Jim recommends that we not remove windowed mode support until we remove all Flash plugin support (bug 1455897).

What's the rationale for that?

(In reply to Jeff Muizelaar [:jrmuizel] from comment #11)

Jim recommends that we not remove windowed mode support until we remove all Flash plugin support (bug 1455897).

What's the rationale for that?

Jim?

(In reply to Jeff Muizelaar [:jrmuizel] from comment #11)

What's the rationale for that?

People still fall back on windowed mode in a small number of cases. I've actually been trying to find a good test case because I'm concerned about webrender shipping. But so far I haven't found one. I think it has something to do with wmode direct + blocked graphics drivers and flash acceleration settings.

On average about 3% of flash instantiations fall back.

Alice, curious, have you ever experienced the old windowed mode behavior with webrender?

This game http://6bob.net/play/Angry-Birds-Space,
Click on 'PLAY GAME'
Click on 'Skip Ad'
Click on 'Run Adobe Flash and then click 'Allow'

This page http://weathernews.jp/

Seems no problem with WebRender.

And this test page https://www.baseonmars.co.uk/bugs/wmode/
On wmode='window', Japanese IME candidate window flashes once and pops up at incorrect place. But this is also reproduced without Webrender.

I believe I see whats happening here. The gist is that NPDrawingModelSyncWin, which is what the unexpected telemetry refers to, does not actually use windowed plugin mode. It promotes to direct with async rendering, as suggested here [1]. If you set dom.ipc.plugins.asyncdrawing.enabled to false, this mode is the result.

I've confirmed with the debugger and some local experimentation that NPDrawingModelSyncWin in Flash does not result in NPP_SetWindow calls. At this point, I agree that trying to remove the dead plugin code is not worth the effort. All of it is on the way out. I'm pro-WONTFIX.

[1] https://wiki.mozilla.org/Plugins/Async_Drawing#Current_WMODE_Handling

FYI: The things I said in comment 16 are true for WebRender runs. Non-async drawing (the way to get NPDrawingModelSyncWin) is already broken without Web-Render (bug 1582307). Since we can assume most of the users showing up in telemetry aren't using WebRender, this seems like it would be a problem for that 3%, but I only found bug 1582307 through experimentation with settings -- there are no bug reports to my knowledge. I don't have a good explanation at this point but I'll throw in that the only time I run into Flash in personal web use is on Comcast's site, and I think their plugin is not visible. So the bug wouldn't be detected. Maybe this needs more investigation after all.

We really already blocked off windowed plugin support -- it was dead code -- but we are removing that dead code as part of clearing all of NPAPI in bug 1682030 so I'll just dupe it.