Bug 17383
| Summary: | g_object_unref() critical errors on console | ||
|---|---|---|---|
| Product: | WebKit | Reporter: | Alp Toker <alp> |
| Component: | New Bugs | Assignee: | Nobody <webkit-unassigned> |
| Status: | RESOLVED FIXED | ||
| Severity: | Normal | CC: | alp, jmalonzo, zecke |
| Priority: | P2 | Keywords: | Gtk |
| Version: | 528+ (Nightly build) | ||
| Hardware: | All | ||
| OS: | All | ||
Alp Toker
This is a refcounting issue introduced in r30226 or one of the preceding/following revisions.
In certain applications (eg. DevHelp), WebView content is no longer visible and the page is always empty.
There are some potential circular refcounting issues here that may need to be looked at too, but the priority should be to get things working and worry about plugging the leaks later.
| Attachments | ||
|---|---|---|
| Add attachment proposed patch, testcase, etc. |
Alp Toker
The unref messages appear when leaving eg. http://maps.google.com/
Holger Freyther
Let us:
a) revert the refcount changes
b) settle for the traditional scheme that FrameLoaderClient refs and unrefs the
WebFrame and that a WebFrame might outlive a WebCore::Frame.
Feel free to do a) now to give us the time to find a solution that does not involve b).
Alp Toker
Lack of content is an unrelated issue, downgrading to P2.
The console errors are still important (equivalent to a null dereference with optimized GTK+ builds) so need to be fixed, but we can take a bit more time over that.
Alp Toker
(In reply to comment #2)
> Let us:
> a) revert the refcount changes
> b) settle for the traditional scheme that FrameLoaderClient refs and unrefs
> the
> WebFrame and that a WebFrame might outlive a WebCore::Frame.
>
> Feel free to do a) now to give us the time to find a solution that does not
> involve b).
>
Any word on this? The console messages are getting a bit unpopular with users.
Jan Alonzo
Do you guys think this is related to bug #18064?
Holger Freyther
(In reply to comment #5)
> Do you guys think this is related to bug #18064?
Yes, it might be related to this bug. I'm pretty confident that #18411 is fixing both issues. But in both cases a reduced test-case would be appreciated, so we can really be sure the crash is gone.
Jan Alonzo
Just tested with the patch at bug #18411 and the assert failed warnings are still there, so I guess they're not related.
Alp Toker
Holger's recent page changes fixed this issue. Re-open bug if the issue persists.