dbaron wrote: Is the test machine using user_pref("layout.fire_onload_after_image_background_loads", true); ? It probably should be...

And maybe the reftest framework should warn if the pref is set incorrectly?

The reftest code could probably even set the pref such that it doesn't affect the profile by setting it in the default branch -- this wouldn't work if the profile explicitly set it the other way, though.

while we're talking about "nice-to-haves", it'd be great if reftest produced its own profile on startup. I could do that from outside the harness, but if it were self-contained, it would be that much easier. In any case, the xserve's had this added to its profile.

Making reftest produce its own profile would make it significantly harder to run reftests with particular prefs set -- something I need to do sometimes, e.g., to test the high DPI case.

Re comment 4: It would make it easier to run reftests in an automated way, which is the primary use-case of reftest. As this would likely not be part of the reftest chrome code, but a runtests.pl-like script, it should not make it any harder to run reftest on a particular profile. robcee, have you filed a bug on that yet?

no, I hadn't, actually. Thanks for the reminder. See bug 374458.