Confirm
313
edits
Changes
no edit summary
= next meeting =
([https://fx-team.etherpad.mozilla.org/australis-meeting-notes next meeting has not yet occurred])
=8/29/2013=
* Milestone 9, now
** http://24.media.tumblr.com/tumblr_mdjbws2fiY1r0bv4uo1_400.gif
* Perf regression and other landing blockers
** bug 889758 (tpaint regression)
*** Closed, but being monitored.
** bug 880611 (ts_paint regression)
*** We're at 1% here, but vladan thinks we should understand where the cost is coming from before giving up. CustomizableUI.initialize only accounts for 1/4 of that regression. It's possible the rest comes from TabsInTitlebar (that'll be measured on start-up next).
** bug 902024 (TART regression)
*** TART has landed in the talos test suite, and we have data coming in. Big thanks to jmaher and avih for making that happen
** MattN created a spreadsheet displaying TART regression data: https://docs.google.com/spreadsheet/ccc?key=0Asj8iLTl0K0UdGJzVnVYYkJTSDREbndRWWFTLWNyS1E&usp=sharing
*** We're kicking ass on Windows 7 and 8 on the talos hardware
*** We're regressing on every other platform. Talk to mconley for the graphs and breakdown. We're going to focus our efforts on Win XP first.
*** avih and MattN are feeding local low-powered machine data into MattN's custom graph server. We're monitoring performance there, and will focus there after we neutralize on the talos hardware.
**** graph: http://bit.ly/15if1Dr
**** compare-talos: http://compare-talos.mattn.ca/?oldRevs=e42dce3209da&newRev=33cd07a5f9a2&server=graphs.mattn.ca&submit=true
*** Graphics is resource constrained, and a number of bugs we've isolated as bottlenecks are in their department. Might get some BenWa cycles (1-2 hours per day maybe) and maybe Bas, once he's done TRIBE (tomorrow).
**** Bug 907544 - Pass the D3DSurface9 down into Cairo so that it can release the DC and LockRect to get at the bits
**** Bug 907546 - Avoid DC readback on windows at all cost.
**** Bug 909807 - When drawing text to an opaque ARGB32 surface with GDI we can avoid fallback
*** tn has looked at a Win XP TART profile, and besides the tabs toolbar repaint (https://bugzilla.mozilla.org/show_bug.cgi?id=908796) he's not seeing anything obvious that we're doing wrong.
**** He suggests we land the toolbar repaint workaround and generate a new profile for him to look at.
** ts/tpaint regression data: https://docs.google.com/spreadsheet/ccc?key=0Asj8iLTl0K0UdDcxemJEWVVXWTlKek1wR09iS2FsOXc#gid=0
* Bug tracking
** 8/22 to 8/29:
*** P1: +6 (Mainly TART bugs)
*** P2: +1
*** P3: ----
*** P4: +1
*** P5: +19
*** Poth: -42 (Triaged bugs and placed them in correct priorities)
** Overall we went from tracking 200 bugs on Aug 22 to 185 bugs on Aug 29, a net decrease of 15 bugs while also adding TART bugs. In total, 20 bugs were fixed within the past week.
** Cross-team coordination
*** Great response and output from the perf team (avih and vladan) in helping with TART (and jmaher too!)
*** Seth says he has a 90% patch for bug 764299 - Cache rasterized SVG images some of the time
**** This should get us good perf (tpaint, TART, and maybe ts_paint) wins on all platforms
***** We should check with him that it affects clip-paths
*** Gfx team is over booked with b2g work
**** P1 bugs tracked for gfx team:
***** 907546 - Avoid DC readback on windows at all cost.
***** 907544 - Pass the D3DSurface9 down into Cairo so that it can release the DC and LockRect to get at the bits
****** benwa surprised us today with a WIP patch for this bug
***** 907426 - CreateSimilarSurface allocates a DDB win32 surface
***** 907542 - Use DIBSections for texture memory on Windows Vista+
*** Add-on SDK team has quite a few P1 & P2 bugs that are assigned to them, need an update from them on the state of their work.
**** P1 bugs tracked for add-on sdk team:
***** bug 910190 - Permanent Jetpack orange on UX: TEST-UNEXPECTED-FAIL | tests/test-ui-button.test button are not in private windows | is null | followed by other errors
****** Gijs patched this: https://github.com/mozilla/addon-sdk/pull/1223
***** bug 906690 - Permanent Jetpack orange on UX: TEST-UNEXPECTED-FAIL | tests/test-ui-sidebar.testButtonIconSet | the icon is set properly in panel - icon32.png vs. icon16.png
****** This is (hopefully) fixed in the SDK, will be uplifted in the next 24 hours, from what I heard on #jetpack today.
****** However, that was based on it landing on fx-team today. They're on WW (in Paris, I think?) and none are online right now, so I'm skeptical this will actually happen (last time we heard this it also took considerably more time than this)
* Roundtable
** [mconley] Check out http://aris-at-mozilla.blogspot.ca/
*** I've emailed the author with tips on how to register a new CustomizableUI toolbar, because his blog said this wasn't possible.
*** Hopefully I haven't just opened myself up to a firehose of hate and rage - but I figure, if we can work with this author, everybody wins.
* Action Items (Last Week)
** (mconley) Push on getting TART landed on test infrastructure
*** Currently that's 2 weeks, but can we make it faster?
**** Done.
** (mconley) Talk to perf team on how we can set a target for us to shoot for. We need a treshold to aim for, otherwise we won't know when we're done. This is absolutely critical.
*** Done - neutralize perf on talos infrastructure on key subtests that avih and I boiled down to.
** (mconley) Before the test lands on talos, how do we measure so that we can work effectively while waiting for the test to land
*** Done - MattN's spreadsheet is badass. We also have compare-talos working for us, and Datazilla will let us overlay data next week
*** Can we be sure that these numbers scale across platforms? Will a relative improvement / regression on one platform match to another platform?
*** In the event that we show no regression against m-c when the test lands, Australis aside, what's the plan for detecting regressions on low-end hardware? Do we get some low-end hardware into the test machine pool?
** (dolske) mull over what/if we should do a broader post/update on current Australis status
* Action Items (New)
** (MattN) look for worst regressions on subtests across platforms
=8/22/2013=
