TLSSocket.bufferSize is always +1 before and after write()

[micooz]

• Version: v8.4.0
• Platform: Linux 4.4.0-92-generic x86_64 GNU/Linux
• Subsystem: net,tls

reproduce:

const tls = require('tls');
const fs = require('fs');

const key = fs.readFileSync('key.pem');
const cert = fs.readFileSync('cert.pem');

const server = tls.createServer({key, cert}, (socket) => {
 socket.pipe(socket);
 socket.on('end', () => {
 server.close();
 });
});

server.listen(0, () => {
 const client = tls.connect(server.address().port, {rejectUnauthorized: false}, () => {
 console.log(client.bufferSize); // 1

 client.write(Buffer.alloc(10));
 console.log(client.bufferSize); // 11

 client.end();
 });

 client.on('finish', () => {
 console.log(client.bufferSize); // 1
 });
});

bufferSize is obviously wrong in this case, this affects users who rely on this property to make judgments, for example:

if (socket.bufferSize <= 0) {
 // oops! never reach here...
}

why?

node/lib/_tls_wrap.js

Lines 427 to 437 in 3cf27f1

	TLSSocket.prototype._init = function(socket, wrap) {
	var self = this;
	var options = this._tlsOptions;
	var ssl = this._handle;
	// lib/net.js expect this value to be non-zero if write hasn't been flushed
	// immediately
	// TODO(indutny): revise this solution, it might be 1 before handshake and
	// represent real writeQueueSize during regular writes.
	ssl.writeQueueSize = 1;

It seems a legacy code here. After I searched the entire lib, writeQueueSize is only use for calculating bufferSize and decide if should wait until write flushed.

node/lib/net.js

Lines 465 to 471 in 3cf27f1

	Object.defineProperty(Socket.prototype, 'bufferSize', {
	get: function() {
	if (this._handle) {
	return this._handle.writeQueueSize + this._writableState.length;
	}
	}
	});

node/lib/net.js

Lines 768 to 773 in 3cf27f1

	// If it was entirely flushed, we can write some more right now.
	// However, if more is left in the queue, then wait until that clears.
	if (req.async && this._handle.writeQueueSize !== 0)
	req.cb = cb;
	else
	cb();

Simply remove writeQueueSize works fine for me, and I also opened a PR to fix the problem, all local tests passed but I still not very sure whether it's a right approach, so please make a review.