Bug 186044

Summary: [GTK] Webkit Web Process crashes on certain Chase.com pages.
Product: WebKit Reporter: Ryan Farmer <rfarmer84>
Component: WebKitGTKAssignee: Nobody <webkit-unassigned>
Status: RESOLVED INVALID    
Severity: Normal CC: bugs-noreply, calvaris, mcatanzaro
Priority: P2    
Version: Other   
Hardware: PC   
OS: Linux   
Attachments:
Description Flags
Backtrace
none
Second attempt at a backtrace. none

Description Ryan Farmer 2018-05-28 22:59:37 PDT
Created attachment 341473 [details]
Backtrace

On GNOME Web 3.28.2.1 with WebkitGTK 2.20.2, the Webkit Web Process crashes sometimes on Chase Bank's website and the tab must be reloaded. 

I generated a backtrace  of the most recent incident which is attached.
Comment 1 Ryan Farmer 2018-05-28 23:10:42 PDT
Excuse me for having an absent minded moment.

I forgot that I was running a scratch build of Webkit for an unrelated reason.

I backed it out to the latest Fedora package (2.20.2-1) and I'll post a new backtrace.
Comment 2 Ryan Farmer 2018-05-28 23:32:41 PDT
Created attachment 341478 [details]
Second attempt at a backtrace.

This is my second attempt. 


I rolled back to the latest version of WebkitGTK (2.20.2-1) provided in Fedora 28 and installed a bunch more debuginfo packages and then ran coredumpctl gdb again with "thread apply all bt full". 

Eventually, gdb has an internal error and says that continuing further may prove to be unreliable, so I cut this off at that point, hoping you'll see something useful.

If you want me to continue past that point, let me know. I tried it once and it would go on for a while, have another error, and so on.
Comment 3 Michael Catanzaro 2018-05-29 16:07:53 PDT
gdb has become a disaster. :(

It's enough to see that this is an Epiphany password manager bug. Could you please report on bugzilla.gnome.org?

Could you also try to get a normal 'bt full' backtrace without the 'thread apply all'? That way we can get a better trace for the crashing thread. The other threads are just getting in the way, due to the gdb bug.
Comment 4 Michael Catanzaro 2018-05-29 16:16:03 PDT
It's hitting one of these, in ephy_password_manager_save:

  g_assert (EPHY_IS_PASSWORD_MANAGER (self));
  g_assert (origin);
  g_assert (target_origin);
  g_assert (password);
  g_assert (!username_field || username);
  g_assert (!password_field || password);

We need the 'bt full' to have a chance at figuring out which. It's unlikely that gdb will die before printing this much, since it will be at the top of the backtrace and won't involve any C++.
Comment 5 Ryan Farmer 2018-05-29 17:42:36 PDT
I'm not liking this new gitlab thing, but I created an account and am reporting this there:

See: https://gitlab.gnome.org/GNOME/epiphany/issues/11

I had to attach the backtrace to a reply to my own bug because gitlab said it attached it and then didn't. *sigh*
Comment 6 Michael Catanzaro 2018-05-30 06:27:46 PDT
(In reply to Ryan Farmer from comment #5)
> I'm not liking this new gitlab thing, but I created an account and am
> reporting this there:
> 
> See: https://gitlab.gnome.org/GNOME/epiphany/issues/11
> 
> I had to attach the backtrace to a reply to my own bug because gitlab said
> it attached it and then didn't. *sigh*

Yeah, GitLab is definitely not as good at bug tracking. Thanks. Is it blocking you from reporting new bugs on Bugzilla already? Epiphany bugs have (obviously) not been migrated yet.... Whatever.