10273 - NECKO: about:blank as linked stylesheet causes crash

Status: VERIFIED FIXED

(Core :: Networking, defect, P3)

Description

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

If about:blank is the href for a linked stylesheet, viewer crashes.

Tested on: 1999-07-20-10-M9, Linux.

Test case to be attached.

Comment 1

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Attachment: test case causing crash.

Comment 2

[Peter Linss]

Amusingly enough, this takes Nav down too (hard), so don't even try it. (I had
to reboot my machine to get networking going again after loading the test in
4.6)

Comment 3

[Peter Linss]

The problem is that for "about:blank" NS_NewUnicharStreamLoader calls the
callback function synchronously. As far as I know, that should never happen. Is
this due to the "about:blank" workaround that went in for ender? Will necko have
the same behavior? If so, send this bug back to me and I'll make the CSSLoader
deal with it.

Comment 4

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

I'm seeing what's probably the same thing happening in
http://bugzilla.mozilla.org/showattachment.cgi?attach_id=753 with a nonexistent
resource: URL (the old xul.css).  The stack trace is very similar.

Comment 5

[Gagan]

Pl. verify with Necko.

Comment 6

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

I'm seeing similar crashes for nonexistant file: stylesheets.

Comment 7

[Paul MacQuiddy]

dbaron, have you tried this with a necko build?

Comment 8

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Not yet.  I should be able to in a few days.

Comment 9

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Actually, I believe this may have been Linux-only, so I can't test it for a
while.

Comment 10

[Warren Harris]

I'll look at this.

Comment 11

[Paul MacQuiddy]

this is crashing on all platforms

Comment 12

[Warren Harris]

This is still a problem -- it hits an assert in nsCSSLoader.cpp. How important
is this to change? Making necko work differently for this case could be done
(basically this is a case where calling AsyncRead does the work before
returning), but it seems somewhat unnecessary from necko's perspective. Maybe
nsCSSLoader should change.

Comment 13

[Peter Linss]

I accept the fact that Necko doesn't *need* to use async callbacks for this
case, but it is extremely bad if Necko uses mixed models. The call site code can
be/is very different for synchronous reads vs async reads, especially when the
behavior varies by URL scheme and *platform*.

Please make the network code just always be async. It is a simpler model and
will avoid many similar problems in the future.

Comment 14

[cpratt]

Although it doesn't crash my M10 build on Windows (build ID 1999082008), it does
cause garbage to be drawn in the main browser window. Changing OS to ALL. (Mac
OS build is currently not working.)

Comment 15

[Warren Harris]

Can we move this to m10 then?

Comment 16

[leger]

If this is not to be fixed in M9, then we'll need it Release Noted.  so, ya
wanna move to M10?

Comment 17

[cpratt]

I don't believe QA have a problem with fixing this in the M10 timeframe.

Comment 18

[cpratt]

Results using the 1999082016 build (latest M9) build:
Windows NT: Apprunner crashes; Talkback report generated.
Linux: Apprunner crashes.
Mac OS: Apprunner crashes.

Comment 19

[Warren Harris]

Ok, let's move it to m10.

Comment 20

[Paul MacQuiddy]

I already added to release notes.

Comment 21

[Warren Harris]

I just checked in fixes for this.

Comment 22

[Paul MacQuiddy]

http://bugzilla.mozilla.org/showattachment.cgi?attach_id=952 no longer crashes,
marking verified.

Comment 23

[leger]

Bulk move of all Necko (to be deleted component) bugs to new Networking

component.