Buffer issues with external two bytes string

[ghostoy]

Node Buffer doesn't work correctly with two bytes external strings. Once passing such string to Buffer constructor, the content of the Buffer is just first half of the string in UCS2 encoding.

Steps to reproduce:

1. Run node with --expose_externalize_string flag
2. Run following code

var str = 'ABCD';
var buf = new Buffer(str);
console.log(buf.toString('hex')); // 41424344
externalizeString(str, true); // force two bytes string
var extbuf = new Buffer(str);
console.log(extbuf.toString('hex')); // 41004200

As documented for new Buffer(str, [encoding]):

new Buffer(str, [encoding])

• str String - string to encode.
• encoding String - encoding to use, Optional.

Allocates a new buffer containing the given str. encoding defaults to 'utf8'.

Therefore the expected content of extbuf should also be encoded in UTF8 (i.e. 41424344 in this case) for any JS string by default.

This issue should be the root cause of an node-webkit issue nwjs/nw.js#2439

[vkurchatkin]

Interesting variation:

var str = 'ABCD';
var buf = new Buffer(str);
console.log(buf.toString('hex')); // 41424344
externalizeString(str, true); // force two bytes string
var extbuf = new Buffer(str);
console.log(extbuf.toString('hex')); // 41004200

var buf = new Buffer('ABCD');
console.log(buf.toString('hex')); // 41004200

Although second 'ABCD' was not explicitly externalized, it still is. Only works for same string literals, so probably this is an effect of constant pool in action

[vkurchatkin]

This line causes the error : https://github.com/joyent/node/blob/master/src/string_bytes.cc#L331 Probably memcpy should be used only for one byte external strings

[ghostoy]

Also this line should return doubled length: https://github.com/joyent/node/blob/master/src/string_bytes.cc#L285