Multiprocess Firefox and the SDK

This article explains how developers can test whether SDK-based add-ons are compatible with multiprocess Firefox, and how to fix them if they are not.

The project to make Firefox multiprocess is also known as "electrolysis" or "e10s". For much more information, see the multiprocess Firefox pages.

Firefox is moving to a multiprocess architecture in which it uses one operating system process for the browser UI (also known as the chrome process), and another process to execute code running in web pages (also known as the content process).

The main compatibility consequence of this is that code running in the chrome process will no longer get direct access to web content.

Because SDK add-ons run in the chrome process (except for their content scripts):

The SDK contract

The SDK makes this promise to add-on developers:

High-level APIs will just work in multiprocess Firefox. If they don't, it's an SDK bug.
Low-level APIs might not work. If you're using low-level APIs, review your use of them, test, and potentially refactor.

In practice most low-level APIs will work fine, but low-level APIs that give direct access to web content will not work.

Testing your add-on

To ease the transition for add-on developers, we've provided add-ons with compatibility shims for many APIs. With these shims in place, code that you would not expect to work in multiprocess Firefox - for example, code that directly accesses web content - will still work.

These shims use wrappers called Cross Process Object Wrappers, to make it look as if code in one process can directly access objects in a different process.

However, these shims are just a temporary migration aid, not a permanent solution. You should aim to make your add-on work without the shims.

Also, the shims don't fix all possible problems, so even a shimmed add-on may not work with multiprocess Firefox.

So the process for making your add-on multiprocess compatible is:

get a multiprocess version of Firefox
test it with the shims enabled, and fix any problems
test it with the shims disabled, and fix any problems.

Getting multiprocess Firefox

First, download and install Firefox Nightly or Firefox Developer Edition. Both these versions are multiprocess by default. You can enable/disable multiprocess Firefox in "about:preferences":

You can double-check that you're running multiprocess Firefox by hovering over the name of a tab: multiprocess Firefox will say "- e10s" after the tab title:

Testing with shims enabled

By default, your add-on is given the compatibility shims. So if you just test your add-on on multiprocess Firefox, you're testing it with the shims active.

You might still see some problems: the shims don't cover everything. You might also see bad performance, as the shims can be bad for responsiveness.

Testing with shims disabled

To test whether an add-on works without the shims, include the "multiprocess" permission in your package.json:

"permissions": {
    "multiprocess": true
}

Setting this permission will disable the shims for your add-on, so you can test to see if it's really multiprocess compatible or not.

Common problems and solutions

Direct access to web content

The most common problem is if you are directly accessing web content, using a low-level API like window/utils. For example:

var contentDocument = require("sdk/window/utils")
  .getMostRecentBrowserWindow().content.document;

If all you need to do is access and modify web pages, the best alternative here is just to do it in a content script, loaded using page-mod or tabs.attach().

Usage of certain XPCOM APIs

More subtly, if you're using any low-level APIs, and especially if you are using require("chrome") to get access to underlying XPCOM APIs, then it's definitely worth reviewing the Limitations of chrome scripts document. This lists APIs and patterns that used to work in the chrome process, that will no longer work. The general solution to this class of problem is to use the API in the content process.

For example: some add-ons, like ad-blockers, want to use nsIContentPolicy to register a content policy. In multiprocess Firefox, if you register the nsIContentPolicy in the chrome process then it will never see any attempts to load web content, because they happen in the content process. Instead, you must register the content policy in the content process.

For an SDK add-on, the best way to do this is to use the SDK's remote/parent module to load modules into the content process. For an example of how to do this, see Using nsIContentPolicy with remote/parent.

SDK internal incompatibilities

Unfortunately, the SDK itself isn't completely multiprocess-compatible. This means that if you disable shims for your add-on, then it might not work, even if you are only using high-level APIs. For example:

var selection = require("sdk/selection");

function myListener() {
  console.log(selection.text);
}

selection.on('select', myListener);

This add-on will not work if you've set the "multiprocess" permission, because sdk/selection depends on the shims. If this is the case for you, there's nothing you can do to fix it apart from using a different API.

We're working on fixing these problems: see bug 1004745 and its dependencies.

Examples

Using nsIContentPolicy with "remote/parent"

This example shows how to use the "remote/parent" module to register a content policy in the content process.

The add-on consists of three files:

package.json
index.js
content-policy.js

Original version

In the original version, the "package.json" file can just be the default file generated by jpm init.

The index.js looks like this:

require("sdk/preferences/service").set("[email protected]", "all");
console.log("running");

var policy = require('./content-policy');
policy.init();

This just makes sure logging is enabled for the add-on (we would need to remove this in production, of course), then loads the "content-policy" module, then calls its init() export.

Most of the work is in "content-policy.js":

var {components, Cu, Cc, Ci} = require("chrome");
var unload = require("sdk/system/unload");

Cu.import("resource://gre/modules/XPCOMUtils.jsm");
Cu.import("resource://gre/modules/Services.jsm");

let policy =
{
  classDescription: "Test content policy",
  classID: components.ID("{12345678-1234-1234-1234-123456789abc}"),
  contractID: "@adblockplus.org/test-policy;1",
  xpcom_categories: ["content-policy"],

  init: function()
  {
    let registrar = components.manager.QueryInterface(Ci.nsIComponentRegistrar);
    registrar.registerFactory(this.classID, this.classDescription, this.contractID, this);

    let catMan = Cc["@mozilla.org/categorymanager;1"].getService(Ci.nsICategoryManager);
    for each (let category in this.xpcom_categories)
      catMan.addCategoryEntry(category, this.contractID, this.contractID, false, true);

    unload.when(
    (function()
    {
      for each (let category in this.xpcom_categories)
        catMan.deleteCategoryEntry(category, this.contractID, false);

      // This needs to run asynchronously, see bug 753687
      Services.tm.currentThread.dispatch(function()
      {
        registrar.unregisterFactory(this.classID, this);
      }.bind(this), Ci.nsIEventTarget.DISPATCH_NORMAL);
    }).bind(this));
  },

  // nsIContentPolicy interface implementation
  shouldLoad: function(contentType, contentLocation, requestOrigin, node, mimeTypeGuess, extra)
  {
    if (contentLocation.schemeIs("http") || contentLocation.schemeIs("https")) {
      console.log("shouldLoad: " + contentType + " " +
                            (contentLocation ? contentLocation.spec : "null") + " " +
                            (requestOrigin ? requestOrigin.spec : "null") + " " +
                            node + " " +
                            mimeTypeGuess + "\n");
    }
    return Ci.nsIContentPolicy.ACCEPT;
  },

  shouldProcess: function(contentType, contentLocation, requestOrigin, node, mimeTypeGuess, extra)
  {
    return Ci.nsIContentPolicy.ACCEPT;
  },

  // nsIFactory interface implementation
  createInstance: function(outer, iid)
  {
    if (outer)
      throw Cr.NS_ERROR_NO_AGGREGATION;
    return this.QueryInterface(iid);
  },

  // nsISupports interface implementation
  QueryInterface: XPCOMUtils.generateQI([Ci.nsIContentPolicy, Ci.nsIFactory])
};

exports.init = function() {
  policy.init();
}

This is a slightly adapted version of the example policy written by Wladimir Palant in a stackoverflow answer. The policy just listens for attempts to load content, then if they are using an "http" or "https" scheme, logs some information about them.

You can build this into an add-on, and it will work fine in singleprocess Firefox. In the Browser Console, you'll see messages being logged as you load pages.

Because we haven't set the "multiprocess" permission in package.json, it will work fine in multiprocess Firefox as well, because nsIContentPolicy is shimmed.

Ported version

First, set the "multiprocess" permission by adding a key like this in package.json:

  "permissions": {
    "multiprocess": true
  }

Now try running the add-on in multiprocess Firefox. You should no longer see any output in the Browser Console, because you registered the content policy in the chrome process, but the loads are happening in the content process.

So let's make it work again. The new index.js should look like this:

require("sdk/preferences/service").set("[email protected]", "all");
console.log("running");

const { processes, remoteRequire } = require("sdk/remote/parent");
remoteRequire("./content-policy", module);

// For every current and future process
processes.forEvery(process => {
  process.port.emit("init");
});

The first two lines are the same as the old version.

After that, we're:

importing processes and remoteRequire() from "sdk/remote/parent"
loading "content-policy.js" into the content process, using remoteRequire()
sending an "init" message to "content-policy.js", using the processes object.

The new "content-policy.js" looks like this:

var {components, Cu, Cc, Ci} = require("chrome");
var unload = require("sdk/system/unload");

Cu.import("resource://gre/modules/XPCOMUtils.jsm");
Cu.import("resource://gre/modules/Services.jsm");

let policy =
{
  classDescription: "Test content policy",
  classID: components.ID("{12345678-1234-1234-1234-123456789abc}"),
  contractID: "@adblockplus.org/test-policy;1",
  xpcom_categories: ["content-policy"],

  init: function()
  {
    let registrar = components.manager.QueryInterface(Ci.nsIComponentRegistrar);
    registrar.registerFactory(this.classID, this.classDescription, this.contractID, this);

    let catMan = Cc["@mozilla.org/categorymanager;1"].getService(Ci.nsICategoryManager);
    for each (let category in this.xpcom_categories)
      catMan.addCategoryEntry(category, this.contractID, this.contractID, false, true);

    unload.when(
    (function()
    {
      for each (let category in this.xpcom_categories)
        catMan.deleteCategoryEntry(category, this.contractID, false);

      // This needs to run asynchronously, see bug 753687
      Services.tm.currentThread.dispatch(function()
      {
        registrar.unregisterFactory(this.classID, this);
      }.bind(this), Ci.nsIEventTarget.DISPATCH_NORMAL);
    }).bind(this));
  },

  // nsIContentPolicy interface implementation
  shouldLoad: function(contentType, contentLocation, requestOrigin, node, mimeTypeGuess, extra)
  {
    if (contentLocation.schemeIs("http") || contentLocation.schemeIs("https")) {
      console.log("shouldLoad: " + contentType + " " +
                            (contentLocation ? contentLocation.spec : "null") + " " +
                            (requestOrigin ? requestOrigin.spec : "null") + " " +
                            node + " " +
                            mimeTypeGuess + "\n");
    }
    return Ci.nsIContentPolicy.ACCEPT;
  },

  shouldProcess: function(contentType, contentLocation, requestOrigin, node, mimeTypeGuess, extra)
  {
    return Ci.nsIContentPolicy.ACCEPT;
  },

  // nsIFactory interface implementation
  createInstance: function(outer, iid)
  {
    if (outer)
      throw Cr.NS_ERROR_NO_AGGREGATION;
    return this.QueryInterface(iid);
  },

  // nsISupports interface implementation
  QueryInterface: XPCOMUtils.generateQI([Ci.nsIContentPolicy, Ci.nsIFactory])
};

const { process } = require("sdk/remote/child");

process.port.on("init", () => {
  policy.init();
});

This is exactly the same as the last version, except for the last few lines, where we:

import the process object from "sdk/remote/child"
use process to listen for the "init" message from the other side, and call our policy.init() function when that happens.

That's it. Now we can run the add-on in multiprocess Firefox, and it will work without needing any shims.