List items
Items from the current list are shown below.
Blog
5 Jul 2024 : Day 279 #
Today I'm continuing with the plan I hatched yesterday. I'm half way through Step 6 ("Build and check everything still works"), having built the packages overnight.
This morning I've done some basic testing: installed the packages, run the browser to view a page with some WebGL content, run the harbour-webview app.
I was a bit perturbed to find the browser failed to render WebGL if I executed it from the app drawer rather than the command line. But this got fixed when I deleted the startupCache:
So, Step 6 is completed.
Step 7: Take a deep breath. Commit the staged changes.
After making all of the changes and discarding anything unnecessary, there are now nineteen modified files to commit.
Deep breath.
Step 8: Admire the resulting, much cleaner, GLSCreenBuffer commit now at HEAD.
Looking through the diff, I think it'll make a decent patch. It remains uncomfortably big — not ideal — but I've now spent a lot of time cutting it down to the bare minimum and I'm satisfied that this is the best I can hope for, at least for now. Certainly it's not going to be worth spending more time trying to cut it down further.
Which leave me with just one more thing to do.
Step 9: Another deep breath. Force push everything to the server. Done.
At least, the plan was to force push. But interestingly, it turns out I don't need to force push at all. It's been so long since I started work on restoring GLScreenBuffer that I thought I must have pushed some of the changes to the remote. But it turns out I hadn't.
This is great news! It means I won't be doing anything destructive by pushing. No need to take any deep breaths after all!
So there we are. I can finally draw a line under the WebView rendering changes. It's a line drawn in pencil of course as I'm sure I'll find things I need to come back to, but for now it's time to move on.
The next thing I want to tackle is another Web-related acronym: WebRTC. I intentionally disabled this back on Day 38 in order to get the build to go through. And as we'll see there's a reason I didn't tackle it head on at the time. Now there's no avoiding it: it's time to swallow the bitter pill and get to it.
There are five patches which relate to WebRTC. Some look tractable. But others... well... I don't fancy the chances of that 2.4 MiB patch applying cleanly:
The problem is that the README.md file mentioned in the explanation is missing from the source tree. A bit of a search throws up a potentially useful comment in Bugzilla that links through to some text that's likely the README.md being referenced. It says the following:
Honestly that doesn't seem as bad as I'd feared. But this will of course constitute only one of the five patches that need to be applied. So tomorrow I'll start looking at the other patches to see what's really going to be involved here.
If you'd like to read any of my other gecko diary entries, they're all available on my Gecko-dev Diary page.
This morning I've done some basic testing: installed the packages, run the browser to view a page with some WebGL content, run the harbour-webview app.
I was a bit perturbed to find the browser failed to render WebGL if I executed it from the app drawer rather than the command line. But this got fixed when I deleted the startupCache:
$ rm ~/.local/share/org.sailfishos/browser/.mozilla/startupCache/*It's luck that I noticed this, but I'm glad I did: glitches like this can cause an almighty headache if you don't discover them early on. I also tested the Settings app to check the WebView still works on the various Account creation pages. Using the ESR 91 WebView I was successfully able to view the Twitter (nee X) Terms of Service; I didn't go as far as agreeing to them of course.
So, Step 6 is completed.
Step 7: Take a deep breath. Commit the staged changes.
After making all of the changes and discarding anything unnecessary, there are now nineteen modified files to commit.
$ git diff --stat --cached dom/base/nsGlobalWindowOuter.cpp | 6 - gfx/gl/GLContext.cpp | 34 +-- gfx/gl/GLContext.h | 6 +- gfx/gl/GLContextEGL.h | 6 +- gfx/gl/GLContextProviderEGL.cpp | 242 +++------------------ gfx/gl/GLLibraryEGL.cpp | 11 +- gfx/gl/GLLibraryEGL.h | 6 +- gfx/gl/GLScreenBuffer.cpp | 2 +- gfx/gl/GLTextureImage.cpp | 26 +-- gfx/gl/SharedSurface.cpp | 2 - gfx/gl/SharedSurface.h | 17 +- gfx/gl/SharedSurfaceEGL.h | 4 - gfx/gl/SharedSurfaceGL.cpp | 12 +- gfx/gl/SharedSurfaceGL.h | 16 ++ gfx/layers/CompositorTypes.h | 1 - gfx/layers/client/TextureClient.cpp | 42 +--- gfx/layers/client/TextureClient.h | 3 - gfx/layers/client/TextureClientSharedSurface.cpp | 1 - gfx/layers/ipc/CompositorVsyncScheduler.cpp | 1 + 19 files changed, 93 insertions(+), 345 deletions(-)Perhaps the more interesting thing to check is what will happen to the commit I'm merging this into. Part of the aim here has been to reduce the size of this commit to make it more manageable and I'm interested to know whether or not this has been successful. Here's the incumbent commit as it currently looks:
$ git diff --stat HEAD~..HEAD dom/base/nsGlobalWindowOuter.cpp | 6 + gfx/gl/GLContext.cpp | 74 +++++- gfx/gl/GLContext.h | 21 +- gfx/gl/GLContextEGL.h | 6 +- gfx/gl/GLContextProviderEGL.cpp | 240 +++++++++++++++++- gfx/gl/GLContextProviderImpl.h | 23 ++ gfx/gl/GLLibraryEGL.cpp | 11 +- gfx/gl/GLLibraryEGL.h | 6 +- gfx/gl/GLScreenBuffer.cpp | 274 +++++++++++++++++++++ gfx/gl/GLScreenBuffer.h | 132 +++++++++- gfx/gl/GLTextureImage.cpp | 34 ++- gfx/gl/SharedSurface.cpp | 116 ++++++++- gfx/gl/SharedSurface.h | 122 ++++++++- gfx/gl/SharedSurfaceEGL.cpp | 89 +++++-- gfx/gl/SharedSurfaceEGL.h | 34 ++- gfx/gl/SharedSurfaceGL.cpp | 64 ++++- gfx/gl/SharedSurfaceGL.h | 32 ++- gfx/gl/SurfaceTypes.h | 2 + gfx/gl/TextureImageEGL.cpp | 221 +++++++++++++++++ gfx/gl/TextureImageEGL.h | 80 ++++++ gfx/gl/moz.build | 1 + gfx/layers/CompositorTypes.h | 1 + gfx/layers/client/TextureClient.cpp | 42 +++- gfx/layers/client/TextureClient.h | 3 + gfx/layers/client/TextureClientSharedSurface.cpp | 50 ++++ gfx/layers/client/TextureClientSharedSurface.h | 25 ++ gfx/layers/ipc/CompositorVsyncScheduler.cpp | 2 - gfx/layers/opengl/CompositorOGL.cpp | 24 +- 28 files changed, 1631 insertions(+), 104 deletions(-)After merging in these changes to the commit, it'll look like this:
$ git diff --stat HEAD~ gfx/gl/GLContext.cpp | 40 ++- gfx/gl/GLContext.h | 17 +- gfx/gl/GLContextProviderEGL.cpp | 36 +++ gfx/gl/GLContextProviderImpl.h | 23 ++ gfx/gl/GLScreenBuffer.cpp | 274 +++++++++++++++++++++ gfx/gl/GLScreenBuffer.h | 132 +++++++++- gfx/gl/GLTextureImage.cpp | 10 + gfx/gl/SharedSurface.cpp | 114 ++++++++- gfx/gl/SharedSurface.h | 105 +++++++- gfx/gl/SharedSurfaceEGL.cpp | 89 +++++-- gfx/gl/SharedSurfaceEGL.h | 30 ++- gfx/gl/SharedSurfaceGL.cpp | 74 +++++- gfx/gl/SharedSurfaceGL.h | 46 ++++ gfx/gl/SurfaceTypes.h | 2 + gfx/gl/TextureImageEGL.cpp | 221 +++++++++++++++++ gfx/gl/TextureImageEGL.h | 80 ++++++ gfx/gl/moz.build | 1 + gfx/layers/client/TextureClientSharedSurface.cpp | 49 ++++ gfx/layers/client/TextureClientSharedSurface.h | 25 ++ gfx/layers/ipc/CompositorVsyncScheduler.cpp | 1 - gfx/layers/opengl/CompositorOGL.cpp | 24 +- 21 files changed, 1334 insertions(+), 59 deletions(-)There are big gains in reducing the diff for GLContextProviderEGL.cpp and the only losers are SharedSurfaceGL.cpp and SharedSurfaceGL.h. But the wins are much greater than the losses, so I'm happy with this. That means it's time to apply the changes to the previous commit.
Deep breath.
$ git log -1
commit 7437a9d17284bd9a4427502b1af6e94a9d7ee356 (HEAD ->
FIREFOX_ESR_91_9_X_RELBRANCH_patches)
Author: David Llewellyn-Jones <david@flypig.co.uk>
Date: Wed May 22 08:35:49 2024 +0100
Restore GLScreenBuffer and TextureImageEGL
TextureImageEGL was removed with Bug 1716559:
https://phabricator.services.mozilla.com/D117904
GLScreenBuffer was removed with Bug 1632249:
https://phabricator.services.mozilla.com/D75055
Both are used as part of the Sailfish OS WebView offscreen render
pipeline. This patch reintroduces these two classes and allows them to
be made use of.
$ git commit --amend
[FIREFOX_ESR_91_9_X_RELBRANCH_patches 1765dd0460a9] Restore GLScreenBuffer and
TextureImageEGL
Date: Wed May 22 08:35:49 2024 +0100
21 files changed, 1334 insertions(+), 59 deletions(-)
create mode 100644 gfx/gl/TextureImageEGL.cpp
create mode 100644 gfx/gl/TextureImageEGL.h
Step 7 completed. On to Step 8.Step 8: Admire the resulting, much cleaner, GLSCreenBuffer commit now at HEAD.
Looking through the diff, I think it'll make a decent patch. It remains uncomfortably big — not ideal — but I've now spent a lot of time cutting it down to the bare minimum and I'm satisfied that this is the best I can hope for, at least for now. Certainly it's not going to be worth spending more time trying to cut it down further.
Which leave me with just one more thing to do.
Step 9: Another deep breath. Force push everything to the server. Done.
At least, the plan was to force push. But interestingly, it turns out I don't need to force push at all. It's been so long since I started work on restoring GLScreenBuffer that I thought I must have pushed some of the changes to the remote. But it turns out I hadn't.
This is great news! It means I won't be doing anything destructive by pushing. No need to take any deep breaths after all!
$ git push
Enumerating objects: 92, done.
Counting objects: 100% (92/92), done.
Delta compression using up to 16 threads
Compressing objects: 100% (51/51), done.
Writing objects: 100% (51/51), 22.70 KiB | 2.84 MiB/s, done.
Total 51 (delta 40), reused 0 (delta 0), pack-reused 0
remote: Resolving deltas: 100% (40/40), completed with 37 local objects.
To github.com:llewelld/gecko-dev-mirror.git
c6ea49286566..1765dd0460a9 FIREFOX_ESR_91_9_X_RELBRANCH_patches ->
FIREFOX_ESR_91_9_X_RELBRANCH_patches
$ cd ..
$ git push
Enumerating objects: 13, done.
Counting objects: 100% (13/13), done.
Delta compression using up to 16 threads
Compressing objects: 100% (7/7), done.
Writing objects: 100% (7/7), 1.46 KiB | 373.00 KiB/s, done.
Total 7 (delta 5), reused 0 (delta 0), pack-reused 0
remote: Resolving deltas: 100% (5/5), completed with 5 local objects.
To github.com:llewelld/gecko-dev.git
9f64ce35a187..5b9c330998db sailfishos-esr91 -> sailfishos-esr91
Done!So there we are. I can finally draw a line under the WebView rendering changes. It's a line drawn in pencil of course as I'm sure I'll find things I need to come back to, but for now it's time to move on.
The next thing I want to tackle is another Web-related acronym: WebRTC. I intentionally disabled this back on Day 38 in order to get the build to go through. And as we'll see there's a reason I didn't tackle it head on at the time. Now there's no avoiding it: it's time to swallow the bitter pill and get to it.
There are five patches which relate to WebRTC. Some look tractable. But others... well... I don't fancy the chances of that 2.4 MiB patch applying cleanly:
$ grep -rInl "webrtc" * --include="*.patch" | xargs ls
-lah | awk '{print t$9 "\t" $5}'
0077-sailfishos-webrtc-Adapt-build-configuration-for-Sail.patch 5.6K
0078-sailfishos-webrtc-Regenerate-moz.build-files.-JB-537.patch 2.4M
0079-sailfishos-webrtc-Disable-desktop-sharing-feature-on.patch 2.1K
0080-sailfishos-webrtc-Enable-GMP-for-encoding-decoding.-.patch 33K
0081-sailfishos-webrtc-Implement-video-capture-module.-JB.patch 29K
In practice the really big patch — patch 0078 — is due to changes to the build system which have been auto-generated. Specifically it relates to components that would usually be built using GN, but where the build configuration has been automgically converted into the moz.build files used by the Gecko build system. Here's the text from the docs explaining what's going on here;GN is a third-party build tool used by chromium and some related projects that are vendored in mozilla-central. Rather than requiring GN to build or writing our own build definitions for these projects, we have support in the build system for translating GN configuration files into moz.build files. In most cases these moz.build files will be like any others in the tree (except that they shouldn’t be modified by hand), however those updating vendored code or building on platforms not supported by Mozilla automation may need to re-generate these files. This is a per-project process, described in dom/media/webrtc/third_party_build/gn-configs/README.md for webrtc. As of writing, it is very specific to webrtc, and likely doesn’t work as-is for other projects.
The problem is that the README.md file mentioned in the explanation is missing from the source tree. A bit of a search throws up a potentially useful comment in Bugzilla that links through to some text that's likely the README.md being referenced. It says the following:
Generate new gn json files and moz.build files for building libwebrtc in our tree
/!\ This is only supported on Linux and macOS. If you are on Windows, you can run the script under WSL.
1. The script should be run from the top directory of our firefox tree.
Adding new configurations to the build
Edit the main function in the python/mozbuild/mozbuild/gn_processor.py file.
/!\ This is only supported on Linux and macOS. If you are on Windows, you can run the script under WSL.
1. The script should be run from the top directory of our firefox tree.
./mach python python/mozbuild/mozbuild/gn_processor.py dom/media/webrtc/
third_party_build/gn-configs/webrtc.json
2. Checkin all the generated/modified files and try your build!Adding new configurations to the build
Edit the main function in the python/mozbuild/mozbuild/gn_processor.py file.
Honestly that doesn't seem as bad as I'd feared. But this will of course constitute only one of the five patches that need to be applied. So tomorrow I'll start looking at the other patches to see what's really going to be involved here.
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