Summary: | segmentation fault in swfdec mozilla plugin | ||
---|---|---|---|
Product: | WebKit | Reporter: | Riccardo Magliocchetti <riccardo.magliocchetti> |
Component: | Plug-ins | Assignee: | Marco Barisione <marco.barisione> |
Status: | RESOLVED FIXED | ||
Severity: | Normal | CC: | alp, christian, compnerd, hausmann, marco.barisione, marc.ordinasillopis, otte, stempubuntu |
Priority: | P2 | Keywords: | Gtk |
Version: | 528+ (Nightly build) | ||
Hardware: | PC | ||
OS: | Linux | ||
Attachments: |
Description
Riccardo Magliocchetti
2008-09-11 05:20:05 PDT
Yeah, it seems Webkit doesn't provide a ws_info field as it should according to http://developer.mozilla.org/en/NPWindow The code to use it was added because using the wrong Visual might cause an X error, see https://bugs.freedesktop.org/show_bug.cgi?id=16717 and https://bugzilla.mozilla.org/show_bug.cgi?id=445250 I may be experiencing this bug as well -- not entirely sure if it's the same or different, but it looks the same to me. This occurs whenever I leave a page with a Flash element and swfdec, using epiphany-webkit 2.23.91 and webkit r36705. Again, the first time a Flash element is loaded, thing is fine, but when one tries to leave this page, thing become sad and crashes. :( I've attached the full backtrace I encountered. Hope it helps. Arch Linux w/ kernel 2.6.26 (x86_64) Webkit r36705 Gnome 2.22.3 / epiphany-webkit 2.23.91 swfdec 0.8.0 Thanks in advance. Created attachment 23602 [details]
A backtrace from a segfault caused by swfdec 0.8.0 and webkit.
A backtrace from a segfault caused by swfdec 0.8.0 and webkit.
According to http://developer.mozilla.org/en/Gecko_Plugin_API_Reference/Drawing_and_Event_Handling and http://developer.mozilla.org/en/NPP_SetWindow , passing a null window indicates that the window is being destroyed. I seem to have misread that as window *and* ws_info, sorry for that. It should be easy to fix, all's needed is to change the order of deleting ws_info and calling the plugin in PluginView::stop in both WebCore/plugins/gtk/PluginViewGtk.cpp:276 and WebCore/plugins/qt/PluginViewQt.cpp:234. I can't patch myself right away, any takers? I've included a patch that seems to resolve the issue. Thanks for your help and attention. : ) Created attachment 23935 [details]
Stop segfault on leaving page with plugin load
I was doing the same thing(In reply to comment #5) > I've included a patch that seems to resolve the issue. Thanks for your help and > attention. : ) i was doing the exact same thing :) it still crash when you close the page with the flash file (webkit r37056 and midori 0.2.1). Should file a new bug? Created attachment 23936 [details]
Stop segfault on leaving page with plugin load - Qt
I don't get that crash with epiphany-webkit. Perhaps you're experiencing a Midori bug? I was doing the same, and this is why I assigned the bug to me :) Can you please provide a ChangeLog entry with the patch? (maybe a patch for both gtk and qt is enough) Did you test the qt patch? When the patch is ready for review you should set the review field to "?", so a reviewer can review it. Even fixing this bug I can still see problems (using GtkLauncher) with both swfdec and the adobe flash plugin, if you want you can open a bug report and try to fix that too. (In reply to comment #9) > I don't get that crash with epiphany-webkit. Perhaps you're experiencing a > Midori bug? Nope, it happens with GtkLauncher too, filed bug 21240. Thank you for the help Marco. I've uploaded a new patch with ChangeLog and marked it ready for review. Thus far, I've been unable to get QtLauncher to function with swfdec, so I haven't been able to test the patch. I hope to figure this out shortly, and will mark that as needing review once I get that funny business sorted. Created attachment 24017 [details]
Stop segfault on leaving page with plugin load -Gtk w/ changelog
Comment on attachment 24017 [details]
Stop segfault on leaving page with plugin load -Gtk w/ changelog
Patches that need a review should be marked as "?", not "+".
It's nice to have a link to the bug report in the changelog entry but I don't know if it's a rule or what.
The changelog entry should also say how the bug was fixed, not only describe the bug.
(In reply to comment #14) > (From update of attachment 24017 [details] [edit]) > Patches that need a review should be marked as "?", not "+". > > It's nice to have a link to the bug report in the changelog entry but I don't > know if it's a rule or what. > > The changelog entry should also say how the bug was fixed, not only describe > the bug. > Thanks Marco. I misclicked on the +, meant ?. Submitting updated patch. Created attachment 24080 [details]
Patch for GTK with better changelog.
(In reply to comment #16) > Created an attachment (id=24080) [edit] > Patch for GTK with better changelog. > Hi, in r37535 a fix for a plugin crasher was landed. It adds a quirk. I wonder if your patch achieves the same thing without the need for a quirk. If so, we can back out r37535 and land your patch instead. Can you check this out? Thanks (In reply to comment #17) > Hi, > > in r37535 a fix for a plugin crasher was landed. It adds a quirk. I wonder if > your patch achieves the same thing without the need for a quirk. If so, we can > back out r37535 and land your patch instead. > > Can you check this out? Thanks Hi Alp, Webkit r37567 + this patch + swfdec -> ok Webkit r37567 + this patch + flash 10 rc -> ok Webkit r37567 + this patch + reverted r37535 with a patch generated by trac + flash 10 rc -> segmentation fault in adobe plugin Please land this patch :) Comment on attachment 24080 [details]
Patch for GTK with better changelog.
Looks sane. I did land this patch.
Comment on attachment 24080 [details]
Patch for GTK with better changelog.
Clearing review flag as this has been applied.
Landed Jeff's patch for the Qt port in 38125 Closing, since it is landed; please reopen if needed. |