on April 17th we have a TART regression on Linux64 (3.22%) and WinXP (3.64%) from the landing of this changeset: http://hg.mozilla.org/integration/mozilla-inbound/rev/7340e1b5c7e1 Here is a graph for winxp: http://graphs.mozilla.org/graph.html#tests=[[293,131,37]]&sel=1397665160000,1397837960000&displayrange=7&datatype=running here is a view on tbpl to ensure it was this changeset and retriggers to ensure there isn't noise: https://tbpl.mozilla.org/?tree=Mozilla-Inbound&fromchange=76e1244494ef&tochange=f854b5d978b5&jobname=Windows%20XP%2032-bit%20mozilla-inbound%20talos%20svgr Here is more information about the test: https://wiki.mozilla.org/Buildbot/Talos/Tests#TART.2FCART

Terrence, can you look at this?

This is a backout of the same patch that caused a throughput regression last week. There is a direct tradeoff here between throughput and latency: you can't have both. I think we should take this regression instead of the other: the code that's getting removed is quick ugly and we're doing strictly less work without it.