<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
</head>
<body text="#000000" bgcolor="#FFFFFF">
<br>
<br>
<div class="moz-cite-prefix">On 2018-11-20 8:13 AM, Dave Townsend
wrote:<br>
</div>
<blockquote type="cite"
cite="mid:CAPMxTNrZ-450jJ=gSrBfA_x-5AcvR1mLJLmCMf+vDzYrWQR1CA@mail.gmail.com">
<meta http-equiv="content-type" content="text/html; charset=UTF-8">
<div dir="ltr">
<div class="gmail_quote">
<div dir="ltr">On Tue, Nov 20, 2018 at 6:06 AM Dão Gottwald
<<a href="mailto:dgottwald@mozilla.com"
moz-do-not-send="true">dgottwald@mozilla.com</a>>
wrote:<br>
</div>
<blockquote class="gmail_quote" style="margin:0 0 0
.8ex;border-left:1px #ccc solid;padding-left:1ex">
<div dir="ltr">
<div class="m_1902225482008599589gmail-row
m_1902225482008599589gmail-messageRow
m_1902225482008599589gmail-sameAuthor
m_1902225482008599589gmail-chat
m_1902225482008599589gmail-self
m_1902225482008599589gmail-type_buffer_msg
m_1902225482008599589gmail-userParent"><span
class="m_1902225482008599589gmail-message"><span
class="m_1902225482008599589gmail-content">I suspect
this is a side effect from how pageload tests
operate: they're in the startup path and start
loading pages quicker than a user would normally be
able to. While this is an interesting benchmark,
it's probably not the right thing to optimize for
between tpaint and loading pages when the browser is
already up. How do people feel about accepting this
kind of regression, and more generally <span
class="m_1902225482008599589gmail-message"><span
class="m_1902225482008599589gmail-content">taking
startup out of the pageload benchmarks?</span></span></span></span></div>
</div>
</blockquote>
<div><br>
</div>
<div>If the pageload tests aren't measuring the right thing
then we should absolutely fix that. And I think if possible
we should do that before considering whether to accept the
regression here. Without the test changes we don't actually
know if this is causing a regression to pageload or not.</div>
</div>
</div>
</blockquote>
<br>
Including startup time in our page load test times seems like a bad
idea. We already have dedicated tests for startup time (ts_paint)
and probably 99% of real world page loads are not happening during
Firefox startup (especially because session restore is off by
default).<br>
</body>
</html>