Gecko-dev Diary
Starting in August 2023 I'll be upgrading the Sailfish OS browser from Gecko version ESR 78 to ESR 91. This page catalogues my progress.
Latest code changes are in the gecko-dev sailfishos-esr91 branch.
There is an index of all posts in case you want to jump to a particular day.
Click the icon for the Gecko-dev Diary RSS feed.
Gecko
5 most recent items
13 Apr 2024 : Day 215 #
This morning I had a really fruitful discussion with everyone attending the fortnightly Sailfish Community Meeting. As you'll know I've reached the point where it's clear that rendering is happening to the surface, but there's something not working between the surface and the screen that's preventing any colour getting to the screen.
There were two really excellent outcomes from the meeting. The first involved potential solutions to this problem and I'd like to spend a little time discussing that now. The second... well I'll come to that later.
So let's first look at those potential solutions. The first thing was that there's a process that happens between the sailfish-browser code and the gecko code that performs the WebView rendering. It takes the offscreen render and ensures it gets to the screen. And as Raine (rainemak) explained during the meeting, that starts in the DeclarativeWebContainer::clearWindowSurface() method. That's part of the sailfish-browser code.
Recently I've been playing around with the CompositorOGL::EndFrame() method. That's at the other end of the process. So that means the part of the code I need to check should be somewhere between these two points. Here's some of the relevant conversation (slightly amended by me for brevity and clarify):
So that's some clear things to check, specifically the code around clearWindowSurface() and the calling of glMakeCurrent() (which is most likely called fMakeCurrent() if it's in the gecko code). So this is the reason for the first part of my upcoming plan:
But there was more useful info from Raine in addition to this. He also suggested to take a look at the DrawUnderlay() method — if it exists — or the functionality that's replaced it.
Finally, he also suggested to check the updatePaintNode() method which is part of the qtmozembed codebase:
All of these are excellent suggestions and they're now on my list of things to check. Yesterday I'd started working through the patches and I'll have to finish that task before I move on to these suggestions, since the patches have ramifications for other parts of the code as well, but I'm really hoping — and quite energised by the prospect — that these tips from Raine might lead to a solution for the offscreen rendering problems.
It was also great to see that the discussion during the meeting also led to some helpful OpenGL debugging suggestions on the forum. I've already thanked Tone (tortoisedoc) previously but now I must also add Damien Caliste (dcaliste) to the list of people who deserve my thanks.
So that was the first nice outcome, what about the second? Well that's of a slightly different nature. During the meeting thigg also made an unexpected by also really superb offer:
I can't get enough of Thilo's pictures and am already looking forward to being able to share them in these diaries when the time comes.
That's it for today, but I'll return tomorrow with updates on patches.
If you'd like to read any of my other gecko diary entries, they're all available on my Gecko-dev Diary page.
There were two really excellent outcomes from the meeting. The first involved potential solutions to this problem and I'd like to spend a little time discussing that now. The second... well I'll come to that later.
So let's first look at those potential solutions. The first thing was that there's a process that happens between the sailfish-browser code and the gecko code that performs the WebView rendering. It takes the offscreen render and ensures it gets to the screen. And as Raine (rainemak) explained during the meeting, that starts in the DeclarativeWebContainer::clearWindowSurface() method. That's part of the sailfish-browser code.
Recently I've been playing around with the CompositorOGL::EndFrame() method. That's at the other end of the process. So that means the part of the code I need to check should be somewhere between these two points. Here's some of the relevant conversation (slightly amended by me for brevity and clarify):
rainemak: flypig, have you tried to change gl clear color?
flypig: Yes, I tried glClearColor. The change shows up with fReadPixels, but not on screen.
rainemak: has to be that off screen is not there
flypig: Could you elaborate?
rainemak: flypig, one thing worth looking is that gecko doesn't create any headless things in addition to what you're expecting it to create
flypig: rainemak, how would that manifest itself? I've been careful to ensure there's only one display value
created. Is there anything else to check for this you can think of?
rainemak: flypig, also gl makeCurrent and swap I'd check
rainemak: DeclarativeWebContainer::clearWindowSurface
rainemak: ^ is that being called properly
flypig: That's in sailfish-browser rainemak? I've not checked that; I'll do so.
flypig: Yes, I tried glClearColor. The change shows up with fReadPixels, but not on screen.
rainemak: has to be that off screen is not there
flypig: Could you elaborate?
rainemak: flypig, one thing worth looking is that gecko doesn't create any headless things in addition to what you're expecting it to create
flypig: rainemak, how would that manifest itself? I've been careful to ensure there's only one display value
created. Is there anything else to check for this you can think of?
rainemak: flypig, also gl makeCurrent and swap I'd check
rainemak: DeclarativeWebContainer::clearWindowSurface
rainemak: ^ is that being called properly
flypig: That's in sailfish-browser rainemak? I've not checked that; I'll do so.
So that's some clear things to check, specifically the code around clearWindowSurface() and the calling of glMakeCurrent() (which is most likely called fMakeCurrent() if it's in the gecko code). So this is the reason for the first part of my upcoming plan:
flypig: Alright, I'm going to check on the clearWindowSurface side of things and everything between that and CompositorOGL::EndFrame().
rainemak: flypig, that's at least scheduling glclear
flypig: Okay, that's useful to know. Then maybe I can put lots of readPixels in between the two to find out exactly which step is failing.
rainemak: flypig, that's at least scheduling glclear
flypig: Okay, that's useful to know. Then maybe I can put lots of readPixels in between the two to find out exactly which step is failing.
But there was more useful info from Raine in addition to this. He also suggested to take a look at the DrawUnderlay() method — if it exists — or the functionality that's replaced it.
rainemak: flypig, and draw underlay is getting called properly as well? that call makeCurrent
rainemak: that calls makeCurrent
flypig: One second, let me check.
rainemak: let me think
flypig: That's called in EmbedLiteCompositorBridgeParent::CompositeToDefaultTarget() ?
rainemak: it starts around there
rainemak: did we remove on esr78 DrawUnderlay like methon ( there's still QMozWindowPrivate::DrawOverlay )
rainemak: hmm
rainemak: I'm not seeing that makeCurrent
rainemak: that calls makeCurrent
flypig: One second, let me check.
rainemak: let me think
flypig: That's called in EmbedLiteCompositorBridgeParent::CompositeToDefaultTarget() ?
rainemak: it starts around there
rainemak: did we remove on esr78 DrawUnderlay like methon ( there's still QMozWindowPrivate::DrawOverlay )
rainemak: hmm
rainemak: I'm not seeing that makeCurrent
Finally, he also suggested to check the updatePaintNode() method which is part of the qtmozembed codebase:
rainemak: flypig, QuickMozView::updatePaintNode I'd check this one as browser rendering works
rainemak: i.e. does getPlatform and MozMaterialNode::preprocess work as expected
rainemak: thinking if there's something wrong now with previous scenegraph integration
rainemak: i.e. does getPlatform and MozMaterialNode::preprocess work as expected
rainemak: thinking if there's something wrong now with previous scenegraph integration
All of these are excellent suggestions and they're now on my list of things to check. Yesterday I'd started working through the patches and I'll have to finish that task before I move on to these suggestions, since the patches have ramifications for other parts of the code as well, but I'm really hoping — and quite energised by the prospect — that these tips from Raine might lead to a solution for the offscreen rendering problems.
It was also great to see that the discussion during the meeting also led to some helpful OpenGL debugging suggestions on the forum. I've already thanked Tone (tortoisedoc) previously but now I must also add Damien Caliste (dcaliste) to the list of people who deserve my thanks.
So that was the first nice outcome, what about the second? Well that's of a slightly different nature. During the meeting thigg also made an unexpected by also really superb offer:
thilo[m]: #info thigg, community
thilo[m]: Flypig: you have a milestone in gecko after which you have a big party planned? ;)
flypig: thilo[m], once the offscreen render is working, I'll generate patches, and then it will be time to celebrate (for me at least!).
thilo[m]: Cool, i'll prepare pictures ;)
thilo[m]: Flypig: you have a milestone in gecko after which you have a big party planned? ;)
flypig: thilo[m], once the offscreen render is working, I'll generate patches, and then it will be time to celebrate (for me at least!).
thilo[m]: Cool, i'll prepare pictures ;)
I can't get enough of Thilo's pictures and am already looking forward to being able to share them in these diaries when the time comes.
That's it for today, but I'll return tomorrow with updates on patches.
If you'd like to read any of my other gecko diary entries, they're all available on my Gecko-dev Diary page.
Comments
Uncover Disqus comments