Johnny changed things so that the document's onload event now waits for all the image onload events to fire before it fires. The event was already waiting for the images to load; just not for their onload events to fire. This added 30ms (about 3%) to the Tp time on btek, because the image onload events are fired asynchronously (which has to be done because libpr0n, unlike necko, will respond with data synchronously) and the Tp time is measured as the time from load start to load event firing. So the question is, how much of that total 1020ms time is spent waiting on things like this after the page is actually loaded? How much time does it take for the page to look loaded to the user (note that this could well be over 1020ms depending on when reflow happens in relation to onload events).
It's too bad that as soon as people come up with a measurable substitute for whatever it is they care about they start treating it as more important than the real thing. I'm reminded of IQ tests and SATs.Posted by bzbarsky at May 10, 2003 1:00 PM