114563 - window.close() crash in TestGtkEmbed

Status: RESOLVED INCOMPLETE

(Core Graveyard :: Embedding: GTK Widget, defect)

Description

[Yanko Kaneti]

In Alex Barnes' word from galeon-devel

"If you go to http://www.tesco.com/register/ and click the "need some
help registering" link (need popups enabled), galeon displays the popup,
but then segfaults when you press the close button on the page."

it crashes in the visibilty callback after its been called zillion times

Comment 1

[Yanko Kaneti]

here is another url that exhibits the behavior directly

http://www.it-vikar.no/web_nyheter_detalj.lasso?Nyhetsnr=n611id

both links in it are window.close()

Comment 2

[Yanko Kaneti]

this bug is still absolutely reproducable for me with both 0.9.7 and recent cvs

Comment 3

[Christopher Blizzard (:blizzard)]

Sounds the same.  duping.

*** This bug has been marked as a duplicate of 121279 ***

Comment 4

[Christopher Blizzard (:blizzard)]

Ahh, it's not the same.

Comment 5

[Christopher Blizzard (:blizzard)]

This only happens if it's the last window that's up.  I think it's because XPCOM
is being accessed after it has been shut down.  Looks like a window destruction
ordering issue.

Comment 6

[Christopher Blizzard (:blizzard)]

This is a reocurring problem.  There are two right ways to fix this:

o Require all callers to hold a reference to the embedding layer until _after_
all of the widgets have been destroyed.  Essentially require them to use
push_startup() and pop_startup().

o Require that any event that might cause the last window to be destroyed will
unwind its stack in such a way that it won't need anything from the XPCOM library.

I'm starting to think that the previous one is the proper solution.  I could put
code in the embedding library that would make it so that this would be tracked
and a warning would be spewed if XPCOM was shut down during the last widget
destruction.

Comment 7

[Johnny Stenback (:jst)]

Dan might have some ideas on this.

Comment 8

[Christopher Blizzard (:blizzard)]

I need some feedback.  I need to get this issue taken care of (fix the test case
or the issue) before I can feel comfortable seeing 0.9.8 go out the door.

Comment 9

[Christopher Blizzard (:blizzard)]

Attachment: fix the test case

This fixes the test case.

Comment 10

[Christopher Blizzard (:blizzard)]

Taking this off the 0.9.8 radar since it's really a bug with the test case, not
in the product.

Comment 11

[Daniel Veditz [:dveditz]]

Can we close this bug at this point? Fixed? obsolete? wontfix?

Comment 12

[Yanko Kaneti]

Flash from the past. Closing. Sorry if it was waiting on me.