►
From YouTube: WebPerfWG F2F TPAC 2019 - day 2 early afternoon - issues
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
B
C
A
So
the
first
issue,
which
is
we
doesn't
need
a
whole
lot
of
discussion
and
I,
have
a
are
for
it,
basically
we're
aligning
how
processing,
with
course
to
make
sure
that
Tao
is
a
strict
subset.
Of
course,
this
doesn't
yet
I
mean
that
will
allow
will
assume
that
timing
is
allowed.
One
course
is
allowed.
That
could
be
step
Everlast
that
will
probably
help
treat
step,
but
initially
we
just
wanted
along
the
definition
so
that,
if
we
have
equivalent
timing,
allow
origin,
headers
and
course
X
Istanbul,
average
and
headers.
A
A
A
A
That's
possible
and
ie,
so
I
don't
think
we
need
to
litigate
this
now.
This
change
I'm
saying
that
we
want
to
do
anyway
regarding
the
cache
ability
exposure.
Maybe
it's
okay,
maybe
we
can
limp
I,
don't
know,
add
some
limits
around
death
but
exposed
most
other
facts.
So
you
know
just
block
cache
ability,
exposure
if
cores
is
allowed,
but
how
isn't
but
I
know
we
can
we
can
debate
it
in
the
future,
but
yeah.
That's
where
we
want
to
go
because
course
is
enabled
on
way
more
resources
than
tangela
origin
can
be
assumption.
E
A
A
D
And
so
I
think
the
key
point
that
we're
is
that
it's
not
a
slam-dunk
too
many
people
in
the
room
that
it
is
the
right
thing
to
switch
effectively
to
course,
as
allowing
town
that
insuring
that
is
a
clean,
non
privacy
of
leaking
so
yeah,
the
fine
ship
I
think
ensuring
that
is
not
going
to
leak
information
is
is
what
we
want
to
happen
to
make
that
change.
It's
not
the
opposite,
where
we're
gonna
make
the
change,
and
it's
on
the
people
to
them,
of
course,
but
the
question
is
like:
are
there
people.
A
F
F
Of
course,
the
other
concern
that
we
have,
which
I
don't
disagree
with
you
on,
that
we
should
actually
look
at
it
and
not
accept
people
at
their
word
is
that
in
certain
cases
we
may
add
into
parts
of
the
spec
that,
if
cores
such
from
such
and
such
then
that
implies
tab
in
certain
cases
where
we
explicitly
said
it,
and
we
don't
think
it
in
there
may
be
a
problem
that
we
don't
know
we
have
to
get
to
America.
Okay,
those
are
two
separate.
F
A
A
F
F
A
A
Seem
worthwhile
to
discuss
in
like
as
a
high-level
overview,
so
one
issue
is
linked.
Preload
headers
right
now
are
being
used
to
imply
push
semantics
and
are
being
used
by
servers
to
perform
h2
push,
and
we
have
a
request
from
I
believe
HLS
folks
from
Apple
and
Akamai
to
add
new
attributes
on
that
head
like
on
link
preload
headers.
That
will
imply
that
only
a
certain
HTTP
range
needs
to
be
pushed.
It's
unclear.
A
What?
How
will
this
impact
the
pricing
model
in
the
browser,
if
at
all,
and
if
it
won't
and
servers,
will
not
remove
those
range
requests?
Then
this
will
likely
result
in
performance
degradations
because,
like
the
server
will
push
a
range,
but
then
the
browser
will
request
the
full
the
full
resource.
At
the
same
time,
if
it
will
impact
the
browser
crossing
model,
then
we
will
mean
to
actually
specify
how
preload
now
can
do
range
requests,
which
will
we.
G
C
A
C
G
G
A
A
A
A
A
B
A
A
G
H
A
A
C
G
G
A
D
A
A
No
I
mean
there
is
the
priority.
Hints
proposal
is
a
different
date
on
the
same
problem
which
isn't
currently
shipping
anywhere
because
effectively,
we
didn't
see
it
being
very
effective.
The
suspicion
is
that
it's
due
to
h2
server
preferences
asian
implementations,
so
maybe
once
we'll
have
a
better
organization
scheme
for
each
three
variants
is
actually
worthwhile,
but
for
now
I.
G
G
G
A
At
least
if
he
wants
range
requests
on
preloads,
that
potentially
is
like
it
seems
like
the
legitimate
use
case,
but
it
wasn't
phrased,
as
such
I
mean
if
you
can
preload
like.
If
you
can
issue
fetch
with
requests,
you
should
similarly
be
able
to
yeah
it's
a
load,
but
I
would
love
to
hear
from
him
if
this
is
indeed
the
case,
and
they
really
want
like
they
want.
Preload
supports
on
top
of
yeah
Bush.
One
ask.
A
A
F
A
I
believe
so
as
well
not
come
first
hundred
percent
sure
why
this
test
case
should
produce
one
in
one
according
to
the
current
spec,
but
I'm,
basically
trusting
that
you
know
he
knows
what
he's
talking
about.
So
we
should
probably
hang
it
off
of
the
yeah,
in
which
case
we
will
need
to
move
the
I'm.
Pretty
sure
will
you
can
move
the
definition
of
type
the
enum
into
HTML
itself
and
then
the
funny
there
yeah
does
that's
like.
F
Basically,
what
we're
saying
here
is
you
back
forward
navigation
type
for
the
navigation
entry
happens
or,
basically
you
do
a
history,
traversal
operation,
100%
of
the
time,
according
to
this
back
yeah,
if
you
follow
all
of
this,
you
go
through
that
all
the
time
and
I
think
that
cheese
I
think
that,
ultimately,
what
I
agree
with
him
on
was
his
second
paragraph
eyes
are
his
one
two
through
the
paragraph
is
that
we
probably
wanted
to
stay
back
forward.
This
traversed
the
history
by
a
delta
yeah,
not
just
traverse
the
history.
F
And
I
don't
know
what
we
should
do
with
that.
That
seems
like
more
of
an
open
question,
but
I
do
think
it
should
be
first,
the
history
by
Adela,
okay,
that
that
part
series
cousin,
because
every
juror
sister
might
doubt
that
say
that
that
will
at
least
get
us
out
of
the
loop
where
we're
always
doing
history,
devotion,
yeah,.
J
A
F
So
I
think
he's
saying
if
we
do
a
history
load,
if
we
do
a
traverse
by
Delta
and
we
end
up
with
a
redirect
as
a
result
of
that,
and
then
we
end
up
in
yet
a
third
page
that
we've
never
seen
because
each
and
every
act
is
different.
Perhaps
what
is
that
is
that
history,
navigation
or
is
that
a
new
navigation
I
believe
that
to
be
the
question
exam
yeah,
then
there's
the
related
question
of
what,
if,
instead
of
a
redirect
HTTP
redirect
you're
doing
just
manually
change,
looking
rather
Drive.
F
D
F
D
The
other
place
this
actually
impacts,
at
least
on
the
other.
The
link
to
the
HTML
issue
is
a
serviceworker
which
also
has
a
on
the
fetch
request.
Maybe
I
don't
know
how
service
worker
gets
the
same
state
search
order,
some
objet
would
steam
well.
He
says
he
basically
asked
their
question
on
the
GE
and
says
that
this
information
is
relevant
for
a
service
for
her
and
the
mini
of
your
time.
D
F
D
A
An
aligned
response
at
the
same
time
I
think
the
intention
for
navigation
time
and
like
why
do
we
have
time?
Okay
and
I
think
the
answer
to
that
is
to
be
able
to
distinguish
different
types
of
navigation
that
have
different
performance
characteristics
and
from
that
perspective,
if
it's
a
back
forward
navigation
that
then
redirected
to
a
brand
new
page
that
isn't
in
the
VF
cache,
then
it
shouldn't
be
a
back
like
right.
The
time
should
be
different
from
that,
like
from
the
use
case,
lens
I,
think.
K
D
A
D
A
D
K
H
K
B
K
D
D
J
H
A
A
D
K
F
D
Now
is
newsreel
in
opposition,
and
in
this
case,
if
you're
redirecting
it's
clearly
neither
or
written
or
history,
so
the
it
seems
like
the
service
worker
answer
is
simple.
Theoretically,
always
together
it's
easier
to
answer.
If
you
know
what
the
service
worker
wants
to
do,
which
is
hey,
it's
now
being
navigated
to
a
component
thing.
D
D
F
H
F
I
think
the
issue
would
be
you
have
a
page
that
does
some
conditional
location
dynatrap
in
your
history
and
you
back
forward
to
it
and
when
you
back
forward
to
it
it
in
some
cases
will
look
at
your
page
revenues
at
all.
So
in
the
cases
it
won't
or
before
that
location,
not
a
giraffe
actually
happens,
the
navigation
side
would
be
back
forward.
A
A
L
D
I
read
that
right.
Well,
probably
like
I,
don't
know
where
it
is.
We
probably
need
to
review
the
language
of
the
spec
and
confirm
the
thing,
to
course,
doesn't
usually
all
finishes
when
the
spec.
That
is
also
true.
But
yes,
as
you
looked
at
this
back
to
see
if
it
covered
the
case,
and
it
didn't
see
the
language
that
covered
the
cake,
so
it
might
need
yeah.
It
probably
needs
clarification
at
least
yeah.
That's
my
Guinness
I
think
the
redirects
scenario
is
probably
similar
in
that.
If
you
look
at
a
spec,
it
says
hey.
D
If
the
navigation
is
an
8,
blah
blah
and
then
the
question
as
well
as
it
starts
as
a
blonde
ends
in
the
block
you
know,
and
so
the
it's
probably
about
being
crisp
about
where
and
then
in
the
navigation,
and
that
was
the
thing
he
was
also
getting
out
with
this
language
in
the
issue
which
is
like,
and
actually
probably
navigation,
doesn't
really
talk
about
this.
So
if
you,
because
this
concept
is
only
for
Service
Worker,
an
omniscient
yeah.
A
A
D
A
But
in
that
world
of
TT
caches
prefetch
as
implement.
That
today
enables
to
bypass
a
bunch
of
protections
because
it
is
fetches
that
are
triggered
by
potentially
one
origin
and
used.
So
cross-origin
prefetch
is
of
eventually
bypassing
those
protections,
because
those
are
Fetchit
that
are
triggered
by
one
origin
and
then
consumed
by
another.
A
So
we
had
a
discussion
last
year
to
basically
modify
prefetch
in
ways
that
it's
only
relevant
for
a
navigation
only
relevant
for
navigation
requests,
hongyi,
reusable
by
navigation
requests,
emitted
credentials
or
same
origin
credentials.
It
doesn't
really
know
refer,
many
will
be
directs
bypass,
the
serviceworker,
a
bunch
of
various
restrictions
which
they
are
not
like.
It's
been
ongoing
discussion,
but
eventually
Anna
asked
four
days
ago
a
very
relevant
question
where
he
wonders
if
a
new
keyword
wouldn't
be
appropriate,
given
that
it
it
would
be
something
like
the
new
and
revamp
and
privacy.
A
C
A
C
C
A
Lots
of
existing
discussions,
recommendations
and
users
be
done,
and
this
is
gonna
be
very
different.
Yes
and
I
also
suspect
that
it
would
be
very
hard
to
explain
to
people
a
why
the
old
thing
is
no
longer
working
and
be
why
they
have
to
add
all
these.
You
know
as
document
and
press
origin
and
yeah.
A
A
Keyword,
yes,
they
will
have
to
jump
through
at
least
two
hoops
in
order
to
make
it
work.
I
know
that
with
pre-loaded
fonts
just
up
the
cross
origin,
one
is
a
50%
of
the
people,
thought
that
you
could
just
restrict
prefetch
to
imply
on
the
distance,
but
I
think
the
problem,
but
that
is
that
Felicity.
A
A
A
J
A
C
K
D
A
A
I
A
It's
not
actually
a
problem
so
that
I
think
I
misunderstood
when
I
said.
This
is
something
that
you
need
to
discuss.
Maybe
we
can
drop
multiple
contacts
as
tight
because
it
gives
you
information.
You
already
have
elsewhere,
that's
the
issue,
but
it's
not
something
that
seems
right
worthy
of
discussion.
That's
just
generally.
Whenever
long
tests.