1211401 - Implement chrome.runtime.sendMessage on b2g

Status: RESOLVED FIXED

(WebExtensions :: Untriaged, defect, P3)

Comment 1

[Dale Harvey (:daleharvey)]

This is required for the password manager on Firefox OS, a content_script embedded in arbitrary web content needs the ability to send a message and receive a result from the main password manager process

Comment 2

[Dale Harvey (:daleharvey)]

https://developer.chrome.com/extensions/runtime#method-sendMessage is the chrome documentation for this, note that chrome will only allow sendMessage from a content_script if the origin has been explicitly whitelisted by the addon, for the password manager however we need this ability from arbitrary origins

Comment 3

[Dale Harvey (:daleharvey)]

Hey Fabrice, so described what I need here and will start trying to implement it. 

I would really prefer to be able to prototype the implementation for desktop and then make it work on b2g, is that possible? and any pointers for implementing this api (as oposed to the tabs one)

Cheers
Dale

Comment 4

[[:fabrice] Fabrice Desré]

That seems to be implemented in https://mxr.mozilla.org/mozilla-central/source/toolkit/components/extensions/ext-runtime.js#25

Can you check on desktop? If that works fine we'll investigate why it's failing on b2g.

Comment 5

[Dale Harvey (:daleharvey)]

Ok so I wasted a good few weeks expecting this to not exist, was pretty dumb of me to not look into it further, apologies. Initial testing on desktop looks great, going to flesh out the demo on desktop and will then test it on b2g and see what the status is there.

Comment 6

[Dale Harvey (:daleharvey)]

So done my initial testing, its working well on desktop, ran into one major issue @ https://bugzilla.mozilla.org/show_bug.cgi?id=1212743

Now testing on b2g

Comment 7

[Dale Harvey (:daleharvey)]

So first problem

E/GeckoConsole( 4122): [JavaScript Error: "TypeError: chrome.runtime is undefined" {file: "moz-extension://1de1b431-dc9b-42f4-a50d-bafd0b976ab1/js/PasswordManager.js" line: 15}]

https://github.com/daleharvey/fxospwdmgr/blob/master/js/PasswordManager.js#L15

Going to look @ https://bugzilla.mozilla.org/show_bug.cgi?id=1199504 to see if I can figure out how to enable it, but Fabrice if you have an pointers, cheers

Comment 8

[Dale Harvey (:daleharvey)]

Quick update, with the patch @ https://bugzilla.mozilla.org/show_bug.cgi?id=1199504 applied chrome.runtime exists, but sendMessage is not getting sent. I should hopefully be able to debug it but right now being asked to spend time on blockers

Comment 9

[Dale Harvey (:daleharvey)]

This is blocked and currently working on 2.5 blockers, if anyone else was able to get it going would be hugely appreciated

Comment 10

[Adrian Aichner [:anaran]]

I submitted
https://github.com/mdn/webextensions-examples/pull/12
to webextensions-examples
today which I wrote to demo cs to bg messaging (works fine in Nightly Firefox on Windows XP and Linux).

The initial UI shows up in Firefox OS as well (installed via WebIDE), but no message passing takes place, as evidenced by this bug.

Trying to sends messages from bg to cs (by Disable/Enable Add-on) reports
chrome.tabs is undefined in the bg script.

Comment 11

[[:fabrice] Fabrice Desré]

That's expected as the chrome.tabs API is not implemented yet on b2g.

Comment 12

[[:fabrice] Fabrice Desré]

Trying to enable more tests I looked into that a bit. Here's some debug log:

1878 is the parent process, 1955 the child process running the mochitest.
The child is sending `Extension:Message` but the parent never receives it in the Broker. I don't know why since the ChromeMessageBroadcaster is listening to this message... Bill, how is that different from the desktop e10s setup?

I/Gecko   ( 1878): ZZZ MessageBroker listening to Extension:Message on [object ChromeMessageBroadcaster]
I/Gecko   ( 1878): ZZZ MessageBroker listening to Extension:Connect on [object ChromeMessageBroadcaster]
I/Gecko   ( 1878): ZZZ MessageBroker listening to Extension:Message on [object ChromeMessageBroadcaster]
I/Gecko   ( 1878): ZZZ MessageBroker listening to Extension:Connect on [object ChromeMessageBroadcaster]
I/Gecko   ( 1878): ZZZ MessageBroker listening to Extension:Message on [object ContentFrameMessageManager]
I/Gecko   ( 1878): ZZZ MessageBroker listening to Extension:Connect on [object ContentFrameMessageManager]
I/Gecko   ( 1955): ZZZ MessageBroker listening to Extension:Message on [object ContentFrameMessageManager]
I/Gecko   ( 1955): ZZZ MessageBroker listening to Extension:Connect on [object ContentFrameMessageManager]
I/Gecko   ( 1878): ZZZ Creating Messenger {"id":"{0bab3193-1971-40a4-b3f6-1c7b5f232661}","url":"moz-extension://6ddd8b4e-6ac6-443b-8d1d-8185b01a3bde/_blank.html"}
I/Gecko   ( 1878): ZZZ Messenger.onMessage runtime.onMessage {"id":"{0bab3193-1971-40a4-b3f6-1c7b5f232661}","url":"moz-extension://6ddd8b4e-6ac6-443b-8d1d-8185b01a3bde/_blank.html"}
I/Gecko   ( 1878): ZZZ Messenger.onMessage runtime.onMessage {"id":"{0bab3193-1971-40a4-b3f6-1c7b5f232661}","url":"moz-extension://6ddd8b4e-6ac6-443b-8d1d-8185b01a3bde/_blank.html"}
I/Gecko   ( 1878): ZZZ SingletonEventManager callback
I/Gecko   ( 1878): ZZZ adding broker message listener, filter is {"id":"{0bab3193-1971-40a4-b3f6-1c7b5f232661}"}
I/Gecko   ( 1878): ZZZ MessageBroker.addListener message filter is {"id":"{0bab3193-1971-40a4-b3f6-1c7b5f232661}"}
I/Gecko   ( 1955): ZZZ MessageBroker listening to Extension:Message on [object ContentFrameMessageManager]
I/Gecko   ( 1955): ZZZ MessageBroker listening to Extension:Connect on [object ContentFrameMessageManager]
I/Gecko   ( 1955): ZZZ Creating Messenger {"id":"{0bab3193-1971-40a4-b3f6-1c7b5f232661}","frameId":0,"url":"http://mochi.test:8888/tests/toolkit/components/extensions/test/mochitest/file_sample.html"}
I/Gecko   ( 1955): ZZZ Messenger.onMessage runtime.onMessage {"id":"{0bab3193-1971-40a4-b3f6-1c7b5f232661}","frameId":0,"url":"http://mochi.test:8888/tests/toolkit/components/extensions/test/mochitest/file_sample.html"}
I/Gecko   ( 1955): ZZZ Messenger.sendMessage {"id":"{0bab3193-1971-40a4-b3f6-1c7b5f232661}","frameId":0,"url":"http://mochi.test:8888/tests/toolkit/components/extensions/test/mochitest/file_sample.html"} ["moz-extension://6ddd8b4e-6ac6-443b-8d1d-8185b01a3bde/test_file.html","moz-extension://6ddd8b4e-6ac6-443b-8d1d-8185b01a3bde/test_file.html"] undefined
I/Gecko   ( 1955): ZZZ broker.sendMessage [object ContentFrameMessageManager] Extension:Message ["moz-extension://6ddd8b4e-6ac6-443b-8d1d-8185b01a3bde/test_file.html","moz-extension://6ddd8b4e-6ac6-443b-8d1d-8185b01a3bde/test_file.html"] to {"extensionId":"{0bab3193-1971-40a4-b3f6-1c7b5f232661}","messageId":1}

Comment 13

[[:fabrice] Fabrice Desré]

Replacing https://mxr.mozilla.org/mozilla-central/source/toolkit/components/extensions/ExtensionContent.jsm#85 by Service.cpmm (like in ext-runtime.js) sends the message properly to the broker in the parent, but we then fail at ExtensionUtils.jsm#557 because |this.delegate| is an empty object. I'm not sure why yet.

Comment 14

[[:fabrice] Fabrice Desré]

Attachment: runtime-sendmessage.patch

That makes things work for me on b2g (and eg. test_ext_geturl.html passes), but I haven't push to try yet.

Comment 15

[[:fabrice] Fabrice Desré]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=8d0dcebaee90

Comment 16

[[:fabrice] Fabrice Desré]

Looks like this broke test_ext_sendmessage_doublereply.html and test_ext_sendmessage_reply.html

Comment 17

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

Can you see what getWindowMessageManager is returning in this case?
http://mxr.mozilla.org/mozilla-central/source/toolkit/components/extensions/ExtensionContent.jsm#208
http://mxr.mozilla.org/mozilla-central/source/toolkit/components/extensions/ExtensionContent.jsm#178

Comment 18

[[:fabrice] Fabrice Desré]

(In reply to Bill McCloskey (:billm) from comment #17)
> Can you see what getWindowMessageManager is returning in this case?
> http://mxr.mozilla.org/mozilla-central/source/toolkit/components/extensions/
> ExtensionContent.jsm#208
> http://mxr.mozilla.org/mozilla-central/source/toolkit/components/extensions/
> ExtensionContent.jsm#178

We get a ContentFrameMessageManager

Comment 19

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

It looks like the global message manager just doesn't work in b2g. The parent of the TabParent's message manager is null. The parent gets set here:

https://dxr.mozilla.org/mozilla-central/source/dom/base/nsFrameLoader.cpp#2579

So if there's no chrome window (and I assume there isn't in b2g) then the parent will be null.

Olli, do you think it would be okay to make every frame message manager be a direct child of the global message manager on b2g? I don't see what harm it could do since no one seems to be using the global message manager there.

Comment 20

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

If I use the globalmessagemanager as the parent, there are still a couple other bugs to be fixed, but the test passes.

Comment 21

[Olli Pettay [:smaug][[e-mail]]]

mozbrowser's message manager, when mozbrowser is in content (not in chrome), doesn't on purpose have parent. (I would need to look at the blame to find some reasoning, but at least if was the safest approach when mozbrowser was implemented).


If we can ensure global message manager isn't used currently, making mozbrowser's mm child of that 
sounds ok to me (in case mozbrowser isn't in chrome window). Another special case, but mozbrowser 
has been special anyway.

Comment 22

[Olli Pettay [:smaug][[e-mail]]]

But, question, why isn't process message manager used here?

Comment 23

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

On desktop, at least, we need to know which tab sent the message. Eventually we'll need that for b2g too.

Comment 24

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

Hi Kershaw, we're trying to make a change in this bug where mozbrowser message managers are children of the global message manager. This causes some test failures:
https://treeherder.mozilla.org/#/jobs?repo=try&revision=5d647709f169

Fabrice looked at them and they seem to be caused by special powers message listeners running multiple times. Do you think this could be related to bug 1038620? I'm having trouble understanding when the nested setup is used during b2g tests.

Comment 25

[Kershaw Chang [:kershaw]]

(In reply to Bill McCloskey (:billm) from comment #24)
> Hi Kershaw, we're trying to make a change in this bug where mozbrowser
> message managers are children of the global message manager. This causes
> some test failures:
> https://treeherder.mozilla.org/#/jobs?repo=try&revision=5d647709f169
> 
> Fabrice looked at them and they seem to be caused by special powers message
> listeners running multiple times. Do you think this could be related to bug
> 1038620? I'm having trouble understanding when the nested setup is used
> during b2g tests.

If I am not wrong, nested oop is not used during b2g tests currently. So, I think this is not related to bug 1038620.

Comment 26

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

Attachment: mm-fix

This makes <iframe mozbrowser>s use the global message manager as their parent, as discussed above.

Comment 27

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

Attachment: test-fix

Doing so allows us to clean up the test harness a little bit. Before we needed a special case for b2g so that we could pass in the <iframe> message manager. Now we can always use the global message manager.

The try results look pretty good:
https://treeherder.mozilla.org/#/jobs?repo=try&revision=a8fc2e6c5ba6
Fabrice, does that xpcshell stuff look okay to you?

Comment 28

[[:fabrice] Fabrice Desré]

Yes the testing part looks fine to me. I'm checking what I can enable on top of that in bug 1224703

Comment 29

[Andrew Halberstadt [:ahal]]

Comment on attachment 8696184 [details] [diff] [review]
test-fix

Review of attachment 8696184 [details] [diff] [review]:
-----------------------------------------------------------------

This lgtm assuming b2g mochitest was the only thing passing in a non-global message manager. Might be good to include b2g reftests in the try run just in case.

Comment 30

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/a22303d0c043

Comment 31

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/a22303d0c043