►
From YouTube: WebPerfWG Triage call - April 23rd 2019
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
A
A
C
Yeah
one
conflict
does
May
30th
fork
money,
buddy,
mace,.
A
A
C
A
A
E
A
Next
up,
let's
talk
about
resource
timing,
so
we
have
multiple
open
issues
that
are
like
the
first
one.
Let's
talk
about
specifying
tau
for
304
responses,
that's
issue
201,
so
Nikolas
opened
this
issue
a
while
ago
and
we
discussed
it
a
while
ago
on
February
and
we
discussed
it
on
one
of
the
or
last
calls
and
said
that
we
will
try
to
see
what
the
scores
do
in
those
cases
so
talking
to
an
F
and
Catherine.
A
A
A
A
A
Not
entirely
so
it
almost
matches
it,
but
for
Korres,
if
one
of
the
redirections
in
the
redirect
chain
goes
off
of
main
origin
and
then
like
goes
cross
origin
and
then
goes
back
to
same
origin.
The
same
origin
response
at
the
end
of
the
redirect
chain
has
to
provide
a
timing,
allow
origin
to
itself
or
access
control,
allow
origin
to
itself
we're
in
Tao.
We
don't
have
that
restriction
and
we
previously
talked
about
potentially
adding
a
similar
processing
model,
but
only
as
part
of
l3.
C
C
Will
take
how
long
it
will
take
the
speck
l3,
but
some
websites
may
start
depend
on
the
current
behavior.
Keep
the
while,
probably
even
economy
had
I'm,
not
yeah.
A
A
C
A
C
B
C
I
mean
also,
it
seems
too
risky.
If
the
website
starts
depending
on
this
right,
then
I
will
be
harder
to
make
a
change
in
the
future
and
if
the
browser's
already
invoice
for
the
you
know,
where's
the
one
like
that
fits
on
vendor
who
is
probably
taking
it.
My
you
know
for
like
I'm
excited
that
my
name
and
up
the
inquiry
was
in
Guangdong
right.
We
don't
want
that
who
hasn't
implemented.
C
A
A
A
A
A
So
from
my
person,
it
would
be
significantly
simpler
to
specify
this
once
we
have
the
concept
of
tainted
origins.
Once
we
have
the
concept
of
like
it
would
be
trivial
to
implement
that
once
we
have
fetch
integration
and
would
be
significantly
more
complex
to
specify
that
before
and
I
don't
see
a
huge
value
from
a
privacy
perspective
like
Todd.
You
mentioned
that
earlier
and
it's
not
like
there
aren't
any
attack
vectors
that
I'm
aware
of
that
are
exposed
by
this.
So
this
is
not
something
that
opens
up
any
privacy
attacks
that
I'm
aware
of.
A
It
so
having
time
allow
origin
on
the
redirect
chain
is
something
that
is
useful.
Forcing
the
origin
to
include
itself
in
a
timing.
Allow
origin
after
a
cross
origin.
Redirect
is
not
something
that
includes
any
privacy,
related
problems
and
I'm
aware
so
the
cross
origin
host
will
still
has
to
include
timing.
Allow
origin,
but
the
cores
model
also
forces
the
same
origin
host
to
include
that,
like
include
that
opt
in
to
itself.
A
From
my
perspective,
the
main
point
of
this
is
to
be
consistent
with
Korres
and
in
order
for
us
to
be
able
to
say
that
timing
allow
origin
is
magically
a
subset
of
chords
and
therefore
a
course.
Opt-In
is
also
timing.
Allow
origin
opt
in
that
all
relies
on
a
lot
of
l3
work
and
a
lot
of
fetch
integrations.
So
I
don't
see
a
point
in
doing
that
before.
C
C
C
B
C
C
You
know
when
we
fixed
one
system,
I
mean
that's
the
Ducks
that
comes
on
right,
like
if
we
don't
change
the
spectra.
Now
somebody
will
come
in
later
and
a
Firefox
working
in
a
fixed
implementation
to
match
the
current
ltos
back,
which
will
say
you
know
in
they
say
if
the
redirect
they
ends
up
in
the
same
origin,
we
don't
check
right,
whereas
if
we
did
update
aspect,
chances
are,
if
we
were
updated
to,
you
know
check
that.
So
it's
more
that,
like
it,
you
know
well
implementation
updates
that
you
know
fixes
their
box.
C
A
B
G
Well,
certainly,
yes,
I
thought
you
were
going
to
ask
me
beyond
CEO
and
I
would
have
said
yes
to
that
as
well.
I
mean
it's
a
corner
case
at
the
end
of
the
day.
As
long
as
everyone
agrees,
the
spec
is
correct.
It
just
bugs
in
implementation.
I
think
that
that
I
can
I
can
call
it
the
director
2004
to
let
us
move
forward
with
that
and
well.
D
D
Haven't
my
gym
is
on
the
calls
well,
but
I
haven't
thought
through
it
very
much,
but
I
actually
just
made.
I
just
opened
a
bug
for
Bugzilla
to
say
that
we
need
to
implement
this
because
it's
currently
broken
so
I
see
any
reason
why
we
couldn't
be
able
to
align,
but
I've
been
taking
more
notes
than
substantively
thinking
about
the
competition.
So
I
might
need
to
go
back
and
look
at
this
again
at
offline.
You.
A
Okay
and
okay
and
I
would
also
like
to
like,
from
a
spec
perspective,
I
think
that
I
already
have
draft
in
an
earlier
commit
that
then
we
decided
not
to
go
in
so
this
back
piece
is
yeah.
It's
probably
not
gonna,
be
that
complex,
okay,
so.
A
A
That
would
mean
that
resource
timing,
entries
are
likely
to,
like
developers,
are
likely
to
see
more
resource
timing
entries
that
include
that
include
rent
requests,
so
double
the
requests.
Double
the
image
request
that
they're
currently
seeing
so
I
open
an
issue
open
2:05
on
that
and
Illya
commented
that
this
seems
to
be
a
subset
of
issue
number
21,
so
handling
of
multi
request
fetches,
which
makes
sense.
So
if
everyone
are
cool
with
that
definition,
I
I
will
close
this
in
close
to
a
five,
and
then
we
can
focus
on
21
is
part
of
l3.
F
A
A
F
A
D
D
A
A
G
Say
yes,
yes,
it's
useful,
like
the
tool
is
reminding
me
that
I'm,
a
muted,
I
I
think
it's
it's
depend
on
whether
those
tests
are
corner
cases
or
not,
I
think
if
they
are
corner
cases-
and
we
all
agree
that
yes
corner
cases
and
we
should
move
forward
without
having
to
a
permutation
for
this,
we
should
go
ahead.
On
the
other
hand,
if
some
of
those
tests,
we
believe
the
Magna
nosy
actor
is
substantive,
then
we
cannot
just
corner
cases
and
we
should
we
should
block
on
them.
That's
my
philosophy.
A
B
D
D
B
B
D
C
A
Okay,
thanks.
A
C
C
It's
just
like
it,
at
least
for
next
two
weeks
days
ago,
with
little
components
face
to
face
coming
up
and
all
the
stuff
is
happening,
and
they
know
that
DC's
getting
closest
a
lot
of
things
happening
around
here
so
but
yeah,
okay,
you
know
yeah
I
know
they
it
I
would
have
to
do
it.
So
it's
just
you
know,
trying
to
put
in
the
agenda,
but
it's
3d
lots
of
stuff
happening,
but
so
yeah.
Sorry,
okay,.
A
Okay,
so
let's
move
on
to
high
resolution
time
so
can
performance
up
now,
timers
be
frozen
for
background
tabs,
etc.
So
there
is
currently
a
chrome
behavior.
That
is
arguably
a
bug
where
chrome
is
freezing
performance
now,
for
some
like,
for
example,
in
background,
add
tabs
as
part
of
the
tab,
freezing
or
tab
throttling,
and
there
are
multiple
people
myself
included
in
chrome.
That
argued
that
this
is
a
bug
and
we
should
change
our
behavior
and
Ben.
A
C
A
C
H
Advocate
here
the
case
where
I'm
trying
to
measure
how
long
my
function
takes
to
execute
and
I
started,
executing
my
function
and
then
might
have
this
background.
There
was
it
computers
put
to
sleep
or
something
like
that,
and
then
eventually
it
wakes
up
again
and
my
function.
It
finishes
executing
if
we
don't
pause
performance
thought
now,
then
it
looks
like
my
function
took
a
huge
amount
of
time
to
execute
words.
If
we
do
pause
it,
it's
more
reflective
of
the
actual
time
it
took
for
the
method,
but.
A
C
H
Will
you
put
the
computer
to
sleep
I,
don't
know
the
details
here
and
I'm,
not
I,
think
probably
either
way
it
probably
doesn't
make
sense
to
freeze
timers
I,
just
try
to
think
of
what
the
potential
benefits
would
be
to
freezing
them,
and
there
might
be
some
case
along
these
lines
where
it
is
actually
nicer
from
the
web
developers
perspective.
If
they're
trying
to
measure
at
the
duration
of
a
function,
yeah.
B
Edge
has
the
behavior
of
not
freezing
times
when
we
freeze
background
temps
and
that's
been
our
behavior,
since
we
first
did
freezing
in
Windows
8,
and
we
took
the
position
that
developers
should
be
able
to
measure
time
and
if
they
need
to
measure
the
other
signals
disambiguate
time,
then
they
should
use
things
like
the
background
visibility.
The
other
events,
as
as
the
signal
to
break
down
this
long
tree,
that
makes
it.
A
D
A
D
A
Not
I'm
not
sure
I
have
a
like.
Maybe
then
has
a
relevant
test
issue
here
and
he
yeah,
so
it's
his.
He
has
a
bend,
has
a
webpage
on
the
related.
So
from
that
issue,
oh
not
even
been
from
that
issue.
There
is
a
jsfiddle
like
if
you
follow
the
chrome
bug,
there's
a
jsfiddle
where
you
can
see
the
issue
happening,
so
that
would
be
one
way
of
testing
whether
like
what
Firefox
does
in
that
scenario,
thank.
A
A
Maybe
she
we
should
add
a
note,
but
I
think
my
intent
from
this
discussion
was
mostly
to
understand
what
you
know
that
we're
not
missing
the
mark
here
and
people
think
that
freezing
is
the
right
behavior
and
it
seems
like
they
don't.
Maybe
I
think
that
normatively
there's
nothing
to
change.
Maybe
it's
worthwhile
to
add
a
note.
Yeah.
A
A
Another
high-resolution
time
issue:
issue
64
as
part
of
the
ping
review
or
following
the
ping
review.
One
of
the
ping
members
opened
this
issue,
suggesting
that
we
will
mandate
gating
performance
now
behind
existing
permission
prompts
so
only
allow
performance
DUP
now
to
be
enabled
if
fullscreen
is
enabled
if
MIDI
devices
are
enabled
or
other
various
current
permissions.
I
replied
on
that
thread.
Basically
like
from
my
perspective.
A
This
will
not
help
to
mitigate
specter,
which
I
think
is
what
he's
aiming
at.
The
current
implementations
have
already
significantly
limited
performance
that
now,
as
a
result
of
Spector,
some
of
them
to
do
a
millisecond
level,
others
to
sub
millisecond,
but
still
significantly
clamped,
as
well
as
jittered
and
I,
also
had
multiple
so
folks
from
Chrome,
as
well
as
folks
from
Firefox
saying
that
it
doesn't
really
make
sense
to
go
that
route
as
there
are
other
other
timers.
A
C
C
A
Yeah
yeah,
I,
agree
and
I
stated
like
in
my
response.
Basically
all
the
other
attempts
that
I
know
that
we
asked
a
you
were
involved
in,
such
as
cross
origin,
opener
policy
and
cross-origin
resource
policy
and
Korb
and
other
like.
There
are
many
other
mitigations
for
that
that
were
like
paths
where
we
think
we
can
actually
make
a
difference
and
blocking
timers
locking
you
know
relatively
coarse.
Timers
is
not
one
of
those
paths
and.
B
C
Think
if
you
were
to
more
concretely
implement
such
a
proposal,
alright,
what
we're
gonna
do
is
we
want
to
keep
the
performance
not
now
function
it
just
like
Precision's
of
it
changes
depending
on
whether
things
are
long
I
think
we
did
discuss
something
like
that.
While
we
were
having
cross-browser
Spectre
mitigations,
where
would
allow
with
some
permissions
who
came
in
to
be
able
to
use
Pokemon
so
now,
stuff,
I,
don't
think
we
ended
up
doing
that,
because,
just
because
users,
allow
doesn't
mean
your
websites
can
just
go
random
resource
inspector
attack
or
like
yakked.
A
D
D
A
A
A
I
have
a
PR
that
I
put
together.
That
replaces
the
unix
time
to
actual
ACMA
262
definitions
of
time
and
the
clock
drift
part
seems
seems
reasonable
to
leave
it,
as
is
as
it's
not
really
referencing.
Anything
that
is
normative
or
impacting
implementations,
so
essentially,
I
would
love
reviews
on
that.
A
A
Which
rings
like
which
brings
me
to
the
next
question?
Assuming
that
we
closed
everything
above
and
we
fix
Chrome's
issues,
can
we
move
forward
with
high
resolution
time?
Can
we
ship
and
I
know
that
we
wanted
to
do
that
for
a
tea
pack
and
Philippe
wanted
to
have
a
another
security
review
related
to
like
from
security
folks
related
to
Spectre
threats?
That
review
has
passed
so
any
objections
to
moving
forward
with
HR
time.
G
A
A
Awesome,
okay,
and
in
the
five
minutes
we
have
left.
Let's
see
what
so,
let's
pick
the
like
skip
the
big
complex,
be
ours
and
that
we
won't
have
time
to
review.
I
think
that
we
need
to
republish
L
to
the
out
based
on
the
level
two
branch,
because,
like
right
now,
the
published
version
issue
124
on
performance
timeline,
I
think
that
the
currently
published
version
includes
the
buffered
flag
and
that
was
removed
from
the
level
two
branch.
So
we
need
to
republish
so
Philippe
is
that
something
you'd
be
able
to
help
with?
Yes,.
A
F
A
F
A
We're
good
to
go
on
that
and
in
the
3
minutes
we
have
left
so
there's
an
ongoing
discussion
related
to
timing,
allow
origin
in
the
context
of
navigation
timing.
So
this
is
an
issue
that
Boris
opened
issue
104
on
the
navigation
timing,
spec.
This
is
an
issue
that
Boris
opened
and
I
think
that,
like
I
gave
this
issue
some
thought
today
and
it
seems
like
for
navigation
timing.
The
role
of
timing.
Allow
origin
is
a
bit
inverse
so
for
resource
timing.
A
A
You
know
the
the
the
way
that
I
say
that
it
should
be
opting
in
is
the
way
that
it
should
indeed
opt
in.
So
I
love
some.
You
know
some
reviews
on
that
particular
issue
and
folks
telling
me
that
I'm
wrong
or
right.
A
A
B
A
He
wasn't
happy
with
with
the
tower
definition,
but
maybe
aligning
that
definition
with
cores
will
make
him
happier
and
more
open
to
do
such
reviews
real
skate
from
your
end.
Is
there
someone
that
should
look
at
this
from
a
security.
C
So
I
think
the
main
thing
is
that
the
document
could
be
created
by
script
right.
That's
that
me,
like
case
in
which
there's
no
resource,
so
in
that
case
an
HTM
inspector.
A
few
cases
where
the
origins
documents
origins
is
defined,
so
I
think
what
we
would
have
to
do
is,
and
so
there's
a
call,
also
concept
of
the.
C
A
A
And
it
will
be
fine,
no
no
extra
information
will
be
exposed,
but
for
cases
where
there
is
a
previous
document,
I
think
that
we
need
to
make
sure
that
the
that
document
has
opted
in
and
in
cases
where
there
is
a
redirect
chain.
I
think
that
the
redirection
need
to
opt
in
to
the
current
document
so
to
the
end
of
the
redirect
rather
than
to
its
beginning-
and
this
is
currently
undefined
so,
like
that's.