XMPP implementation in JavaScript.

remoting/webapp/xmpp_stream_parser.js

+// Copyright 2014 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+'use strict';
+
+/** @suppress {duplicate} */
+var remoting = remoting || {};
+
+/**
+ * XmppStreamParser is used to split data received over XMPP connection into
+ * individual stanzas. Data is fed to the parser using appendData() method.
+ * onStartCallback is called once for the stream header and after that

[Jamie, 2014/08/29 02:14:09]

There is no onStartCallback.

[Sergey Ulanov, 2014/08/29 23:40:31]

Done.

+ * onStanzaCallback is called for each complete stanza.
+ *
+ * @param {function(Element):void} onStanzaCallback
+ * @param {function(string):void} onErrorCallback
+ * @constructor
+ */
+remoting.XmppStreamParser = function(onStanzaCallback, onErrorCallback) {
+  this.onStanzaCallback_ = onStanzaCallback;
+  this.onErrorCallback_ = onErrorCallback;
+
+
+  /** @type {Array.<ArrayBuffer>} */

[Jamie, 2014/08/29 02:14:09]

Please add descriptions of these members.

[Sergey Ulanov, 2014/08/29 23:40:31]

Done.

+  this.data_ = [];
+  this.depth_ = 0;
+  this.error_ = false;
+  this.startTag_ = '';
+  this.startTagEnd_ = '';
+  this.currentStanza_ = '';
+}
+
+/** @param {ArrayBuffer} data */
+remoting.XmppStreamParser.prototype.appendData = function(data) {
+  base.debug.assert(!this.error_);
+
+  this.data_.push(data);
+
+  // Check if the newly appended data completes XML tag or a piece of text by
+  // looking for '<' and '>' char codes. This has to be done before converting
+  // data to string because the input may not contain complete UTF-8 sequence.
+  var tagStartCode = '<'.charCodeAt(0);
+  var tagEndCode = '>'.charCodeAt(0);
+  var spaceCode = ' '.charCodeAt(0);
+  var tryAgain = true;
+  while (this.data_.length > 0 && tryAgain && !this.error_) {
+    tryAgain = false;
+
+    // If we are not currently in a middle of a stanza then skip spaces (server
+    // may send spaces periodically as heartbeats) and make sure that the first
+    // character starts XML tag.
+    if (this.depth_ <= 1) {

[kelvinp, 2014/08/29 01:37:10]

base.debug.assert(this.depth_ > 0);

[Sergey Ulanov, 2014/08/29 23:40:30]

depth_ can be 0 here.

+      var view = new DataView(this.data_[0]);
+      var firstChar = view.getUint8(0);
+      if (firstChar == spaceCode) {
+        tryAgain = true;
+        this.extractToken_(1);

[Jamie, 2014/08/29 02:14:09]

This doesn't look right. It seems to be testing whether or not the first
character of the first buffer is a space, and if so extracting everything up to
the first character of the *last* buffer. If the assumption is that there is
only one buffer in this case, you should assert that.

Could this whole tryAgain loop be replaced with a loop that strips leading
whitespace, or is it important to error out if there's more than one leading
space?

[Sergey Ulanov, 2014/08/29 23:40:31]

That's for catching this. I've simplified this code now to use a single
ArrayBuffer for pending data. That was useless optimization.
The while loop is necessary to handle the case when we receive a new packet with
multiple XMPP tokens. The inner loop finds end of the first tag and extracts it
from |data_|.

+        continue;
+      } else if (firstChar != tagStartCode) {
+        this.processError_('Received unexpected text data: ' +
+                           base.decodeUtf8(this.data_[0]));

[Jamie, 2014/08/29 02:14:10]

Your earlier comment stated that this might be incomplete UTF-8. We should add a
unit test for decodeUtf8 to ensure that it does something sensible in this case
(decoding only complete characters, for example).

[Sergey Ulanov, 2014/08/29 23:40:31]

Good point. decodeUtf8() throws in that case. I changed this code to handle this
case properly.

+        return;
+      }
+    }
+
+    var view = new DataView(this.data_[this.data_.length - 1]);

[Jamie, 2014/08/29 02:14:10]

It's not obvious what this second loop is doing. I think a comment would be
helpful.

[Sergey Ulanov, 2014/08/29 23:40:31]

Done.

+    for (var i = 0; i < view.byteLength; ++i) {
+      var currentChar = view.getUint8(i);
+      if (currentChar == tagStartCode) {
+        if (this.data_.length > 1 || i > 0) {
+          this.processText_(this.extractToken_(i));
+          tryAgain = true;
+          break;
+        }
+      } else if (currentChar == tagEndCode) {
+        var tag = this.extractToken_(i + 1);
+        if (tag.charAt(0) != '<') {
+          this.processError_('Received \'>\' without \'<\': ' + tag);
+          return;
+        }
+        this.processTag_(tag);
+        tryAgain = true;
+        break;
+      }
+    }
+  }
+}
+
+/** @param {string} text */
+remoting.XmppStreamParser.prototype.processText_ = function(text) {
+  // Tokenization code in appendData() shouldn't allow text tokens in between
+  // stanzas.
+  base.debug.assert(this.depth_ > 1);
+  this.currentStanza_ += text;
+}
+
+/** @param {string} tag */
+remoting.XmppStreamParser.prototype.processTag_ = function(tag) {
+  base.debug.assert(tag.charAt(0) == '<');
+  base.debug.assert(tag.charAt(tag.length - 1) == '>');
+
+  this.currentStanza_ += tag;
+
+  var openTag = tag.charAt(1) != '/';
+  if (openTag) {
+    ++this.depth_;
+    if (this.depth_ == 1) {
+      this.startTag_ = this.currentStanza_;
+      this.currentStanza_ = '';
+
+      // Create end tag matching the start.
+      var tagName =
+          this.startTag_.substr(1, this.startTag_.length - 2).split(' ', 1)[0];
+      this.startTagEnd_ = '</' + tagName + '>';
+
+      // Try parsing start together with the end
+      var parsed = this.parseTag_(this.startTag_ + this.startTagEnd_);
+      if (!parsed) {
+        this.processError_('Failed to parse start tag: ' + this.startTag_);
+        return;
+      }
+    }
+  }
+
+  var closingTag =
+      (tag.charAt(1) == '/') || (tag.charAt(tag.length - 2) == '/');
+  if (closingTag) {
+    // The first start tag is not expected to be closed.
+    if (this.depth_ <= 1) {
+      this.processError_('Unexpected closing tag: ' + tag)
+      return;
+    }
+    --this.depth_;
+    if (this.depth_ == 1) {
+      this.processCompleteStanza_();
+      this.currentStanza_ = '';
+    }
+  }
+}
+
+remoting.XmppStreamParser.prototype.processCompleteStanza_ = function() {
+  var stanza = this.startTag_ + this.currentStanza_ + this.startTagEnd_;
+  var parsed = this.parseTag_(stanza);
+  if (!parsed) {
+    this.processError_('Failed to parse stanza: ' + this.currentStanza_);
+    return;
+  }
+  this.onStanzaCallback_(parsed.childNodes[0]);

[kelvinp, 2014/08/29 01:37:10]

childNodes also contains all nodes, including text nodes consisting entirely of
whitespace (in theory it should not happen as you are removing empty spaces in
the code), would it be safer to just use firstElementChild?  Same for parseTag_.

[Sergey Ulanov, 2014/08/29 23:40:31]

Done.

+}
+
+/** @param {string} text */
+remoting.XmppStreamParser.prototype.processError_ = function(text) {
+  this.onErrorCallback_(text);
+  this.error_ = true;
+}
+
+/**
+ * Helper that assembles block of data from pieces in this.data_ up to specified
+ * position in the last buffer and then removes that data from this.data_ .
+ *
+ * @param {number} endPos Specifies how many bytes should be taken from the last
+ *   buffer in this.data_ .
+ * @returns {string}
+ */
+remoting.XmppStreamParser.prototype.extractToken_ = function(endPos) {

[Jamie, 2014/08/29 02:14:09]

What is a "token" in this context? This seems to be extracting bytes up to a
certain limit and then returning the corresponding string representation;
there's no tokenization going on, so I think the name is misleading. I suggest
calling it something like extractStringUpTo.

[Sergey Ulanov, 2014/08/29 23:40:31]

Renamed extractStringFromBuffer_.

+  var size = endPos;
+  for (var i = 0; i < this.data_.length - 1; ++i) {
+    size += this.data_[i].byteLength;
+  }
+
+  var buffer = new Uint8Array(size);
+  var pos = 0;
+  for (var i = 0; i < this.data_.length - 1; ++i) {
+    buffer.set(new Uint8Array(this.data_[i]), pos);
+    pos += this.data_[i].byteLength;
+  }
+  if (endPos > 0) {
+    buffer.set(new Uint8Array(this.data_[this.data_.length - 1], 0, endPos),
+               pos);
+    pos += endPos;
+  }
+  base.debug.assert(pos == size);
+
+  // Remove copied data from |data_|.
+  var dataLeft = this.data_[this.data_.length - 1].slice(endPos);

[Jamie, 2014/08/29 02:14:10]

I suggest calling this dataRemaining, since it's arguably at the right-most end
of the array, not the left :)

[Sergey Ulanov, 2014/08/29 23:40:31]

Done, though some people would say that arrays start on the right and end on the
left :)

+  if (dataLeft.byteLength == 0) {
+    this.data_ = []
+  } else {
+    this.data_ = [dataLeft];
+  }
+
+  return base.decodeUtf8(buffer.buffer);
+}
+
+/**
+ * @param {string} text
+ * @return {Element}
+ */
+remoting.XmppStreamParser.prototype.parseTag_ = function(text) {
+  /** @type {Document} */
+  var result = new DOMParser().parseFromString(text, 'text/xml');
+  if (result.querySelector('parsererror') != null)
+    return null;
+  return result.firstChild;
+}