[Webkit-unassigned] [Bug 251336] [SOUP] Very slow to load the CNN or Globe and Mail websites

bugzilla-daemon at webkit.org bugzilla-daemon at webkit.org
Wed Oct 25 07:04:30 PDT 2023


https://bugs.webkit.org/show_bug.cgi?id=251336

--- Comment #16 from Michael Catanzaro <mcatanzaro at redhat.com> ---
(In reply to Michael Catanzaro from comment #14)
> Something very weird that I noticed: cnn.com consistently loads *much*
> faster using my personal jhbuild-built WebKitGTK (at 266205 at main) than it
> does in Tech Preview (2.41.6).

So we've figured out that this bug is caused by flatpak itself:

 * Performance is good when building with jhbuild instead of flatpak
 * Performance is good when using GIO_USE_PORTALS=1 and GDK_DEBUG=portals
 * Performance is good when building and running the org.gnome.Epiphany.Debug.json flatpak-builder manifest in GNOME Builder

Notably, when using the flatpak-builder manifest in GNOME Builder, we are running the exact same software/dependencies as Epiphany Tech Preview, but not actually running under 'flatpak run', so this eliminates the potential that a bad version of some dependency (I was suspecting libsoup) is to blame for the difference. It also means we really have to test under 'flatpak run' itself to reproduce the problem.

Unfortunately this means it's really hard to set up an edit/compile/test workflow to debug the issue.

Patrick thinks we should debug (a) whether anything slow is happening during main context iterations that might be preventing the next main context iteration from running in a reasonable time, and also (b) soup-client-message-io-http2.c. However, I don't know how to do this reasonably. I think I would need to build and install a new flatpak runtime each time I want to add a printf somewhere. This makes it difficult to make any progress.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.webkit.org/pipermail/webkit-unassigned/attachments/20231025/034c3de8/attachment.htm>


More information about the webkit-unassigned mailing list