In bug 1300490 we saw a performance regression after landing the new icon code. * Only nytimes spiked but not the twitter test - Why? * The old and new icon code both run asynchronously, so it shouldn't affect page loading too much. Why does it? * To avoid performance issues we intentionally load icons only *after* the page has loaded. So why does it affect our performance tests? To answer those questions and debug the performance regression I'll have to look closer at the code and how all components interact - and at what points in time we log the throbber messages for the performance measurements.

No change (mozilla-central): https://treeherder.mozilla.org/#/jobs?repo=try&revision=495c97047fa2 Icon loading disabled: https://treeherder.mozilla.org/#/jobs?repo=try&revision=79674e20e7e1

Disabled loading and building icon requests: https://treeherder.mozilla.org/#/jobs?repo=try&revision=00e33c2154df88d60748608f0276a0b30e07d193

I measured the current code (no change), loading disabled and code completely removed for local-nytimes and remote-nytimes on try. I couldn't find any noticeable difference. In fact often the tests with no code change ran faster - again pointing into the direction that this doesn't affect page loading much. In addition to that I looked at the code and the icon load is triggered a little bit before logging the throbber messages. They listen to different events and it doesn't make sense to align them so that icon loading is happening after the throbber messages for performance measurements. Right now there doesn't seem to be a measurable improvement to make.