Looking at a reflow profile, we spend quite a bit of time drawing inner box shadows on the URL bar and search bar (about 132ms and 47ms respectively). Hiding these box shadows during the transition could be a really cheap win for customization transition performance.

Going to start here...

I whipped up a patch to remove the box-shadows on the URL bar and the search input, and a comparison reflow profile saw what might be a 10ms win on painting. That's really small fries right now. I'm not going to WONTFIX this bug, but I'm going to see if I can find some bigger fish. I'll post my patch in a second.

Here's a list of visible things that have box-shadow's during the transition on Windows 7: TabsToolbar:before, nav-bar, customization-titlebar-visibility-button, customization-toolbar-visibility-button, customization-reset-button, panel-arrowcontent, PanelUI-footer, PanelUI-footer-inner

I'm curious to know how much the box-shadow that's creating the Windows Aero glass fog is playing into this. Preliminary local tests suggest it's a non-trivial amount, but I've pushed to Try to see if I can get a more definitive answer. Baseline: https://tbpl.mozilla.org/?tree=Try&rev=9d601acf25f2 Patch to remove fog: https://tbpl.mozilla.org/?tree=Try&rev=a4f1199e12d3 If taking out the fog puts a nice dent in our CART results, we might be able to use an image for the fog - at least during the customize mode transition (see bug 908067).

compare-talos: http://compare-talos.mattn.ca/?oldRevs=9d601acf25f2&newRev=a4f1199e12d3&server=graphs.mozilla.org&submit=true Breakdown for Win 7: http://compare-talos.mattn.ca/breakdown.html?oldTestIds=34130835&newTestIds=34130935&testName=cart&osName=Windows%207&server=graphs.mozilla.org I think it might be worth seeing if we can swap in the fog image during the customize mode transition.

Here's a list of things with box-shadows that includes anonymous content: "TabsToolbar:before,nav-bar,toolbarbutton-icon,toolbarbutton-icon,toolbarbutton-icon,toolbarbutton-menubutton-dropmarker:before,dropmarker-icon,toolbarbutton-icon,toolbarbutton-icon,toolbarbutton-icon,toolbarbutton-icon"

Retargeting for Windows, since this is where performance hurts the most - though I can easily see us doing this on OS X if needs be.

Ok, I've salvaged MattN's work from bug 908067, and used shorlander's newer images. Right now I'm just using the static fog image during the customize mode transition, but who knows - we might want it for the normal fog as well, since box-shadows are slow as molasses. Anyhow, here's a try push for comparing: Baseline: https://tbpl.mozilla.org/?tree=Try&rev=fa48d6ebc486 Static fog patch: https://tbpl.mozilla.org/?tree=Try&rev=28c131a7990c Compare-talos: http://compare-talos.mattn.ca/?oldRevs=fa48d6ebc486&newRev=28c131a7990c&server=graphs.mozilla.org&submit=true

Here's a distilled profile of just the customize mode enter transition (3 enters): http://people.mozilla.org/~bgirard/cleopatra/#report=8a51b9a8ecf10789eca95fd280cd11c1c0f08f46 Just so it's clear how much box-shadows are contributing here, it looks like 114 samples were gathered inside nsDisplayBoxShadowOuter::Paint, which is about 38 samples per each enter of customize mode. Using the static fog image, I got 94 samples gathered in nsDisplayBoxShadowOuter::Paint - so my original hypothesis that the static fog was responsible for the majority of the box shadow cost appears to have been incorrect.