Bug 260519 – [GTK] web process crash when pressing down-arrow and page-down

• Component: WebKitGTK
• Status: NEW

[Haelwenn (lanodan) Monnier]

Created attachment 467387 [details]
backtrace of WebProcess

System:
- Gentoo Linux
- LLVM/Clang 16.0.6
- Wayland (sway)

Steps:
- Open browser like MiniBrowser
- Let it load webkitgtk.org or any other website
- Press down arrow or page down (instant crash)

Logs:

Could not determine the accessibility bus address
ERROR: Failed to connect to RealtimeKit: Could not connect: No such file or directory
/var/tmp/notmpfs/portage/net-libs/webkit-gtk-2.41.91-r410/work/webkitgtk-2.41.91/Source/WTF/wtf/linux/RealTimeThreads.cpp(190) : void WTF::RealTimeThreads::realTimeKitMakeThreadRealTime(uint64_t, uint64_t, uint32_t)

ASSERTION FAILED: inProgrammaticScroll == (options.type == ScrollType::Programmatic)
/var/tmp/notmpfs/portage/net-libs/webkit-gtk-2.41.91-r410/work/webkitgtk-2.41.91/Source/WebCore/page/scrolling/AsyncScrollingCoordinator.cpp(333) : virtual bool WebCore::AsyncScrollingCoordinator::requestScrollToPosition(ScrollableArea &, const ScrollPosition &, const ScrollPositionChangeOptions &)

ERROR: 0x7b9768000c00 - [PID=1801] WebProcessProxy::didClose (web process crash)
/var/tmp/notmpfs/portage/net-libs/webkit-gtk-2.41.91-r410/work/webkitgtk-2.41.91/Source/WebKit/UIProcess/WebProcessProxy.cpp(1050) : virtual void WebKit::WebProcessProxy::didClose(IPC::Connection &)
ERROR: 0x7b9768000c00 - [PID=1801] WebProcessProxy::processDidTerminateOrFailedToLaunch: reason=Crash
/var/tmp/notmpfs/portage/net-libs/webkit-gtk-2.41.91-r410/work/webkitgtk-2.41.91/Source/WebKit/UIProcess/WebProcessProxy.cpp(1058) : void WebKit::WebProcessProxy::processDidTerminateOrFailedToLaunch(ProcessTerminationReason)
ERROR: 0x7b97b50c4100 - [pageProxyID=10, webPageID=11, PID=1801] WebPageProxy::processDidTerminate: (pid 1801), reason=Crash
/var/tmp/notmpfs/portage/net-libs/webkit-gtk-2.41.91-r410/work/webkitgtk-2.41.91/Source/WebKit/UIProcess/WebPageProxy.cpp(8839) : void WebKit::WebPageProxy::resetStateAfterProcessTermination(ProcessTerminationReason)
ERROR: 0x7b97b50c4100 - [pageProxyID=10, webPageID=11, PID=1801] WebPageProxy::dispatchProcessDidTerminate: reason=Crash
/var/tmp/notmpfs/portage/net-libs/webkit-gtk-2.41.91-r410/work/webkitgtk-2.41.91/Source/WebKit/UIProcess/WebPageProxy.cpp(8897) : void WebKit::WebPageProxy::dispatchProcessDidTerminate(ProcessTerminationReason)

** (MiniBrowser:1753): WARNING **: 05:00:33.953: WebProcess CRASHED

[Michael Catanzaro]

So you compiled with assertions enabled? That's unusual, but very useful to reveal bugs like this. Unfortunately your backtrace is useless because you don't have debuginfo. You'll need to recompile with -g.

[Haelwenn (lanodan) Monnier]

> So you compiled with assertions enabled? 

I didn't set an option for enabling assertions, might be due to a WebKit buildsystem change or some side-effect of not having "Release" mode but "Gentoo" mode, which AFAIK is based on "RelWithDebInfo".
Or might be due to removing an old hack where -DNDEBUG was appended to CPPFLAGS, removed it as it would fail to compile (Source/WebCore/page/LocalFrameViewLayoutContext.cpp:281:13: error: use of undeclared identifier 'showRenderTree').

> Unfortunately your backtrace is useless because you don't have debuginfo. You'll need to recompile with -g.

It had -ggdb so not sure why it doesn't…

[Michael Catanzaro]

(In reply to Haelwenn (lanodan) Monnier from comment #2)
> Or might be due to removing an old hack where -DNDEBUG was appended to
> CPPFLAGS, removed it as it would fail to compile

I think that would indeed cause debug assertions to be enabled. -DNDEBUG is added by CMake itself when using a Release or RelWithDebInfo build, but if Gentoo mode doesn't do that, it would explain why you get debug assertions.

I think running with debug assertions enabled is a good idea, sort of. Problem is it is not well-tested, so basically only Gentoo users will ever notice these assertion failures. Pragmatically, it's probably not a good idea for only Gentoo to have these enabled. But I suppose it's good for WebKit upstream to receive bug reports for problems we wouldn't otherwise know about. :)

> (Source/WebCore/page/LocalFrameViewLayoutContext.cpp:281:13: error: use of
> undeclared identifier 'showRenderTree').
> 
> > Unfortunately your backtrace is useless because you don't have debuginfo. You'll need to recompile with -g.
> 
> It had -ggdb so not sure why it doesn't…

Strange. Maybe try gdb instead of lldb then? I have actually never seen a backtrace from lldb on Linux. What I can say for sure is the bug won't be fixed without a better backtrace.