1396569 - Figure out tresize regression in Quantum Render builds caused by the unified search bar

Status: RESOLVED DUPLICATE of bug 1416651

(Core :: Graphics, defect, P3)

Description

[:Paolo Amadini]

While bug 1387416 causes definite Talos improvements, it also causes a regression on "tresize" (optimized, e10s) on the Quantum Render build:

https://treeherder.mozilla.org/perf.html#/compare?originalProject=try&originalRevision=72efa487393444a6ae6c0836cbb7cb994ad00248&newProject=try&newRevision=41bd6918b93fe704504faa9aa461f3f1b6a18ba1&framework=1

(Tick "Show only important changes" to filter it.)

Gijs noted the following in bug 1387416 comment 26:

"FWIW, part of the perf impact here might be the flexible spacers which are being added by CUI, which may cause a layout rejig. Ditto for some late icon loading (cf. bug 1394914, bug 1374224). I expect the fact that there are fewer (high) flex items here would influence the degree to which things resize from those operations, which in turn might mean more stuff gets invalidated, which could potentially have negative perf implications for rendering. But that's just me guessing - maybe Milan can confirm if this is at all plausible?"

"I don't think that should stop this patch landing, we should instead try to work out how to have flexible spacers in the markup so CUI doesn't need to add them dynamically (and add minimum sizes to icons in the other bugs I mentioned)."

Comment 1

[:Gijs (out for now; he/him)]

(In reply to :Paolo Amadini [Away] from comment #0)
> "I don't think that should stop this patch landing, we should instead try to
> work out how to have flexible spacers in the markup so CUI doesn't need to
> add them dynamically (and add minimum sizes to icons in the other bugs I
> mentioned)."

We've done all of this, and in any case we didn't need to block on this, so marking fix-optional for 57.

Note that the general trend of QR tresize over the last few weeks is up, and although that's probably not solely due to this change, I'm wondering if we should be concerned. Milan probably knows more about this?

Comment 2

[Jeff Muizelaar [:jrmuizel]]

This isn't webrender related.

Comment 3

[Kartikaya Gupta (email:[e-mail])]

Sounds like it's a regression in QuantumRender which is WR-related. But given the age of this bug we can just roll it into bug 1416651.