Once all the WebDriver commands work with the JSWindowActor implementation successfully in all the Firefox builds we can go ahead and remove all the code related to the listener.js framescript.

We might still want to wait 1 or 2 releases after bug 1669169 reached mozilla-release to be sure to not regress anything.

This bug shouldn't block the Fission project.

Given that we enabled actors by default (bug 1669169) for Firefox 84, we will keep all the framescript code around for Firefox 85 (to be released on 2021-01-26). That way we have two major releases to bake the actor code, and to fix regressions as reported by users.

With the development on Firefox 86 - its merge day is set to 2020-12-14 - we should then plan to remove all the framescript code.

Tracking marionette-fission-mvp bugs for Fission Beta milestone (M7).

Moving to the reserved backlog given that we cannot finish this bug this year. See comment 1 for details.

Note that we are a bit behind with landing relevant code. As such we may have to delay this bug a bit further. Lets check back early in January what the stability of Marionette tests show.

Just to see how it looks like I pushed a try build yesterday:
https://treeherder.allizom.org/jobs?repo=try&revision=3b0652c62be5390e0c624e9d0f28203aed0c96ce

By doing that work I noticed a couple of things we still have to do before we can get rid of the old code.

Blocked by various dependencies. Moving to P3 for now.

Depends on D100942

Depends on D100943

Depends on D100944

I just pushed the the code as WIP so that we have it publicly available. As mentioned before we should still wait probably until the 86 release before landing those pieces.

Does this marionette-fission-reserve bug need to block shipping Fission MVP?

For now, I am assuming marionette-fission-reserve bugs don't need to block shipping Fission MVP.

As discussed with Julian we will do the required work for bug 1683193 once the old framescript code has been removed, simply to avoid an unnecessary duplicated implementation in an area that gets removed now.

Also bug 1661591 shouldn't be an actual blocker because it's not happening that often, and isn't forcing us to disable the actor support for end-users.

Given that no further bug reports have been filed for both Marionette and geckodriver since actors are enabled and the framescript code not running anymore, I'll go ahead and start the review process for these patches.

There is actually one reported issue at https://github.com/mozilla/geckodriver/issues/1837 that I want to investigate / fix first, before landing these changes.

(In reply to Henrik Skupin (:whimboo) [⌚️UTC+1] from comment #15)

There is actually one reported issue at https://github.com/mozilla/geckodriver/issues/1837 that I want to investigate / fix first, before landing these changes.

This should actually not block. It's a rare situation that should basically not happen. We should only allow a single connection at a time. I filed bug 1691047 for restricting it.

With bug 1690308 another instance of the framescript decorator for the Marionette test had to be added. Once merged I will have to update the 2nd patch here.

Backed out for marionette failures on test_navigation.py

Backout link: https://hg.mozilla.org/integration/autoland/rev/4c29672b9a35eb4d0e7f58a557e86c8066ded069
Log link: https://treeherder.mozilla.org/logviewer?job_id=329902538&repo=autoland&lineNumber=19680

I forgot to rebase and to remove the newly added usage of the framescript decorator.

https://hg.mozilla.org/mozilla-central/rev/b4021ee7f0dc
https://hg.mozilla.org/mozilla-central/rev/1e6088681f84
https://hg.mozilla.org/mozilla-central/rev/ec293f45d826
https://hg.mozilla.org/mozilla-central/rev/791cc5bb75e7