ZWhlbWVuY2Ugb2YgYW55IGNhcm' +
'5hbCBwbGVhc3VyZS4=';
assert.strictEqual(expected, (Buffer.from(quote)).toString('base64'));
let b = Buffer.allocUnsafe(1024);
let bytesWritten = b.write(expected, 0, 'base64');
assert.strictEqual(quote.length, bytesWritten);
assert.strictEqual(quote, b.toString('ascii', 0, quote.length));
// check that the base64 decoder ignores whitespace
const expectedWhite = expected.slice(0, 60) + ' \n' +
expected.slice(60, 120) + ' \n' +
expected.slice(120, 180) + ' \n' +
expected.slice(180, 240) + ' \n' +
expected.slice(240, 300) + '\n' +
expected.slice(300, 360) + '\n';
b = Buffer.allocUnsafe(1024);
bytesWritten = b.write(expectedWhite, 0, 'base64');
assert.strictEqual(quote.length, bytesWritten);
assert.strictEqual(quote, b.toString('ascii', 0, quote.length));
// check that the base64 decoder on the constructor works
// even in the presence of whitespace.
b = Buffer.from(expectedWhite, 'base64');
assert.strictEqual(quote.length, b.length);
assert.strictEqual(quote, b.toString('ascii', 0, quote.length));
// check that the base64 decoder ignores illegal chars
const expectedIllegal = expected.slice(0, 60) + ' \x80' +
expected.slice(60, 120) + ' \xff' +
expected.slice(120, 180) + ' \x00' +
expected.slice(180, 240) + ' \x98' +
expected.slice(240, 300) + '\x03' +
expected.slice(300, 360);
b = Buffer.from(expectedIllegal, 'base64');
assert.strictEqual(quote.length, b.length);
assert.strictEqual(quote, b.toString('ascii', 0, quote.length));
}
assert.strictEqual(Buffer.from('', 'base64').toString(), '');
assert.strictEqual(Buffer.from('K', 'base64').toString(), '');
// multiple-of-4 with padding
assert.strictEqual(Buffer.from('Kg==', 'base64').toString(), '*');
assert.strictEqual(Buffer.from('Kio=', 'base64').toString(), '*'.repeat(2));
assert.strictEqual(Buffer.from('Kioq', 'base64').toString(), '*'.repeat(3));
assert.strictEqual(Buffer.from('KioqKg==', 'base64').toString(), '*'.repeat(4));
assert.strictEqual(Buffer.from('KioqKio=', 'base64').toString(), '*'.repeat(5));
assert.strictEqual(Buffer.from('KioqKioq', 'base64').toString(), '*'.repeat(6));
assert.strictEqual(Buffer.from('KioqKioqKg==', 'base64').toString(),
'*'.repeat(7));
assert.strictEqual(Buffer.from('KioqKioqKio=', 'base64').toString(),
'*'.repeat(8));
assert.strictEqual(Buffer.from('KioqKioqKioq', 'base64').toString(),
'*'.repeat(9));
assert.strictEqual(Buffer.from('KioqKioqKioqKg==', 'base64').toString(),
'*'.repeat(10));
assert.strictEqual(Buffer.from('KioqKioqKioqKio=', 'base64').toString(),
'*'.repeat(11));
assert.strictEqual(Buffer.from('KioqKioqKioqKioq', 'base64').toString(),
'*'.repeat(12));
assert.strictEqual(Buffer.from('KioqKioqKioqKioqKg==', 'base64').toString(),
'*'.repeat(13));
assert.strictEqual(Buffer.from('KioqKioqKioqKioqKio=', 'base64').toString(),
'*'.repeat(14));
assert.strictEqual(Buffer.from('KioqKioqKioqKioqKioq', 'base64').toString(),
'*'.repeat(15));
assert.strictEqual(Buffer.from('KioqKioqKioqKioqKioqKg==', 'base64').toString(),
'*'.repeat(16));
assert.strictEqual(Buffer.from('KioqKioqKioqKioqKioqKio=', 'base64').toString(),
'*'.repeat(17));
assert.strictEqual(Buffer.from('KioqKioqKioqKioqKioqKioq', 'base64').toString(),
'*'.repeat(18));
assert.strictEqual(Buffer.from('KioqKioqKioqKioqKioqKioqKg==',
'base64').toString(),
'*'.repeat(19));
assert.strictEqual(Buffer.from('KioqKioqKioqKioqKioqKioqKio=',
'base64').toString(),
'*'.repeat(20));
// no padding, not a multiple of 4
assert.strictEqual(Buffer.from('Kg', 'base64').toString(), '*');
assert.strictEqual(Buffer.from('Kio', 'base64').toString(), '*'.repeat(2));
assert.strictEqual(Buffer.from('KioqKg', 'base64').toString(), '*'.repeat(4));
assert.strictEqual(Buffer.from('KioqKio', 'base64').toString(), '*'.repeat(5));
assert.strictEqual(Buffer.from('KioqKioqKg', 'base64').toString(),
'*'.repeat(7));
assert.strictEqual(Buffer.from('KioqKioqKio', 'base64').toString(),
'*'.repeat(8));
assert.strictEqual(Buffer.from('KioqKioqKioqKg', 'base64').toString(),
'*'.repeat(10));
assert.strictEqual(Buffer.from('KioqKioqKioqKio', 'base64').toString(),
'*'.repeat(11));
assert.strictEqual(Buffer.from('KioqKioqKioqKioqKg', 'base64').toString(),
'*'.repeat(13));
assert.strictEqual(Buffer.from('KioqKioqKioqKioqKio', 'base64').toString(),
'*'.repeat(14));
assert.strictEqual(Buffer.from('KioqKioqKioqKioqKioqKg', 'base64').toString(),
'*'.repeat(16));
assert.strictEqual(Buffer.from('KioqKioqKioqKioqKioqKio', 'base64').toString(),
'*'.repeat(17));
assert.strictEqual(Buffer.from('KioqKioqKioqKioqKioqKioqKg',
'base64').toString(),
'*'.repeat(19));
assert.strictEqual(Buffer.from('KioqKioqKioqKioqKioqKioqKio',
'base64').toString(),
'*'.repeat(20));
// handle padding graciously, multiple-of-4 or not
assert.strictEqual(
Buffer.from('72INjkR5fchcxk9+VgdGPFJDxUBFR5/rMFsghgxADiw==', 'base64').length,
32
);
assert.strictEqual(
Buffer.from('72INjkR5fchcxk9+VgdGPFJDxUBFR5/rMFsghgxADiw=', 'base64').length,
32
);
assert.strictEqual(
Buffer.from('72INjkR5fchcxk9+VgdGPFJDxUBFR5/rMFsghgxADiw', 'base64').length,
32
);
assert.strictEqual(
Buffer.from('w69jACy6BgZmaFvv96HG6MYksWytuZu3T1FvGnulPg==', 'base64').length,
31
);
assert.strictEqual(
Buffer.from('w69jACy6BgZmaFvv96HG6MYksWytuZu3T1FvGnulPg=', 'base64').length,
31
);
assert.strictEqual(
Buffer.from('w69jACy6BgZmaFvv96HG6MYksWytuZu3T1FvGnulPg', 'base64').length,
31
);
{
// This string encodes single '.' character in UTF-16
const dot = Buffer.from('//4uAA==', 'base64');
assert.strictEqual(dot[0], 0xff);
assert.strictEqual(dot[1], 0xfe);
assert.strictEqual(dot[2], 0x2e);
assert.strictEqual(dot[3], 0x00);
assert.strictEqual(dot.toString('base64'), '//4uAA==');
}
{
// Writing base64 at a position > 0 should not mangle the result.
//
// https://github.com/joyent/node/issues/402
const segments = ['TWFkbmVzcz8h', 'IFRoaXM=', 'IGlz', 'IG5vZGUuanMh'];
const b = Buffer.allocUnsafe(64);
let pos = 0;
for (let i = 0; i < segments.length; ++i) {
pos += b.write(segments[i], pos, 'base64');
}
assert.strictEqual(b.toString('latin1', 0, pos),
'Madness?! This is node.js!');
}
// Regression test for https://github.com/nodejs/node/issues/3496.
assert.strictEqual(Buffer.from('=bad'.repeat(1e4), 'base64').length, 0);
{
// Creating buffers larger than pool size.
const l = Buffer.poolSize + 5;
const s = 'h'.repeat(l);
const b = Buffer.from(s);
for (let i = 0; i < l; i++) {
assert.strictEqual('h'.charCodeAt(0), b[i]);
}
const sb = b.toString();
assert.strictEqual(sb.length, s.length);
assert.strictEqual(sb, s);
}
{
// test hex toString
const hexb = Buffer.allocUnsafe(256);
for (let i = 0; i < 256; i++) {
hexb[i] = i;
}
const hexStr = hexb.toString('hex');
assert.strictEqual(hexStr,
'000102030405060708090a0b0c0d0e0f' +
'101112131415161718191a1b1c1d1e1f' +
'202122232425262728292a2b2c2d2e2f' +
'303132333435363738393a3b3c3d3e3f' +
'404142434445464748494a4b4c4d4e4f' +
'505152535455565758595a5b5c5d5e5f' +
'606162636465666768696a6b6c6d6e6f' +
'707172737475767778797a7b7c7d7e7f' +
'808182838485868788898a8b8c8d8e8f' +
'909192939495969798999a9b9c9d9e9f' +
'a0a1a2a3a4a5a6a7a8a9aaabacadaeaf' +
'b0b1b2b3b4b5b6b7b8b9babbbcbdbebf' +
'c0c1c2c3c4c5c6c7c8c9cacbcccdcecf' +
'd0d1d2d3d4d5d6d7d8d9dadbdcdddedf' +
'e0e1e2e3e4e5e6e7e8e9eaebecedeeef' +
'f0f1f2f3f4f5f6f7f8f9fafbfcfdfeff');
const hexb2 = Buffer.from(hexStr, 'hex');
for (let i = 0; i < 256; i++) {
assert.strictEqual(hexb2[i], hexb[i]);
}
}
// Test single hex character throws TypeError
// - https://github.com/nodejs/node/issues/6770
assert.throws(() => Buffer.from('A', 'hex'), TypeError);
// Test single base64 char encodes as 0
assert.strictEqual(Buffer.from('A', 'base64').length, 0);
{
// test an invalid slice end.
const b = Buffer.from([1, 2, 3, 4, 5]);
const b2 = b.toString('hex', 1, 10000);
const b3 = b.toString('hex', 1, 5);
const b4 = b.toString('hex', 1);
assert.strictEqual(b2, b3);
assert.strictEqual(b2, b4);
}
function buildBuffer(data) {
if (Array.isArray(data)) {
const buffer = Buffer.allocUnsafe(data.length);
data.forEach((v, k) => buffer[k] = v);
return buffer;
}
return null;
}
const x = buildBuffer([0x81, 0xa3, 0x66, 0x6f, 0x6f, 0xa3, 0x62, 0x61, 0x72]);
assert.strictEqual('<Buffer 81 a3 66 6f 6f a3 62 61 72>', x.inspect());
{
const z = x.slice(4);
assert.strictEqual(5, z.length);
assert.strictEqual(0x6f, z[0]);
assert.strictEqual(0xa3, z[1]);
assert.strictEqual(0x62, z[2]);
assert.strictEqual(0x61, z[3]);
assert.strictEqual(0x72, z[4]);
}
{
const z = x.slice(0);
assert.strictEqual(z.length, x.length);
}
{
const z = x.slice(0, 4);
assert.strictEqual(4, z.length);
assert.strictEqual(0x81, z[0]);
assert.strictEqual(0xa3, z[1]);
}
{
const z = x.slice(0, 9);
assert.strictEqual(9, z.length);
}
{
const z = x.slice(1, 4);
assert.strictEqual(3, z.length);
assert.strictEqual(0xa3, z[0]);
}
{
const z = x.slice(2, 4);
assert.strictEqual(2, z.length);
assert.strictEqual(0x66, z[0]);
assert.strictEqual(0x6f, z[1]);
}
['ucs2', 'ucs-2', 'utf16le', 'utf-16le'].forEach((encoding) => {
const b = Buffer.allocUnsafe(10);
b.write('あいうえお', encoding);
assert.strictEqual(b.toString(encoding), 'あいうえお');
});
['ucs2', 'ucs-2', 'utf16le', 'utf-16le'].forEach((encoding) => {
const b = Buffer.allocUnsafe(11);
b.write('あいうえお', 1, encoding);
assert.strictEqual(b.toString(encoding, 1), 'あいうえお');
});
{
// latin1 encoding should write only one byte per character.
const b = Buffer.from([0xde, 0xad, 0xbe, 0xef]);
let s = String.fromCharCode(0xffff);
b.write(s, 0, 'latin1');
assert.strictEqual(0xff, b[0]);
assert.strictEqual(0xad, b[1]);
assert.strictEqual(0xbe, b[2]);
assert.strictEqual(0xef, b[3]);
s = String.fromCharCode(0xaaee);
b.write(s, 0, 'latin1');
assert.strictEqual(0xee, b[0]);
assert.strictEqual(0xad, b[1]);
assert.strictEqual(0xbe, b[2]);
assert.strictEqual(0xef, b[3]);
}
{
// Binary encoding should write only one byte per character.
const b = Buffer.from([0xde, 0xad, 0xbe, 0xef]);
let s = String.fromCharCode(0xffff);
b.write(s, 0, 'latin1');
assert.strictEqual(0xff, b[0]);
assert.strictEqual(0xad, b[1]);
assert.strictEqual(0xbe, b[2]);
assert.strictEqual(0xef, b[3]);
s = String.fromCharCode(0xaaee);
b.write(s, 0, 'latin1');
assert.strictEqual(0xee, b[0]);
assert.strictEqual(0xad, b[1]);
assert.strictEqual(0xbe, b[2]);
assert.strictEqual(0xef, b[3]);
}
{
// #1210 Test UTF-8 string includes null character
let buf = Buffer.from('\0');
assert.strictEqual(buf.length, 1);
buf = Buffer.from('\0\0');
assert.strictEqual(buf.length, 2);
}
{
const buf = Buffer.allocUnsafe(2);
assert.strictEqual(buf.write(''), 0); //0bytes
assert.strictEqual(buf.write('\0'), 1); // 1byte (v8 adds null terminator)
assert.strictEqual(buf.write('a\0'), 2); // 1byte * 2
assert.strictEqual(buf.write('あ'), 0); // 3bytes
assert.strictEqual(buf.write('\0あ'), 1); // 1byte + 3bytes
assert.strictEqual(buf.write('\0\0あ'), 2); // 1byte * 2 + 3bytes
}
{
const buf = Buffer.allocUnsafe(10);
assert.strictEqual(buf.write('あいう'), 9); // 3bytes * 3 (v8 adds null term.)
assert.strictEqual(buf.write('あいう\0'), 10); // 3bytes * 3 + 1byte
}
{
// #243 Test write() with maxLength
const buf = Buffer.allocUnsafe(4);
buf.fill(0xFF);
assert.strictEqual(buf.write('abcd', 1, 2, 'utf8'), 2);
assert.strictEqual(buf[0], 0xFF);
assert.strictEqual(buf[1], 0x61);
assert.strictEqual(buf[2], 0x62);
assert.strictEqual(buf[3], 0xFF);
buf.fill(0xFF);
assert.strictEqual(buf.write('abcd', 1, 4), 3);
assert.strictEqual(buf[0], 0xFF);
assert.strictEqual(buf[1], 0x61);
assert.strictEqual(buf[2], 0x62);
assert.strictEqual(buf[3], 0x63);
buf.fill(0xFF);
assert.strictEqual(buf.write('abcd', 1, 2, 'utf8'), 2);
assert.strictEqual(buf[0], 0xFF);
assert.strictEqual(buf[1], 0x61);
assert.strictEqual(buf[2], 0x62);
assert.strictEqual(buf[3], 0xFF);
buf.fill(0xFF);
assert.strictEqual(buf.write('abcdef', 1, 2, 'hex'), 2);
assert.strictEqual(buf[0], 0xFF);
assert.strictEqual(buf[1], 0xAB);
assert.strictEqual(buf[2], 0xCD);
assert.strictEqual(buf[3], 0xFF);
['ucs2', 'ucs-2', 'utf16le', 'utf-16le'].forEach((encoding) => {
buf.fill(0xFF);
assert.strictEqual(buf.write('abcd', 0, 2, encoding), 2);
assert.strictEqual(buf[0], 0x61);
assert.strictEqual(buf[1], 0x00);
assert.strictEqual(buf[2], 0xFF);
assert.strictEqual(buf[3], 0xFF);
});
}
{
// test offset returns are correct
const b = Buffer.allocUnsafe(16);
assert.strictEqual(4, b.writeUInt32LE(0, 0));
assert.strictEqual(6, b.writeUInt16LE(0, 4));
assert.strictEqual(7, b.writeUInt8(0, 6));
assert.strictEqual(8, b.writeInt8(0, 7));
assert.strictEqual(16, b.writeDoubleLE(0, 8));
}
{
// test unmatched surrogates not producing invalid utf8 output
// ef bf bd = utf-8 representation of unicode replacement character
// see https://codereview.chromium.org/121173009/
const buf = Buffer.from('ab\ud800cd', 'utf8');
assert.strictEqual(buf[0], 0x61);
assert.strictEqual(buf[1], 0x62);
assert.strictEqual(buf[2], 0xef);
assert.strictEqual(buf[3], 0xbf);
assert.strictEqual(buf[4], 0xbd);
assert.strictEqual(buf[5], 0x63);
assert.strictEqual(buf[6], 0x64);
}
{
// test for buffer overrun
const buf = Buffer.from([0, 0, 0, 0, 0]); // length: 5
const sub = buf.slice(0, 4); // length: 4
assert.strictEqual(sub.write('12345', 'latin1'), 4);
assert.strictEqual(buf[4], 0);
assert.strictEqual(sub.write('12345', 'binary'), 4);
assert.strictEqual(buf[4], 0);
}
{
// test alloc with fill option
const buf = Buffer.alloc(5, '800A', 'hex');
assert.strictEqual(buf[0], 128);
assert.strictEqual(buf[1], 10);
assert.strictEqual(buf[2], 128);
assert.strictEqual(buf[3], 10);
assert.strictEqual(buf[4], 128);
}
// Check for fractional length args, junk length args, etc.
// https://github.com/joyent/node/issues/1758
// Call .fill() first, stops valgrind warning about uninitialized memory reads.
Buffer.allocUnsafe(3.3).fill().toString();
// throws bad argument error in commit 43cb4ec
Buffer.alloc(3.3).fill().toString();
assert.strictEqual(Buffer.allocUnsafe(NaN).length, 0);
assert.strictEqual(Buffer.allocUnsafe(3.3).length, 3);
assert.strictEqual(Buffer.from({length: 3.3}).length, 3);
assert.strictEqual(Buffer.from({length: 'BAM'}).length, 0);
// Make sure that strings are not coerced to numbers.
assert.strictEqual(Buffer.from('99').length, 2);
assert.strictEqual(Buffer.from('13.37').length, 5);
// Ensure that the length argument is respected.
['ascii', 'utf8', 'hex', 'base64', 'latin1', 'binary'].forEach((enc) => {
assert.strictEqual(Buffer.allocUnsafe(1).write('aaaaaa', 0, 1, enc), 1);
});
{
// Regression test, guard against buffer overrun in the base64 decoder.
const a = Buffer.allocUnsafe(3);
const b = Buffer.from('xxx');
a.write('aaaaaaaa', 'base64');
assert.strictEqual(b.toString(), 'xxx');
}
// issue GH-3416
Buffer.from(Buffer.allocUnsafe(0), 0, 0);
// GH-5110
{
const buffer = Buffer.from('test');
const string = JSON.stringify(buffer);
assert.strictEqual(string, '{"type":"Buffer","data":[116,101,115,116]}');
assert.deepStrictEqual(buffer, JSON.parse(string, (key, value) => {
return value && value.type === 'Buffer'
? Buffer.from(value.data)
: value;
}));
}
// issue GH-7849
{
const buf = Buffer.from('test');
const json = JSON.stringify(buf);
const obj = JSON.parse(json);
const copy = Buffer.from(obj);
assert(buf.equals(copy));
}
// issue GH-4331
assert.throws(() => Buffer.allocUnsafe(0xFFFFFFFF), RangeError);
assert.throws(() => Buffer.allocUnsafe(0xFFFFFFFFF), RangeError);
// issue GH-5587
assert.throws(() => Buffer.alloc(8).writeFloatLE(0, 5), RangeError);
assert.throws(() => Buffer.alloc(16).writeDoubleLE(0, 9), RangeError);
// attempt to overflow buffers, similar to previous bug in array buffers
assert.throws(() => Buffer.allocUnsafe(8).readFloatLE(0xffffffff),
RangeError);
assert.throws(() => Buffer.allocUnsafe(8).writeFloatLE(0.0, 0xffffffff),
RangeError);
assert.throws(() => Buffer.allocUnsafe(8).readFloatLE(0xffffffff),
RangeError);
assert.throws(() => Buffer.allocUnsafe(8).writeFloatLE(0.0, 0xffffffff),
RangeError);
// ensure negative values can't get past offset
assert.throws(() => Buffer.allocUnsafe(8).readFloatLE(-1), RangeError);
assert.throws(() => Buffer.allocUnsafe(8).writeFloatLE(0.0, -1), RangeError);
assert.throws(() => Buffer.allocUnsafe(8).readFloatLE(-1), RangeError);
assert.throws(() => Buffer.allocUnsafe(8).writeFloatLE(0.0, -1), RangeError);
// offset checks
{
const buf = Buffer.allocUnsafe(0);
assert.throws(() => buf.readUInt8(0), RangeError);
assert.throws(() => buf.readInt8(0), RangeError);
}
{
const buf = Buffer.from([0xFF]);
assert.strictEqual(buf.readUInt8(0), 255);
assert.strictEqual(buf.readInt8(0), -1);
}
[16, 32].forEach((bits) => {
const buf = Buffer.allocUnsafe(bits / 8 - 1);
assert.throws(() => buf[`readUInt${bits}BE`](0),
RangeError,
`readUInt${bits}BE()`);
assert.throws(() => buf[`readUInt${bits}LE`](0),
RangeError,
`readUInt${bits}LE()`);
assert.throws(() => buf[`readInt${bits}BE`](0),
RangeError,
`readInt${bits}BE()`);
assert.throws(() => buf[`readInt${bits}LE`](0),
RangeError,
`readInt${bits}LE()`);
});
[16, 32].forEach((bits) => {
const buf = Buffer.from([0xFF, 0xFF, 0xFF, 0xFF]);
assert.strictEqual(buf[`readUInt${bits}BE`](0),
(0xFFFFFFFF >>> (32 - bits)));
assert.strictEqual(buf[`readUInt${bits}LE`](0),
(0xFFFFFFFF >>> (32 - bits)));
assert.strictEqual(buf[`readInt${bits}BE`](0),
(0xFFFFFFFF >> (32 - bits)));
assert.strictEqual(buf[`readInt${bits}LE`](0),
(0xFFFFFFFF >> (32 - bits)));
});
// test for common read(U)IntLE/BE
{
const buf = Buffer.from([0x01, 0x02, 0x03, 0x04, 0x05, 0x06]);
assert.strictEqual(buf.readUIntLE(0, 1), 0x01);
assert.strictEqual(buf.readUIntBE(0, 1), 0x01);
assert.strictEqual(buf.readUIntLE(0, 3), 0x030201);
assert.strictEqual(buf.readUIntBE(0, 3), 0x010203);
assert.strictEqual(buf.readUIntLE(0, 5), 0x0504030201);
assert.strictEqual(buf.readUIntBE(0, 5), 0x0102030405);
assert.strictEqual(buf.readUIntLE(0, 6), 0x060504030201);
assert.strictEqual(buf.readUIntBE(0, 6), 0x010203040506);
assert.strictEqual(buf.readIntLE(0, 1), 0x01);
assert.strictEqual(buf.readIntBE(0, 1), 0x01);
assert.strictEqual(buf.readIntLE(0, 3), 0x030201);
assert.strictEqual(buf.readIntBE(0, 3), 0x010203);
assert.strictEqual(buf.readIntLE(0, 5), 0x0504030201);
assert.strictEqual(buf.readIntBE(0, 5), 0x0102030405);
assert.strictEqual(buf.readIntLE(0, 6), 0x060504030201);
assert.strictEqual(buf.readIntBE(0, 6), 0x010203040506);
}
// test for common write(U)IntLE/BE
{
let buf = Buffer.allocUnsafe(3);
buf.writeUIntLE(0x123456, 0, 3);
assert.deepStrictEqual(buf.toJSON().data, [0x56, 0x34, 0x12]);
assert.strictEqual(buf.readUIntLE(0, 3), 0x123456);
buf.fill(0xFF);
buf.writeUIntBE(0x123456, 0, 3);
assert.deepStrictEqual(buf.toJSON().data, [0x12, 0x34, 0x56]);
assert.strictEqual(buf.readUIntBE(0, 3), 0x123456);
buf.fill(0xFF);
buf.writeIntLE(0x123456, 0, 3);
assert.deepStrictEqual(buf.toJSON().data, [0x56, 0x34, 0x12]);
assert.strictEqual(buf.readIntLE(0, 3), 0x123456);
buf.fill(0xFF);
buf.writeIntBE(0x123456, 0, 3);
assert.deepStrictEqual(buf.toJSON().data, [0x12, 0x34, 0x56]);
assert.strictEqual(buf.readIntBE(0, 3), 0x123456);
buf.fill(0xFF);
buf.writeIntLE(-0x123456, 0, 3);
assert.deepStrictEqual(buf.toJSON().data, [0xaa, 0xcb, 0xed]);
assert.strictEqual(buf.readIntLE(0, 3), -0x123456);
buf.fill(0xFF);
buf.writeIntBE(-0x123456, 0, 3);
assert.deepStrictEqual(buf.toJSON().data, [0xed, 0xcb, 0xaa]);
assert.strictEqual(buf.readIntBE(0, 3), -0x123456);
buf.fill(0xFF);
buf.writeIntLE(-0x123400, 0, 3);
assert.deepStrictEqual(buf.toJSON().data, [0x00, 0xcc, 0xed]);
assert.strictEqual(buf.readIntLE(0, 3), -0x123400);
buf.fill(0xFF);
buf.writeIntBE(-0x123400, 0, 3);
assert.deepStrictEqual(buf.toJSON().data, [0xed, 0xcc, 0x00]);
assert.strictEqual(buf.readIntBE(0, 3), -0x123400);
buf.fill(0xFF);
buf.writeIntLE(-0x120000, 0, 3);
assert.deepStrictEqual(buf.toJSON().data, [0x00, 0x00, 0xee]);
assert.strictEqual(buf.readIntLE(0, 3), -0x120000);
buf.fill(0xFF);
buf.writeIntBE(-0x120000, 0, 3);
assert.deepStrictEqual(buf.toJSON().data, [0xee, 0x00, 0x00]);
assert.strictEqual(buf.readIntBE(0, 3), -0x120000);
buf = Buffer.allocUnsafe(5);
buf.writeUIntLE(0x1234567890, 0, 5);
assert.deepStrictEqual(buf.toJSON().data, [0x90, 0x78, 0x56, 0x34, 0x12]);
assert.strictEqual(buf.readUIntLE(0, 5), 0x1234567890);
buf.fill(0xFF);
buf.writeUIntBE(0x1234567890, 0, 5);
assert.deepStrictEqual(buf.toJSON().data, [0x12, 0x34, 0x56, 0x78, 0x90]);
assert.strictEqual(buf.readUIntBE(0, 5), 0x1234567890);
buf.fill(0xFF);
buf.writeIntLE(0x1234567890, 0, 5);
assert.deepStrictEqual(buf.toJSON().data, [0x90, 0x78, 0x56, 0x34, 0x12]);
assert.strictEqual(buf.readIntLE(0, 5), 0x1234567890);
buf.fill(0xFF);
buf.writeIntBE(0x1234567890, 0, 5);
assert.deepStrictEqual(buf.toJSON().data, [0x12, 0x34, 0x56, 0x78, 0x90]);
assert.strictEqual(buf.readIntBE(0, 5), 0x1234567890);
buf.fill(0xFF);
buf.writeIntLE(-0x1234567890, 0, 5);
assert.deepStrictEqual(buf.toJSON().data, [0x70, 0x87, 0xa9, 0xcb, 0xed]);
assert.strictEqual(buf.readIntLE(0, 5), -0x1234567890);
buf.fill(0xFF);
buf.writeIntBE(-0x1234567890, 0, 5);
assert.deepStrictEqual(buf.toJSON().data, [0xed, 0xcb, 0xa9, 0x87, 0x70]);
assert.strictEqual(buf.readIntBE(0, 5), -0x1234567890);
buf.fill(0xFF);
buf.writeIntLE(-0x0012000000, 0, 5);
assert.deepStrictEqual(buf.toJSON().data, [0x00, 0x00, 0x00, 0xee, 0xff]);
assert.strictEqual(buf.readIntLE(0, 5), -0x0012000000);
buf.fill(0xFF);
buf.writeIntBE(-0x0012000000, 0, 5);
assert.deepStrictEqual(buf.toJSON().data, [0xff, 0xee, 0x00, 0x00, 0x00]);
assert.strictEqual(buf.readIntBE(0, 5), -0x0012000000);
}
// Regression test for #5482: should throw but not assert in C++ land.
assert.throws(() => Buffer.from('', 'buffer'), TypeError);
// Regression test for #6111. Constructing a buffer from another buffer
// should a) work, and b) not corrupt the source buffer.
{
let a = [0];
for (let i = 0; i < 7; ++i) a = a.concat(a);
a = a.map((_, i) => {return i;});
const b = Buffer.from(a);
const c = Buffer.from(b);
assert.strictEqual(b.length, a.length);
assert.strictEqual(c.length, a.length);
for (let i = 0, k = a.length; i < k; ++i) {
assert.strictEqual(a[i], i);
assert.strictEqual(b[i], i);
assert.strictEqual(c[i], i);
}
}
assert.throws(() => Buffer.allocUnsafe((-1 >>> 0) + 1), RangeError);
assert.throws(() => Buffer.allocUnsafeSlow((-1 >>> 0) + 1), RangeError);
assert.throws(() => SlowBuffer((-1 >>> 0) + 1), RangeError);
if (common.hasCrypto) {
// Test truncation after decode
const crypto = require('crypto');
const b1 = Buffer.from('YW55=======', 'base64');
const b2 = Buffer.from('YW55', 'base64');
assert.strictEqual(
crypto.createHash('sha1').update(b1).digest('hex'),
crypto.createHash('sha1').update(b2).digest('hex')
);
} else {
common.skip('missing crypto');
}
const ps = Buffer.poolSize;
Buffer.poolSize = 0;
assert(Buffer.allocUnsafe(1).parent instanceof ArrayBuffer);
Buffer.poolSize = ps;
// Test Buffer.copy() segfault
assert.throws(() => Buffer.allocUnsafe(10).copy());
const regErrorMsg = new RegExp('First argument must be a string, Buffer, ' +
'ArrayBuffer, Array, or array-like object.');
assert.throws(() => Buffer.from(), regErrorMsg);
assert.throws(() => Buffer.from(null), regErrorMsg);
// Test prototype getters don't throw
assert.equal(Buffer.prototype.parent, undefined);
assert.equal(Buffer.prototype.offset, undefined);
assert.equal(SlowBuffer.prototype.parent, undefined);
assert.equal(SlowBuffer.prototype.offset, undefined);
{
// Test that large negative Buffer length inputs don't affect the pool offset.
// Use the fromArrayLike() variant here because it's more lenient
// about its input and passes the length directly to allocate().
assert.deepStrictEqual(Buffer.from({ length: -Buffer.poolSize }),
Buffer.from(''));
assert.deepStrictEqual(Buffer.from({ length: -100 }),
Buffer.from(''));
// Check pool offset after that by trying to write string into the pool.
assert.doesNotThrow(() => Buffer.from('abc'));
}
// Test that ParseArrayIndex handles full uint32
assert.throws(() => Buffer.from(new ArrayBuffer(0), -1 >>> 0),
/RangeError: 'offset' is out of bounds/);
// ParseArrayIndex() should reject values that don't fit in a 32 bits size_t.
assert.throws(() => {
const a = Buffer.alloc(1);
const b = Buffer.alloc(1);
a.copy(b, 0, 0x100000000, 0x100000001);
}), /out of range index/;
// Unpooled buffer (replaces SlowBuffer)
{
const ubuf = Buffer.allocUnsafeSlow(10);
assert(ubuf);
assert(ubuf.buffer);
assert.strictEqual(ubuf.buffer.byteLength, 10);
}
// Regression test
assert.doesNotThrow(() => Buffer.from(new ArrayBuffer()));
// Test that ArrayBuffer from a different context is detected correctly
const arrayBuf = vm.runInNewContext('new ArrayBuffer()');
assert.doesNotThrow(() => Buffer.from(arrayBuf));
assert.doesNotThrow(() => Buffer.from({ buffer: arrayBuf }));
assert.throws(() => Buffer.alloc({ valueOf: () => 1 }),
/"size" argument must be a number/);
assert.throws(() => Buffer.alloc({ valueOf: () => -1 }),
/"size" argument must be a number/);
assert.strictEqual(Buffer.prototype.toLocaleString, Buffer.prototype.toString);
{
const buf = Buffer.from('test');
assert.strictEqual(buf.toLocaleString(), buf.toString());
}