75576 - imgRequestProxy doesn't need DummyChannel

Status: RESOLVED FIXED

(Core :: Graphics: ImageLib, defect)

Description

[Chris Waterson]

I think that the DummyChannel stuff it 1) not necessary and 2) not being managed
properly, so would propose the following patch. N.B. that I had to fix some bad
assumptions in the doc loader.

This fixes several bugs that I've been fighting in the layout regression tests
where the doc load finishes too early b/c the dummy channel had no load group.

Attaching a patch.

Comment 1

[Chris Waterson]

Attachment: get rid of dummy channel use in imgRequestProxy

Comment 2

[Chris Waterson]

I think that this probably depended on darin's recent necko API changes, so it's
not obvious that the code could've been written this way even a week ago.

Comment 3

[Chris Waterson]

Attachment: with fix to LoadImageWithChannel()

Comment 4

[Chris Waterson]

Hmm, one thing I'm noticing is that the chrome stuff (e.g., scrollbar thumb
images) are now being added to the load group. Not sure if this is correct.

Comment 5

[Stuart Parmenter]

ah.  yes... this is how i had done it originally, but since you couldn't add
nsIRequest objects that weren't nsIChannel's to loadgroups (without things
crashing), i had to change it back.

Comment 6

[Stuart Parmenter]

r=pavlov on the docloader changes... i've merged your patch into my tree.

i had to remove/change:
+  NS_PRECONDITION(mLoadGroup, "recieved >1 OnStopRequest()");
+  if (! mLoadGroup)
+    return NS_ERROR_UNEXPECTED;

a) remove the precondition.  it is legal for it to receive multiple onstops 
(multipart/x-mixed-replace)
b) return NS_OK instead of error

other than that it looks good.  thanks

Comment 7

[Stuart Parmenter]

i'll land this stuff when i land the rest of the changes in my tree today i hope

Comment 8

[Tom Everingham]

[imagelib] keyword added to status

Comment 9

[Chris Waterson]

nsDocLoader changes checked in.

Comment 10

[Stuart Parmenter]

fixed