281467 - Local nonexistent images mean page never finishes loading

Status: RESOLVED FIXED

(Core :: Graphics: ImageLib, defect)

Description

[Boris Zbarsky [:bzbarsky]]

BUILD: 2005-01-25-06 nightly or latest trunk nightly

STEPS TO REPRODUCE:

1)  Create a local HTML file with one line in it:

  <img src="/mozilla-banner.gif">

2)  Load this file in Mozilla.

ACTUAL RESULTS: throbber never stops

EXPECTED RESULTS: throbber stops

NOTES:  If this is done via HTTP (so AsyncOpen() on the channel doesn't fail
like it does for file:// in this case), the throbber stops.  If the image has an
alt attribute (so we replace it with alt text), the throbber stops.

Is there anything I can do to usefully log loadgroups and see what types of
requests are added to them?

Comment 1

[Christian :Biesinger (don't email me, ping me on IRC)]

yeah, loadgroups use nspr logging:
http://lxr.mozilla.org/seamonkey/source/netwerk/base/src/nsLoadGroup.cpp#135

Comment 2

[Boris Zbarsky [:bzbarsky]]

This is an imagelib bug...

Comment 3

[Boris Zbarsky [:bzbarsky]]

Attachment: Patch

The problem is that we stick an imgRequest for the image in the cache, then
fail to open the channel, but leave the imgRequest in the cache.  Then the next
time we try to load the image (either if the same page is loaded again, or if
the same URL is present twice in the page, which was what I was testing), we
get it from cache, add a listener to it, and.... wait.	Since the imgRequest
has not gotten OnStartRequest yet, it assumes it's still waiting on the data,
so we end up with the imgRequestProxys in the loadgroup forever.

Canceling the imgRequest does the right thing here -- removes it from cache.

Comment 4

[Boris Zbarsky [:bzbarsky]]

This is actually preventing the layout regression tests from running, so it'd be
good to get this in ASAP... Darin, any idea when you can look at this?

Comment 5

[Christian :Biesinger (don't email me, ping me on IRC)]

hmm, that patch reminds me of bug 202369, esp the last patch there... (this one
sounds like a better patch, but the problem seems similar)

Comment 6

[Boris Zbarsky [:bzbarsky]]

Yeah, that last thing there is trying to address this same issue.  I prefer my
patch too.  ;)

Comment 7

[Boris Zbarsky [:bzbarsky]]

Comment on attachment 173796 [details] [diff] [review]
Patch

Could this please be approved for 1.8b1?  This is pretty safe, and makes it
possible to do things like run the layout regression tests again...

Comment 8

[Asa Dotzler [:asa]]

Comment on attachment 173796 [details] [diff] [review]
Patch

a=asa for checkin to 1.8b

Comment 9

[Boris Zbarsky [:bzbarsky]]

Fixed.