Here's the stack: nextStep@chrome://global/content/bindings/progressmeter.xml:83:19 FrameRequestCallback*nextStep@chrome://global/content/bindings/progressmeter.xml:105:15 FrameRequestCallback*nextStep@chrome://global/content/bindings/progressmeter.xml:105:15 FrameRequestCallback*_init@chrome://global/content/bindings/progressmeter.xml:109:11 progressmeter-undetermined_XBL_Constructor@chrome://global/content/bindings/progressmeter.xml:113:1 PopupNotifications_showPanel@resource://gre/modules/PopupNotifications.jsm:932:9 PopupNotifications_update@resource://gre/modules/PopupNotifications.jsm:1097:9 _onButtonEvent@resource://gre/modules/PopupNotifications.jsm:1480:5 oncommand@chrome://browser/content/browser.xul:1:1

The layout flush comes from the call to "stack.boxObject.width" here: https://dxr.mozilla.org/mozilla-central/rev/7d2e89fb92331d7e4296391213c1e63db628e046/toolkit/content/widgets/progressmeter.xml#82 ...in the progressmeter-undetermined "_init" method. That .width accessor does indeed flush layout (unsurprisingly), via a call from BoxObject::Width() to BoxObject::GetOffsetRect() to BoxObject::GetFrame(), which calls GetPresShell() with aFlushLayout==true: https://dxr.mozilla.org/mozilla-central/rev/7d2e89fb92331d7e4296391213c1e63db628e046/layout/xul/BoxObject.cpp#110

(In reply to Daniel Holbert [:dholbert] from comment #1) > The layout flush comes from the call to "stack.boxObject.width" here: [...] > ...in the progressmeter-undetermined "_init" method. Sorry, I meant it's in the <method name="_init"> tag -- but really the flush happens every refresh driver tick (!) via a requestAnimationFrame callback called "nextStep"

(In reply to Daniel Holbert [:dholbert] from comment #2) > the flush happens every refresh driver tick (!) via a requestAnimationFrame callback > called "nextStep" Ugh :-(. But this binding seems to only be referenced on Linux: http://searchfox.org/mozilla-central/search?q=progressmeter-undetermined

(In reply to Florian Quèze [:florian] [:flo] (away until August 7th) from comment #3) > (In reply to Daniel Holbert [:dholbert] from comment #2) > > the flush happens every refresh driver tick (!) via a requestAnimationFrame callback > > called "nextStep" > > Ugh :-(. Yeah, that seemed particularly terrible to me, too... > But this binding seems to only be referenced on Linux: > http://searchfox.org/mozilla-central/search?q=progressmeter-undetermined Huh. Interesting. I wonder why that is...

Apparently bug 658829 switched to native progress meters on other platforms.

fxperf:p3 because Linux-only, I would have made it a p2 if other platforms were affected.