Load the page from the URL field. One sees several stages of loading. At first the block has red bits in it, but as the page loads, they disappear and when the page is fully loaded the square is solidly green. If one runs a reftest on this page, the snap has red in the box. It seems that the reftest snap is being made when one layer of the page is loaded, as it does catch the box, but there seem to be other types of loading that still need to happen. Does reftest need to be listening for another event before it takes the snap?

I would guess that this is due to bug 253851

Now that you mention it, I think I recall seeing one of the current reftests that uses background-images failing intermittently. We could set the hidden pref. that makes onload fire after background-images are loaded, but I'm not sure if this would be appropriate, plus, the reftest doesn't have a way to set preferences currently.

Running reftest with the "layout.fire_onload_after_image_background_loads" preference set might be the way to go here, no?

(In reply to comment #3) > Running reftest with the "layout.fire_onload_after_image_background_loads" > preference set might be the way to go here, no? Looks like we'll just be switching that pref by default (see bug 253851). We should make sure to set it for reftest if we end up reverting that change, though.

Sounds like bug 371866 was marked fixed a little too eagerly, although maybe this should just be duped against bug 374458.