See https://areweslimyet.com/mobile, and zoom in to circa 30 jan 2013. There is a 4.53 mb regression in startup resident memory usage; 3.87 mb in explicit memory usage. The regression range (from inbound builds) is:
Sorry, misread the graph data. 835178 is off the hook.
Bug 835503 removes some debug-only code, so I doubt it is at fault.
The diff between the two about:memory reports shows that the largest difference comes from:
2911848 explicit/workers (added)
2911848 explicit/workers/workers() (added)
2911848 explicit/workers/workers()/worker(resource:\\gre\modules\osfile\osfile_async_worker.js, 0x59496400) (added)
Looks like bug 828201 is the winner.
Should we back out the offending cset?
What does "2911848 explicit/workers (added)" mean exactly?
Created attachment 709064 [details]
It means that when comparing the "before" about:memory dump to the "after" about:memory dump, the total in the "explicit/workers" subtree went up by 2911848 bytes. The "(added)" means that the explicit/workers tree didn't even exist in the "before" dump. I've attached the full diff (as generated by my tool at https://github.com/staktrace/awsy-armv6/tree/master/about-memory-differ operating on the raw json files.
There was a talos RSS regression at about the same time: bug 836429.
A nit: when we file an AWSY regression bug can we please include a screenshot?
I backed out the offending change in https://hg.mozilla.org/integration/mozilla-inbound/rev/8728de36d4a8
My AWSY builder is catching up on some backlog so it hasn't run the backout yet; I will resolve this bug once that happens. Note that the talos RSS regression in bug 836429 was fixed by the backout.
Created attachment 709212 [details]
The AWSY website isn't showing the data yet, but the backout cset was processed, and it brought the resident memory down from 81399808 bytes to 77836288 bytes.