1464742 - Code size regressed by porting InspectorUtils::GetCSSValuesForProperty to use Servo

Status: NEW

(Core :: CSS Parsing and Computation, defect, P3)

Description

[Xidorn Quan [:xidorn] UTC+11]

See discussions in bug 1434130 comment 102 and after.

Comment 1

[Eric Rahm [:erahm]]

(In reply to Xidorn Quan [:xidorn] UTC+10 from comment #108)
> Filed bug 1464742 for that. From the graph it doesn't seem to be a very
> significant regression, and also I don't have a great idea on how it can be
> fixed, so I gave it a P3 for now. We can continue discussion there.

Unfortunately after we merged ICU's data into libxul (bug 1445524) it really skewed our numbers making sizeable regressions look "smaller". Focusing on the raw numbers makes more sense -- a 212k regression is sizeable. Zooming in [1] makes it a little more clear.

From a Fission point of view I'm mainly concerned about increases in non-shareable data, so .data, .data.rel.ro, .bss etc. We're tracking those now [2] but don't have historical data back this far.

In this case there's a 33,872 byte increase in .data.rel.ro. That's not huge, but it would be nice to understand what's going on and get it lower if possible.

[1] https://treeherder.mozilla.org/perf.html#/graphs?timerange=7776000&series=autoland,1338582,1,2&series=mozilla-inbound,1299711,1,2&zoom=1524928657766.8357,1525000662347.2383,129956760.51003662,130707928.29188234&selected=autoland,1338582,331976,460069505,2
[2] https://treeherder.mozilla.org/perf.html#/graphs?timerange=7776000&series=mozilla-inbound,1686770,1,2

Comment 2

[Xidorn Quan [:xidorn] UTC+11]

(In reply to Eric Rahm [:erahm] (please no mozreview requests) from comment #1)
> In this case there's a 33,872 byte increase in .data.rel.ro. That's not
> huge, but it would be nice to understand what's going on and get it lower if
> possible.

I think every data added in that change should fully be shareable, so it would be very useful to know what ends up being in .data.rel.ro there. Maybe trait function pointers?