►
From YouTube: WebPerfWG call 2023 08 17 - Bimodal performance timings
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
So
I'll
just
introduce
myself
briefly,
since
this
is
my
first
time
in
this
forum,
so
Mike
Jackson,
I'm,
an
engineer
on
the
Microsoft
Edge
team
and
so
I've
been
working
with
Amir
and
Andy
for
this
bimodal
performance
timings
in
web
app
and
I
think
this
was
previously
discussed
in
this
forum
where
we
were
trying
to.
A
We
were
talking
about
adding
a
extending
the
navigation
type
to
adding
a
user
agent
launch
like
value
to
the
to
the
enum,
and
we
collected
a
bunch
of
feedback
both
in
this
forum
and
in
other
forums.
You
know
this
can
happen
for
all
navigation
types,
not
just
like
a
regular
navigate
but
can
happen.
A
We
can
be
in
situations
where
timing's
thrown
off.
You
know
in
back
forwards,
navigations
and
other
things
like
that.
It
would
be
impractical
to
add
new
email
members
for
all
navigation
types.
The
user
agent
can
be
under
load
at
other
times,
not
just
during
startup,
and
there
was
some
difficulty
expressed
in
this
forum
around
mapping
browser
launch
to
a
HTML
concept.
A
So
with
all
that
in
mind,
we
went
back
and
sort
of
thought
about
it,
a
little
bit
more
and
and
tried
to
restate
our
goals,
which
is
to
allow
developers
to
eliminate
outlier
metrics
when
the
user
agent
is
in
a
in
a
non-optimal
performance.
State-
and
you
know,
one
of
the
things
that
we
looked
at
here
was
like.
When
we
went
and
collected
data
we
were
looking
at,
you
know:
where
were
we
seeing
spikes
and
when
the
user
agent
was
under
load
right?
A
You
know
we'd,
see
Spikes
all
along
the
timeline
from
the
start
time,
all
the
way
through
to
you
know,
request
start
or
between
DNS
lookup
start
and
DNS,
lookup
and,
and
so,
and
so
that
sort
of
then
led
us
to
the
The
Proposal
that
we're
I'm
sharing
today
and
then.
The
other
thing
that
we
wanted
to
do
too
was
avoid
this
being
sort
of
like
used
as
a
proxy
to
determine
what
the
user
agent
was.
The
user
agent
performance
state
is
in
real
time.
A
So
so
that
led
us
to
this
proposal,
which
is
to
instead
of
extending
the
navigation
type
we'd,
add
a
new
property
called
system.
Entropy
there's
been
feedback
on
the
name,
so
I'm
happy
to
work
through
naming,
updates
and
and
do
some
bike
shedding
offline.
A
But
the
idea
here
being
that
for
top
level,
navigations
we'd
return
higher
normal
values
for
other
navigations.
We
just
returned
empty
string
and
that
you
know
what
we
with
the
data
that
we've
sort
of
started
collecting.
You
know
I
think
we're
gonna
Focus.
We
want
to
focus
on
what
user
agent
launches
today,
where,
where
that
concept
is
sort
of
just
something
that
we
defer
to
each
user
agent
to
decide
what
that
meant
for
them.
A
But
then
you
know
in
the
future.
We
could
expand
it
to
any
navigation
where
the
user
agent
is
under
load
and
I've
put
in
a
little
code
snippet
here
for
how
we
expect
people
to
to
sort
of
use
it
to
use
the
API.
A
And
then
the
other
thing
I
wanted
to
mention
in
this
forum
I
think
looking
at
the
notes
for
this
meeting
earlier,
there
were
some
questions
around
how
Chrome
custom,
tabs
or
webview
might
work,
and
so
Chrome
custom
tabs
on
Android
I
think
you
know
it
should
just
work
the
same
way
that
it
would
work
in
the
browser.
A
Webview
webview
doesn't
really
have
a
signal
to
say
whether
or
not
the
host
app
is
during
startup.
So
you
know,
The
Proposal
here
is
at
least
we
just
always
return
normal,
but
there
may
be
situa
like
the
webview
team
might
find
it
helpful
to
expose
an
API
at
some
point
to
allow
the
host
app
to
sort
of
participate
in
that
conversation
and
then
with
restored
tabs.
A
Those
aren't
typically
loaded
until
the
user
selects
them,
and
so
it
really
just
sort
of
in
what
I
think
what
I'm
proposing
here
is
that,
basically,
it
just
behaves
the
same
way
as
a
regular
tab
like
if
the
user
was
able
to
start
the
start.
The
user
agent
click
on
a
tab
stop
typing.
Before
the
user
agent
was
finished
loading
then
you
know
it
would
return
High.
Otherwise,
it
would
just
return
normal
in
that
in
that
state
and
then
the
other
question
that
I've
heard
a
lot
of
is
like
well.
A
Why
can't
we
just
use
compute
pressure
as
I
like
being
able
to
go.
Do
this,
since
you
want
to
know
when
the
user
agent
is
under
under
load
and
really
the
the
issue
there
is
that
since
it's
using
JavaScript
right,
there's
a
whole
bunch
of
events
that
just
you
can't
really
capture
like
by
the
time
the
the
developer
is
able
to
register
for
the
pressure
Observer.