Mike Conley (:mconley) (:⚙️) Reporter
	Description • 9 years ago

Regression: Fx-Team-Non-PGO - Tab Animation Test - WINNT 5.1 (ix) - 8.89% increase
----------------------------------------------------------------------------------
    Previous: avg 5.249 stddev 0.136 of 12 runs up to revision 01f56fe98acc
    New     : avg 5.716 stddev 0.053 of 12 runs since revision 122413f5b051
    Change  : +0.467 (8.89% / z=3.436)
    Graph   : http://mzl.la/1P2fYmO

Changeset range: http://hg.mozilla.org/integration/fx-team/pushloghtml?fromchange=01f56fe98acc&tochange=122413f5b051

Changesets:
  * http://hg.mozilla.org/integration/fx-team/rev/122413f5b051
    : Mike Conley <mconley@mozilla.com> - Bug 1150200 - QI the docShell to an nsIWebNavigation before sending canGoForward or canGoBackward in browser-child.js. r=ttaubert.
    : http://bugzilla.mozilla.org/show_bug.cgi?id=1150200

Bugs:
  * http://bugzilla.mozilla.org/show_bug.cgi?id=1150200 - Can't navigate back to previous page sometimes

	Mike Conley (:mconley) (:⚙️) Reporter
	Comment 1 • 9 years ago

Worst April Fools prank ever. :(

	Tim Taubert [:ttaubert] (inactive)
	Comment 2 • 9 years ago

That's a little hilarious, considering that bug 1137933 is MemShrink bug.

	Mike Conley (:mconley) (:⚙️) Reporter
	Comment 3 • 9 years ago

It's always a tense relationship between space and time.

	Mike Conley (:mconley) (:⚙️) Reporter
	Comment 4 • 9 years ago

First, going to ensure that bug 1150200 did introduce the regression. The graph looks pretty definitive though.

Once confirmed, I'm testing a hypothesis where we only QI the docShell to nsIWebNavigation once - we'll see if that moves the needle back.

Baseline before bug 1150200: https://treeherder.mozilla.org/#/jobs?repo=try&revision=14a12c1a160e
With patch for bug 1150200: https://treeherder.mozilla.org/#/jobs?repo=try&revision=96f9802ed6ce
With idea: https://treeherder.mozilla.org/#/jobs?repo=try&revision=efb98cc97963

	Mike Conley (:mconley) (:⚙️) Reporter
	Comment 5 • 9 years ago

So, I suspect that this was caused by something other than my patch.

Here's the compare talos between the baseline (before patch) and the patch for bug 1150200:

http://compare-talos.mattn.ca/?oldRevs=14a12c1a160e&newRev=96f9802ed6ce&server=graphs.mozilla.org&submit=true

The TART numbers really don't budge at all. So... outside influence? Perhaps something was up with the machines?

Also, a look at the graph shows that the regression has been wiped out anyway by an fx-team merge[1]:

http://graphs.mozilla.org/graph.html#tests=[[293,132,37]]&sel=1427865685672.5945,1428268431618.5405,3.2830188679245285,6&displayrange=7&datatype=geo

So... I think this is a WONTFIX. jmaher, do you concur?

[1]: http://hg.mozilla.org/integration/fx-team/rev/fbfa60dce341

	Joel Maher ( :jmaher ) (UTC -8)
	Comment 6 • 9 years ago

sold!  the graph is back to the original level.