1190679 - (webext-oop) [meta] Run WebExtensions out of process

Status: NEW

(WebExtensions :: General, enhancement, P3)

Description

[Bill McCloskey [inactive unless it's an emergency] (:billm)]

Besides setting the remote=true attribute on the browser elements used for extensions, we'll need to move the API-injected code to the other process (via content scripts or something). We'll also need platform support to ensure that all the <browser> elements for a given add-on run in the same process.

Comment 1

[[:fabrice] Fabrice Desré]

Bill, is this bug about running the background pages OOP or more than that?

Comment 2

[Bill McCloskey [inactive unless it's an emergency] (:billm)]

There are a couple different kinds of pages that would need to run out of process:
- background page
- browserAction popup page
- any tabs that are showing moz-extension: URIs

We have to move them all together since they can get references to each other's windows.

Comment 3

[Edgar Chen [:edgar]]

I'd like to work on the extensions; assigning this to myself. :)

Comment 4

[Bill McCloskey [inactive unless it's an emergency] (:billm)]

I think this could be a lot of work, including some platform changes. Do you have a plan for this Edgar? Otherwise we should discuss the approach.

Comment 5

[Edgar Chen [:edgar]]

I believe this is a huge work. And I am still in study stage (trying to get whole picture from bug 1175770). Do you have any suggestion where we should start with? Thank you.

Comment 6

[Bill McCloskey [inactive unless it's an emergency] (:billm)]

Sorry it took me a while to get to this.

The actual extension code will run in a content process. We'll do this using remote <browser> elements (and maybe remote moz-browser elements on b2g). The main process will load a process script into the extension process and the two processes will communicate using the process message manager.

We'll need some platform support to ensure that all these <browser> elements run in the same extension process. That could be an attribute on the <browser> DOM element. The <browser> elements that will need this attribute are for the background page, the browser action, and any page loaded with a moz-extension URI. For the latter, we'll need special handling in E10SUtils.jsm [1] on desktop to ensure that the page loads in the right process. I'm not sure how to handle that on b2g.

The ext-*.js scripts will have to be split into main process scripts and an extension process scripts. They'll communicate using the process message manager. Most API functions will probably just be forwarded to the main process, with their arguments sent using structured clone. However, there are some APIs that pass functions, so we'll have to do something special there.

The webRequest API will also need special handling. Right now it requires the request handlers to be synchronous, which won't be the case when the extension is OOP. I think we'll have to make it suspend the request until the extension makes a decision about whether to block.

I think the best way to get started here is to start using a non-remote browser element for the background page. Right now the code is loaded directly into a windowless docshell. If that works, then we can start moving some of the ext-*.js code to process scripts (as well as some of the related code in Extension.jsm, such as GlobalManager).

[1] http://mxr.mozilla.org/mozilla-central/source/browser/modules/E10SUtils.jsm#62

Comment 7

[[:fabrice] Fabrice Desré]

(In reply to Bill McCloskey (:billm) from comment #6)
> Sorry it took me a while to get to this.
> 
> The actual extension code will run in a content process. We'll do this using
> remote <browser> elements (and maybe remote moz-browser elements on b2g).
> The main process will load a process script into the extension process and
> the two processes will communicate using the process message manager.
> 
> We'll need some platform support to ensure that all these <browser> elements
> run in the same extension process. That could be an attribute on the
> <browser> DOM element. The <browser> elements that will need this attribute
> are for the background page, the browser action, and any page loaded with a
> moz-extension URI. For the latter, we'll need special handling in
> E10SUtils.jsm [1] on desktop to ensure that the page loads in the right
> process. I'm not sure how to handle that on b2g.

On b2g we had a need to run different apps in the same process (for Tarako) and we did that by adding a parentapp attribute on <iframe mozbrowser>. The ContentParent uses this attribute at http://mxr.mozilla.org/mozilla-central/source/dom/ipc/ContentParent.cpp#1264

> I think the best way to get started here is to start using a non-remote
> browser element for the background page. Right now the code is loaded
> directly into a windowless docshell. If that works, then we can start moving
> some of the ext-*.js code to process scripts (as well as some of the related
> code in Extension.jsm, such as GlobalManager).

Right, that's what I'm fixing in bug 1198970 as part of getting webRequest to work.

Comment 8

[Edgar Chen [:edgar]]

Unassigning myself as I am not able to work on this in short term.

Comment 9

[Andy McKay]

After chatting with Bill, we decided to move this to post 48, but before we implement any APIs beyond Chromes implementation.

Comment 10

[Andy McKay]

Taking off webextensions+ and qf:p1. We've got out of process running on Windows which is the main quantum flow target. We'd like to get out of process working on OS X and Linux, but those will likely come after 57.

The bug for turning on out of process on windows was bug 1357486.

Comment 11

[erosman]

I also had a problem with <select> drop down menu showing but not working in sidebar.

Tomislav suggested setting 'extensions.webextensions.remote' to false and after reloading the extension, the drop down menu worked fine in the sidebar.

I am using FF56.0b1

The default 'extensions.webextensions.remote' is true and users will not be expected to change it.

Comment 13

[Firefox Bug Husbandry Bot]

Bulk move of bugs per https://bugzilla.mozilla.org/show_bug.cgi?id=1483958

Comment 14

[ariasuni]

I suppose that bug 1513656 should be considered a regression.

Comment 15

[BugBot [:suhaib / :marco/ :calixte]]

In the process of migrating remaining bugs to the new severity system, the severity for this bug cannot be automatically determined. Please retriage this bug using the new severity system.