►
From YouTube: TPAC 2019 WebRTC meeting @Thursday part 3
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
B
B
C
This
is
this:
is
us
I
think
I
just
wanted
to
touch
upon
the
fact
that
you
know
the
progress
has
been
made
there
on
the
web.
Rtc
stats
implementation
compared
to
last
year.
I
won't
harp
on
the
fact
that
kite
already
presented
a
bunch
of
things
that
shows
progress
has
been
made
but
like
clearly,
if
you
look
at
the
slide,
you
see
numbers
increasing,
you
can
look
at
the
actual
one
at
web.
Rtc
stats,
dot,
call
stat
I.
A
C
Think
in
terms
of
implement
I
think
the
first
thing
we
need
to
do
is
that
we
started
this
process
of
re
like
fixing
some
of
these
stats,
so
things
that
were
in
some
other
places
are
now
returning
to
inbound
or
remote,
inbound,
RTP
or
inbound
RTP,
local
or
outbound.
Rtp
I
think
that
changes
finally
been
made
a
couple
of
weeks
ago
and
I
think
we're
now
trying
to
stabilize
that
over
the
next
few
few
weeks.
C
So
we'll
need
extra
eyes
on
the
spec,
because
stats
have
moved
around
in
the
last
year
we
made
new
dictionaries
and
we
moved
them
back,
partly
because
of
simulcast,
partly
because
of
pax,
we'll
come
to
some
of
those
things
later,
I
think
implementation,
wise,
that's
the
place.
We
need
to
work
the
most
like
thinks
that,
like
the
inbound
and
the
odd
word,
RTP
is
that's
implementation,
wise
I
would
say
that's
the
place
you
have
to
work
in
terms
of
specification.
I
think
we
are
in
a
good
place.
C
C
Reason
we
call
dot.
That
aspect
was
because
I
think
all
the
stats
that
are
now
being
reported.
We
need
to
start
to
validate
if
they
image
jimin's
or
some
word
Reese
and
we've
not
made
much
progress
on
this
since
last
year,
even
though
I
think
several
people
volunteered
to
to
do
this,
including
us
and
we've,
not
found
the
time
to
actually
put
this
together.
So
I
believe
that
there
is
some
community
work
required
to
do
this.
C
Yeah,
we
call
them
verification,
validation,
so
semantic
verification,
I,
believe
you
mean
that
they
report
a
number.
The
number
exists,
that's
what
verify
does
but
validate
is
like
at
fifty
milliseconds
delay,
and
you
see
what
relays
actually
reported
on
the
link
like
it
should
be
reasonably
close
to
50
or
it
shouldn't
be.
Outrage
is
like
five
hundred
I.
Think
that's
what
we're
going
for
the
other
thing
that
we
like
simple
things
like
counters,
going
up.
G
H
H
H
B
I
J
B
Steak
is
business
so
originally
we
had
the
track
stacks
and
receiving
side
remote,
inbound,
RTP
and
tracks
s
again
model
around
media
stream
tracks,
but
that's
they
aren't
really
track.
We
were
never
really
track,
salad
like
it
makes
track
encode
their
decoders
and
their
receivers.
That's
and
then
we
put
an
RTP
a
rate.
That's
a
female
comic
on
any
minor
people
against
between
coding
and
decoding
stuff.
So
we
had
all
this
all
this
good,
but
then
we
have
the
next
slide
peach.
This
is
the
my
plan
transceivers.
B
And
if
you
replace
track,
we
we
made
a
copy
of
the
visionary.
We
reset
all
the
concerts
and
yeah.
So
so
the
sender
might
accomplish
keeps
increasing,
but
the
track
counters
you
get
a
new
trial
date.
That's
my
counter
starts
at
zero.
Again,
thanks,
that's
that's
multiple
dictionaries
and
so
yeah.
That's
problem
sold
right.
The
next
thanks
like
problem
case.
No
one
ever
implemented
the
sender
and
receiver.
That's
like
we're
still
called
track,
sets
in
chrome
and
there's
still
a
mix
of
reverse
to
the
mixture.
So
senator.
B
K
B
Decoders,
that's
it
a
mix
and
then
the
breaking
point
from
trying
to
change.
This
again
was
one
of
my
single
casters
because
yeah
her
encoders,
that's
on
the
account
on
RTP.
Well,
how
do
you
express,
although
we're
doing
promised
Bob
use
Express
having
the
same
same
striking
senator,
but
you
have
multiple
layers,
so
it's
a
next
slide.
This
is
how
things
look
today.
We
just
move
things
around
basically,
so
now
the
outbound
party
P,
is
the
encoder
and
Rd
beats
next,
that's
any
sense,
but
single
cast
cases
have
x
mark
at
the
objects.
B
This
plunker
your
mayor,
even
though
men
tender
and
that
removed
from
the
Senate
there
are
senators.
That's
we
lose
stuff.
It
was
specific
to
the
camera,
my
sources,
the
source,
the
inputs
we
distract
I
would
put
that
evening
and
we
get
her
stationary
and
then
the
other
leave
outside
unless
things
change
was
the
mood
swings
from
receiver
to
in
done.
B
Our
duty
to
reflect,
what's
going
on
in
this
case,
is
simpler,
because
you
always
only
have
one
in
one
stream
soon
there
one
second
decoder,
stats
and
the
output
of
the
decoder
is
the
same
thing
as
the
access
of
the
destination.
There
is
a
big
mistake,
so
how
things
should
look?
You
practice
anything
simpler,
open
tracks,
experts,
but
it's
the
spec
allows
you
to
express
simulcast
and
revision.
You
proof
all
right.
Let's
move
over
to
issues
and
there's
someone
ask
questions
of
us
all.
These
onions,
yeah.
A
B
B
We
never
documented
the
fact
that
you
have
a
track
stats
on
the
receiving
side
as
well,
because
well
you
you
never
change
attachments
on
the
receiving
side,
so
it's
entirely
sufficient
to
just
have
the
receiver
tracks
test.
So
that
would
be
exactly
the
same
as
the
proposal
of
this
slide
is
add
to
the
obsolete
section
documentary
the
fact
that
you
can
also
have
a
receiver
track
stats,
because
this
is
a
promise
we
should
document
what
implementations
do.
M
B
J
B
On
look
at
the
next
slide
should
we
have
transceivers?
That's
so
thanks,
Alonzo
and
transceivers
best.
This
will
allow
us
to
document
the
main
interface
that
make
sense
included
hours
to
expose
the
direction
and
current
direction,
especially
the
kind
of
direction,
will
be
useful
to
handwrite
analysis
doesn't
exist.
You
can
just
ensure
our
my
counters
increasing
or
not,
and
if
we
have,
as
we
could
also
have
a
reference
to
codec
IDs,
you
guess
that's
most
of
the
output
of
this
function
will
just
be
a
set
of
codec
objects
because
they
listen
free
transaction.
K
B
You
mean
you
need
me,
because
you
need
to
be
able
to
correlate
your
your
stats
objects
with
the
M
sections,
my
proposal,
so
that
will
be
proposal
to
there's.
Also
no
way
to
tell
you
what
you
convict,
stats,
update,
belong
to,
which
and
section
right.
So
you
could
you
could
you?
Could
you
say:
hey,
we
don't
want
direction,
we
don't
print
direction.
So
therefore
we
put
the
paint
on
the
sender.
B
That's
all
make
sense,
that's
perfectly
valid
if
it's
trade,
the
picture
from
last
picture
and
that's
not
the
same
as
the
underlying
model
and
also
when
you
do
get
stats
by
the
tie.
It's
an
asynchronous
operations
by
the
time
leads
and
when
I
expect
them
and
correlate
commits
and
stuff,
maybe
maybe
I
call.
So
let
me
give
you
a
different
result.
So
I
think
I
think
this
is
preferences.
It's
very
easy
to
just
a
snapshot.
P
P
B
J
B
Right
now
you
have
RTP,
that's
objects,
but
you
don't
know
if
they're
active
or
not,
because
maybe
you
don't
have
a
direction
you're
like
okay
can
do
is
increasing.
Is
that
because
the
network
is
batteries
them
because
it's
not
being
used,
so
it's
tough
to
adapt
of
a
little.
You
know
why
it's
not
a
little
bit
right
there.
This.
This
would
tell
you
that
and
I
think
the
implementation
effort
and
adding
direction
in
grant
direction,
and
it's
no.
It
doesn't
it's
nothing.
Much
well.
I'm.
K
O
B
Other
things,
if
you
don't
know
her
debugging,
because
right,
like
sometimes
I've,
seen
bugs
with
base
conditions
where
you
negotiated
before
or
after
some
event
of
the
application,
and
so
it's
maybe
if
you
see
status
s,
oh
the
current
direction,
is
this
that
you
know
might
as
well.
Why
don't
I
do
this?
Well,
that's
aggression
direction
up
or
not,
because
no.
C
Think,
just
just
to
interject
I
think
there
are
two
parts
to
this.
First
is:
do
we
need
a
dictionary
for
this,
like
the
transceiver
dictionary,
which
would
carry
a
some
information
which
includes
me
I
think
that
the
main
reason
to
have
it
and
then
the
second
thing
is
like
once
we
have
the
dictionary
like
do
we
need
the
direction
current
direction
of
what
else?
I?
C
Think
that,
though
these
tube
can
be
decoupled,
I
think
the
first
thing
is
transceiver
stats
makes
sense
because
at
least
gives
one
location
where
you
could
get
a
bunch
of
information
that
you
can
use
to
correlate
other
data
in
other
stand,
objects
and
the
simplest
one
would
be
like
to
take.
The
mid
switch
is
the
most
needed
one
and
put
it
in
the
right
place,
and
then
we
find.
B
D
O
A
C
Learned
like
the
ship
for
like
not
using
getstats
for
debugging,
has
sailed
like
people
use
it
for
debugging.
We
exist
because
they'd
need
it
for
debugging
I.
Don't
think
like
saying
that
this
information
should
not
be
there.
There
is
no
other
mechanism
today,
like
people
use,
get
stats,
call
get
stats
frequently
infrequently
mainly
for
debugging.
E
K
E
C
K
P
B
K
B
P
F
D
B
C
J
B
K
B
You
know
actually
going
on,
we
moved
more
or
less
everything
into
the
RTP
sense
objects,
which
means
we
now
have
a
track:
stats
object,
which
only
contains
stats
members
that
are
present,
l
squared
think
they
did
their
document.
It
involves
a
big
section.
So
if
you
were
to
implement
a
spec
from
scratch
today,
it's
next
a
table
more
or
less
empty.
I
have
a
follow
up,
sign
my
empty
bit
and
you
don't
need
it
anymore.
B
Track
that
would
end
the
current
tracks,
a
table
to
create
a
new
track
set
audience,
so
you
would
have
an
advanced
correlate
with
your
place
track.
Why
is
that
good?
Well,
because
if
you're
pulling
stats
and
you're
you're
trying
to
figure
out
what's
going
on,
you
have
an
event
that
you
can
correlate
to
it,
which
that's
right
like
if
the
resolution
changes
and
yeah
you'll
get
a
different
bitrate
and
it's
not
because
your
network
is
bad.
It's
because
you
change
the
track
so
well.
B
It
would
be
very
straight
to
keep
stats
events
around
if
it's
awful
and
the
purpose
is
to
find
these
nets
object,
doesn't
have
any
purpose
on
its
own
for
one
as
well.
The
other
question
about
whether
or
not
he
should
remove
the
checks-
that's
not
big,
but
assuming
he
needs
me
remove
it.
So
what
do
we
do?
My
Thomas
and-
and
one
thing
to
do-
is
we
don't
need
all
stats
ended
or
removal?
That's
ended
or
we
remove
tracks.
B
That's
object,
one
of
the
two
in
either
case
the
application
cancer
doesn't
have
an
event
to
listen
to,
but
we
don't
need
to
do
anything.
Are
you
do
because
the
application
knows
if
they
call
the
place
track,
so
they
can
just
note,
don't
get
sent
for
or
even
if
you
don't
do
anything
I
show
whenever
you
call
guess.
K
K
C
So
my
challenge
is
that
we
are
not
the
application
we
sit
besides.
The
application
and
our
attempts
to
override
peer
connection
can
be
very
fragile
because
some
other
people
may
override
the
peer
connection.
So
we
have
no
chance
of
doing
anything
except
for
adding
event
listeners,
and
so
that's
one
of
the
reasons
why
we
like
the
on
on
events
right
so
like
we
have
on
track
on
nice
connection,
state
change
there.
It
allows
us
to
poll,
get
stats
exactly
at
points
when
things
change
within
the
app
and,
of
course
we
pull
them.
C
We
can
choose
to
pull
it
more
often
or
less
often,
and
we
have
algorithms
so
that.
But
we
really
when
the
application
changes
something,
then
we
need
to
be
able
to
measure
at
those
points
and
I
think
the
on
stats
ended
was
the
compromise
exactly
for
that,
because
it
would
end
a
track
and
we
would
fire
on
stats,
ended
and
would
get
all
the
stats
for
the
track.
That
was
just
going
to
be
replaced
by
a
new
one
right.
C
So
we
would
get
everything
that
we
needed
and
we
would
get
a
new
snapshot
for
the
old
and
the
new
at
the
same
time,
if
so
now,
with
the
tracks
being
almost
empty
like
removing.
It
means
that
on
stats-
and
it
would
not
fire
because
there
are
no
tracks
that
are
ending
and
there's
no
track
dictionary.
So
there's
no
reason
to
fire
and
we
now
back
to
the
same
conundrum
as
we
had
in
2016-17
that
now,
when
people
call
replace
tack,
we
will
not
have
a
chance
to
know
what
happened.
H
C
But
like
shimming
replaced
track
means
that
someone
else
chims
it
and
we
can
be
good
actors,
but
they
may
not
be.
That
means
we
would
assume
that
we
shim
did
and
everything
works
okay,
but
someone
shim
did
before
us
and
is
not
a
good
actor
and
does
not
allow
us
to
call
her
play
like
listen
to
replace
track
and
that's
the
the
biggest
problem
we
have
like.
H
C
H
B
C
Yeah
actually
like
on
stats
on
stats,
ended
solve
that
problem
right
because
on
stats-
and
it
would
give
me
the
track
and
I
call
I
can
call
get
stats
after
that
thing
is
called
so
at
least
I
have
like
the
snapshot
of
the
last
state,
and
then
I
have
a
new
state
which
would
be
soon
after
which
I'll
call
get
stats
myself.
So
that's
why
I
like
both
on
stats
and
we
could
do
that.
Get
stats.
B
Then
I
think
option
two
is
nice
to
is
instead
of
having
in
all
sense
ended
event,
you
have
a
non
replaced,
Jacqueline,
which
you
know
that
you
might
say
that
that's
weird
to
have
an
event
for
an
operation
with
you
because
well,
there's
presidents
for
this.
You
already
have
this.
It's
called
long
track.
K
B
B
B
P
C
G
C
Like
I
guess,
then
my
pushback
would
be
that
like
then
I
would
want
to
track
stats.
Like
then
I
would
just
say
like.
Why
would
we
keep
track
stats
and
not
obsolete
them?
Because
then
you
just
carry
the
double
information
and
I
have
whatever
I
want
then
like
we've
already
like
maybe
Firefox
is
not
implemented
it,
but
it
is
there.
Currently,
my.
C
Replace
track,
we've
done
a
lot
of
analysis
on
this
and
we
found
that
we're
basically
forcing
our
application
to
pull
get
stats
like
at
very
high
frequency
to
know
something
that
we
should
have
known
by
default.
Like
we
know
when
candidates
change,
we
know
when
Iced
restart
happens.
We
know
so
many
things,
because
there
are
triggers
that
that
get
exposed
but
replace
track.
C
If,
if
my
source
is
changing,
you
should
be
able
to
detect
it
and
this
today,
because
it's
not
implemented
is
already
quite
difficult
and
we
don't
have
a
very
good
answer
for
it
and
we
punted
it
to
the
fact
that
has
not
been
implemented
by
anyone.
But
if
you
take
it
out,
then,
basically
we're
doing
a
disservice,
because
people
are
people
will
demand
it,
and
it's
basically
saying
that.
C
H
E
K
Q
K
D
D
A
A
B
B
B
That
mixes
customer
this
slide
is
proving
that
point,
but
either
either
it's
already
moved
away
or
there
exists
something
similar.
Well,
it's
something
that
really
should
exist
somewhere
else
to
go
through
them.
You
saw
Riley,
that's
already
aggressively
that
one
case,
and
it's
not
applicable
indicated
over
this.
We
don't
need
for
that
track.
Identifier
for.
K
K
B
B
So
it's
ten
see
if
the
track
is
ended,
since
media
source
is
using
track,
sense,
object,
moving,
that's
where
we
should
have
put
and
then
regardless,
so
that
we
have
coverage
for
that,
the
in
Bachman
pace
and
the
other
and
whether
the
tracks
ended
or
not,
has
more
to
do
with
whether
or
not
to
negotiate
something.
So
we
should
not
to
cover
this
transient
apparel
erection,
but
it
was
pushback
her
that
so
the
worst
case
is
we
and
in
bond
market
redone
tendon
and
would
say
whether
or
not
the
RTP
students
is
still
alive
again.
B
B
I
B
No
stuff
yeah,
but
then
again,
there's
no
details.
I
think
time
is
all
very
prescient
in
both
in
the
remaining
thing,
these
things
are
is
try
or
Athene.
I
only
refers
to
pick
some
featurettes
risk
remedy,
CPC
again,
I
think
we
really
do
want
priority.
This
is
an
encoding
parameter
solution
for
that.
There's
a
lot
of
specials,
but
we're
decoding,
parameter,
stats,
care,
nation
being.
B
K
N
D
We
have
set
up
in
weeds.
Where
said:
okay,
something
coming
in
Roma
from
from
a
reservation:
that's
probably
going
to
need
to
be
shot,
let's
sharpen
it
up
or
something
is
coming
in
from
the
camera,
heads,
probably
a
picture,
and
it
probably
needs
to
have
smooth
movements.
This
works
reasonably
well.
It
compels
someone
connected
PC
via
an
HDMI
cable.
D
N
D
D
D
D
Complicated
work,
happily
not
the
activate
button
vocabulary
to
mark
the
video
tracks
might
actually
use
it
and
want
to
use
is
not
a
complex
for
saying
it's
kind
of
video.
Here's
an
absolute
to
do
something
about
it.
This
led
me
to
think
so.
This
might
actually
makes
sense
to
establish
the
vocabulary,
is
dictionary
values
and
then
have
all
the
specs
able
to
refer.
Yes,
we're
using
that
vocabulary
to
describe
the
video
and
then
in
the
spec
is
probably
true.
You
don't
have
this.
D
The
other
specs
tried
to
put
in
rules
for
saying
here's
exactly
what
you
do
in
order
to
you
know
doesn't
make
that
make
that
work.
So
that's
kind
of
different
direction,
but
I
haven't
looked
at
those
prices.
I
had
the
conversation,
so
that's
not
worth
anything.
I
wanted
to
do
that.
That's
part
of
stage
setting.
A
Okay,
so
this
is
issue
22:48
news
originally
posted
over
tcp
sea,
but
it's
not
related
to
this
discussion
and
it's
a
discussion
of
degradation
preference.
The
first
thing
to
note
is
that
you
know
W
BTS.
We
only
set
test
the
ability
to
say
and
get
the
value,
so
you
don't
actually
really
test
the
effect
of
it
and,
unlike
some
other
attributes
that
we
have
like
max
framerate,
it's
not
that
easy
to
think
about
how
you
would
actually
test
the
event.
A
So
so
that's
one
thing
to
note:
is
we
really
in
terms
of
our
testing,
we're
not
really
testing
that
it
does
anything
in
particular?
The
second
thing
to
note
is
that
we've
had
actually
against
the
omen.
Is
it
correct
to
say
this
is
only
been
implemented
in
one
browser
which
is
edge,
I,
think
I,
think
that
might
be
true
carnage,
but
in
that
browser
it's
implemented
widely
differently
than
people
who
are
thinking
of
implementing
it
at
other
browsers.
A
Content
hint
and
took
the
prefer
frame
rate
want
us
an
equivalent
emotion
and
that
this
was
actually
at
least
as
I
understand,
are
very
different
from
what
is
done
in
Chrome,
which
is
that
there
is
a
very
clear
distinction,
as
I
understand.
Far
between
that.
N
A
A
G
So
that's
the
first
thing,
crumbs
positions
that
we
want
to
have
any
relation
preference,
the
way
that
describe
particular
that
Gospels
level.
The
problem
is
currently
we
default
to
maintain
program
management
and
so
changing
the
default
to
balanced.
We
read
risky
because
it
the
degrade
performance
of
different
applications,
so
we
don't
probably
want
to
do
that,
and
also
the
balanced
equation
default
needs
to
be
true
to
Matt
will
get
a
better
resource
so.
G
The
default
to
know
necessarily
advanced
is
probably
good.
That's
not
good
exchange,
but
it's
not
well
it's
working
for
next.
So
that's
why
it
hasn't
done
and
we
need
to
get
into
the
transition
phase
to
really
expose
it
properly.
So
it's
a
bit
of
work.
We
had
a
world
apologies,
no,
it's
something
that
rethinking
go
to.
G
A
B
D
Then
danger
danger
this
was
it
wouldn't
be
possible
to
say
okay
company
in
this
over
there
to
get
them
coming
in
and
if
you
see
a
competent,
you
can
do
the
following
actions.
Now,
let's
overhead
you
by
the
user,
which
wouldn't
say
prince,
prepare
solution
and
detail
different
pre-made
emotion,
men
have
the
definition
of
this
nation
of
competent
over
here
and
have
the
definition
of
the
actions
over
here.
So.
K
Yes,
there
are
two
issues:
the
first
one
I
split
em
up
the
first
one
is
redundancy
and
lack
of
normative
steps,
which
is,
I
think,
don't
just
on
segregation.
Preference
as
well
is
that
inst
don't
really
specifying
their
descriptive
of
the
trying
to
describe
what
an
input
is,
instead
of
it
explain
to
browsers
what
they
should
do
with
that
information,
so,
unfortunately,
it
I
feel
we
feel
that
unspecified
behavior
figures,
the
dominant
browser
implementation
and
then
everyone
else
has
to
reengineer.
K
Reverse
engineer
laugh
and
it's
also
a
missed
opportunity
to
make
that
make
things
number
can
be
testable.
For
instance,
on
the
this
is
easier
to
see
on
the
audio
constraints
conserve,
the
redundancy
is
clear
there.
The
spec
mentions
echo
cancellation,
noise
suppression
and
boost
intelligibility
of
the
incoming
signals,
which
I
assume
is
probably
auto
gain
the
drum,
and
so
what
I
like
about
confidence
is
that
they're,
simpler,
API,
I
love,
summer,
VI
and
so
I.
K
K
It's
a
missed
opportunity
to
make
it
testable,
because
right
now
it
might
set
music
and
then
I
and
I
mark
the
settings
for
echo.
Cancellation
goes
oppression
and
again
control
what
do
I
get
back,
and
it's
not
clear
to
me
well,
it
sounds
like
it's
saying
right
now
that
a
competent
is
does
not
preclude
settings
on
the
object
so
setting
something
I
didn't
take
precedence.
So
actually
it
doesn't
say
that
ready,
yeah
anything.
K
Is
a
secondary
thing
and
it
should
not
override
settings
so
so,
presumably
so
the
example
I
show
here
sending
music
and
then
it
would
be
nice
if
we
don't
really
could
expect
echo
cancellation
good
right,
now,
constraints
outsource
requirement
to
the
application
so
that
it's
constraints
does
not
that's
long.
Many
constraints
is
that
it
doesn't
say
anything
about
default
values
at
all
by
the
way
definition
it
lets
the
browser
do
whatever
it
wants
with
in
an
envelope.
It's
great
many
application.
K
So
having
a
way
to
change
the
default
within
I,
never
thought
would
actually
be
useful
and
testable.
So,
for
instance,
then,
if
you
I
think
right
now,
actually
we
hacked
it
a
little
bit
and
both
Chrome
and
I
think
in
Firefox.
We
don't
believe
open
it
up
to
any
network.
Rum
does,
which
is
that
the
problem
has
added
to
be.
K
We
started
to
make
a
cancellation,
that's
the
first
constraint
that
we
specified
and
I
think
it's
from
that
echo
cancellation
is
false,
then
from
defaults,
most
cancellations
or
suppression,
or
again
you
tell
also
default,
so
they
key
off
the
first
constraint.
Just
because
that's
the
dominant
constraint
right,
which
is
a
very
happy
way
to
control
the
default,
so
this
will
actually
be
a
more
clean
when
different
folders
fall.
K
So
say
that
this
is
an
appeal
to
adding
one
over
to
the
steps.
If
we're
going
to
have
comment
is,
and
also
I
had
a
question
at
the
bottom
of
web.
Where
would
that
go,
and
maybe
I
should
just
segue
in
because
they've
kind
of
discuss
the
same
things
about
they're
useful
to
comment
on
these
individually.
K
That,
but
hints
is
that
there
is
somewhat
akin
hearing
properties
from
attract
like
there's
a
music
track
and
motion
track,
and
you
might
think
they're
invariant
and
ever-present
as
they
describe
infinity.
But
they're.
Not
you
know,
first
of
their
control
surface,
which
means
that
they're
in
runtime
not
find
them
at
any
time
goes
on
right.
K
Yeah,
there's
also
not
specify
any
wearing
or
are
there,
nor
when
the
observable-
and
they
don't
follow
the
medium
meaning
that
the
media
goes
through
the
same
pipes
from
thanks
to
sort
of
sources,
distinct
every
once
in
one
way,
Mike's
like
make
a
connection.
The
element
after
stream,
audio
meter,
recorder
or
even
come
from.
K
I
know
the
last
one
is
like
competence,
come
from
the
JavaScript
application
and
someday
user
agents.
Maybe
today
may
be
able
to
detect
the
content
is
directly
by
just
analyzing
it,
and
it's
unclear
whether
these
user
agents
are
allowed
to
ignore
the
hint
from
JavaScript.
If
it
knows
the
immigrants
consent.
D
D
B
I
think,
if
you're
shaping
feature-based,
we
can
see
this
because
overlaps
with
undefined
behavior,
you
have
a
problem
like
image.
Should
you
care
me
and
if
you
have
an
API,
miss
can
assess.
If
you
know
Chris
function,
then
x
y&z
happens,
then
yeah
an
X
I
said,
should
happen,
even
if
the
browser
is
a
second
to
there's
something
better
than
X
I
said
kiss
you
wear
this
yes
and
play
I
mean
there's.
K
M
D
A
D
N
H
H
D
H
A
Yeah
I
mean
yeah
the
finding
a
Yanni
very
asked
for
the
delay,
particular
how
long
am
I
taking
and
some
of
those
questions
might
be.
Connect
specific.
Like
it's
a
weird
example,
we
added
the
text
content
hands
in
the
idea
that
it
might
result
in
turning
on
content
coding
tools
and
certain
codecs,
but
you
probably
don't
want
to
probably
want
to
start
doing
some
of
those
things
at
a
keyframe
as
opposed
to
immediately
so
they're
over.
A
T
H
B
B
H
K
Think
it
suffered
from
like
we
can
specify
it
I
think
it
should
be
specified
in
normative,
so
that
they're
not
they
would
hinge
and
wise
that
he
don't
have
to
follow
it.
I
think
you
know
we
have
to
sustain
whether
that's
true
or
not,
and
it's
back
to
be
clear
about
that
as
far
as
the
redundancy
of
it
I
think
it's
just
kind
of
like
on
a
focus
that
you
know,
and
obviously
you
can
also
have
a
default
mode.
B
Go
after
because
of
one
one
extreme
or
the
other
like
a
running
stream,
everything
is
a
very
specific
controller
and
then
that
would
be
again
it's.
But
but
even
when
we
were
in
this
gray
area,
where
we
eat
these
place,
then
of
course
we
should
make
it
past
a
specified,
that's
possible
right,
maybe
even
we
can
say
called
function
a
week
instance
day.
They
intended
effects
of
this
API
to
do.
K
This
it
was
an
example
reading
beforehand,
as
I
said,
a
lot
of
bad
things
about
it.
We
have
three:
we
have
a
translation
expression
that
the
auto
gain
control.
If
we
had
a
fourth
one
I,
don't
know
what
it
is.
Maybe
there's
a
fourth
one.
If
we
weren't
had
a
fourth
one,
you
know
we
might
want
to
have
different
defaults,
whether
it's
music
for
or
speech
and
confident,
we'll
pick
up
a
not
default,
still
background.
A
H
S
A
U
I
A
R
D
D
We
had
to
turn,
we
need
to
be
able
to
say:
I
want
the
blunt
propriety
in
queueing
without
turning
on
the
AC
P
or
the
other
way
around,
and
we
wanted
to
do
this
without
requiring
the
identical
callback
expect
from
it
from
their
desk.
You
I
guess
that
assumes
that
you're
doing
both
at
same
time
and
same
controls.
D
D
D
The
nice
thing
about
the
Canon
is
to
be
mm
kind
of
monkey
conscious
around
the
house
prior
it
works
and
and
the
way
I'm
having
having
control
over
both
my
might
actually
get
some
simplified
interface,
and
the
other
purpose
is,
of
course,
just
drop.
It
don't
say
officially
they
care
that
we
don't
want
any
more
words.
The
bad
thing
about
that.
One
is
that
if
you
try,
it
is
left
in
web
RTC
documents.
D
It
means
that
the
requirement
is
278
keyword
very
distant,
which
means
that,
when
on
performance
hand
without
conforming,
which,
if
you
do
the
sensible
thing,
my
favorite
at
the
moment
that
proposal
on
and
which
also
and
that
that
is
for
solving
the
discussion
about
what
to
do
with
features
adjacent
same.
She
seems
to
me,
like
a
psychological
thing,
to
do
next
again,
something.
D
E
V
S
Have
a
testing
artist
when
you
get
to
the
beauty,
mr.
red
yeah,
I,
guess
this
gives
context
that
there
are
two
parts
totally
found.
One
is
making
the
RTC
X
transport
stand-alone
in
the
Indy
style
and
the
other
is
to
add
a
bunch
of
stopped.
You
did
something
the
application
into
twelve,
more
things,
probably
by
the
next
guys
and.
S
A
L
D
A
A
A
A
H
E
E
B
And
yes,
it's
terrifying!
The
thing
about
chrome
implementation
status.
While
the
constructor
is
shaped
regardless,
you
can
inspect
them
anything
useful.
Like
start,
a
gather
start
under
budget
like
the
stuff
you'd
use
for
a
standalone
and
ice
transferred.
Those
things
are
still
behind
orange
in
child
black,
so
you
can
only
have
a
dummy
ice
transport
if
you
don't
have
a
black.
K
E
A
I
mean
we'll
get
into
discussion
of
functionality,
but
basically
it's
what's
been
implemented,
promised
thrown
at
the
same
functionality
as
was,
and
it's
all
the
same
methods
angle.
So
it's
they
went
to
it.
They
were
essentially
two
implementations
and
it's
already
so
at
least
for
that,
for
that
set
of
functionality,
it's
pretty
well
understood.
He
wrote
long
about
other
things.
You
might
want
Witcher,
not
mechanic.
T
A
S
Yeah,
so
then
there
are
two
Possible's
there
you
go,
one
is
to
implement
things
that
build
on
top
of
that
which
I
guess
it's
into
the
discussion.
But
even
if
this
working
group
doesn't
do
anything
on
top
of
it,
it
is
useful
for
teenagers
to
want
to
do
pretty
raincoats,
so
the
stories
will
be
there.
The
second
Avenue
that
can
be
pursued
is
to
improve
it,
and
so
far
four
flex
ice
were
general
consensus
to
do
something.
S
So
I
think
it's
to
the
interesting
work,
which
is
what
do
the
three-dimensional
scene
I
took
a
long
time
to
figure
or
to
find
out
what
it
was
it
took.
I
mean
I
had
an
actually
I
didn't
find
out
like
over
a
year
ago,
so
I
finally
got
in
contact
with
the
very
people
and
a
lengthy
technical
discussion
with
them
and
Linus
nation
of
sorts
and
presented
to
them
should
the
slides
with
them,
and
they
said
yes,
this
would
work
for
us.
So
that's
why
I'm
presenting
London.
K
X
S
Trying
to
route
between,
because
basically
they
wanna
appear
act
like
a
server
in
the
sense
that
it
can
post
information
such
as
local
candidates,
somewhere
that
other
people
can
read
and
then
listen
for
incoming
connections
and
then
scaled
up
to
make
any
connections.
So,
just
like
you
would
run
a
normal
server
if
you
had
a
public
IP
address
for
you,
publishers,
hostname
port
and
then
munis
and
frame,
come
in
connections
and
they
want
to
have
mini.
And
you
want
to
do
that.
V
V
S
Is
one
piece
of
the
puzzle?
Okay,
there
are
several
pieces
and
I'm
gonna
try
and
explain
how
they
all
fit
together
with
a
little
animated
diagram
here,
there's
a
part
that
I
think
it
might
explode
so
watch
out,
and
let
me
know
if
you
partner,
okay,
so
first,
the
JavaScript
application
we've
asked
the
browser
to
gather
ice
candidates.
Perhaps
with
just
English
translation,
gathered
review
the
peer
connection,
technical
description,
then
the
browser
gathers
the
candidates
and
hangs
back
the
JavaScript
application.
S
The
candidates
along
with
I,
see
Frank
password
reset
is
affirmed.
Then
the
application
would
post
this
information,
along
with
the
DTS
fingerprint,
is
necessary
up
to
some
databases
you
can
think
of
it
like
singling
server,
but
instead
of
sending
in
somewhere
this
engine.
That
puts
it
there
for
a
long
period
of
time
and
then
someone's
motivated.
That's
the
one
important
distinction
here
for
some
future
some
future
time.
Some
peer
reads
that
out
of
oak
started
to
do
systemic,
an
event
one.
Yes,
at
the
same
time,
the
application
tells
the
browser
to
a
scheme
that
candidates.
S
S
So
at
some
future
time
we
here
will
pull
from
the
database
to
the
fact
that
the
first
compute
along
the
Left
exists
it'll
get
its
you
Frank
password
and
service
lines
of
candidate,
which
is
tech,
light
and
fingerprint,
and
now
it
will
exempt
to
connect
to
the
browser's
come
and
then
please
holding
that
port
leave
them
with
the
whole
thing.
Annette
Sykes
check
goes
over
there.
Now
the
browser
is
the
use
dragged
out
of
the
ice
check.
Hashes
is
and
gives
it
to
the
JavaScript
application.
S
S
In
first
yes,
so
then
the
JavaScript
application
tells
the
browser
to
fork
to
the
ice
transport
so
that
it
keeps
the
same
local
dicing,
fragment
password
in
Canada
and
tells
us
to
start
connecting
to
the
remote
on
the
right.
With
a
new
ice
transport.
That's
been
forked
with
a
different
remote
in
fragment
password
and
then,
of
course
you
do
details
fingerprinting.
So
then
a
nice
text
will
flow
in
both
directions
and
then
ice
will
work
we'll
get
to
type
of
different
types
of
maximum
and
then
after
work.
S
S
D
W
S
They
said
they
have
something
where
they
can
put
something
and
pull
something
out.
I
said:
okay
over
minute,
yeah,
okay,
so
the
problems
with
the
current
connections
web
RTC
spec
that
does
not
allow
this
is
that
we
currently
don't
have
a
way
to
have
local
candidates
useful
for
more
than
one
connection
and
also
for
a
long
period
of
time,
and
you
have
to
know
the
remote
need
for
a
password
before
you
can
connect,
and
also
you
can't
create
very
good
connections.
Chrome
has
led
to
their
connection
with
something
other
than
500.
Okay
cuz
they
wanted.
S
For
this
are
first
give
you
my
support
team
and
you
can
have
more
than
one
connection
or
the
same
local
candidates.
You
have
retain
candidate
from
flex
sites.
You
can
keep
that
still
Kennedy
live
on
receives
check
is
the
name
of
the
mind-blowing
thing
that
I
came
up
with
my
chosen
and
then
so
avoid.
N
S
S
Core
king
would
work.
We
would
create
a
new
ice
pants
pulled
into
that
like
and
I
serve
them
together
and
it
is
and
then,
when
you
need
it
to,
you,
call
fork,
and
that
would
give
you
back
a
client
transport
to
the
other
side
that
you
will
start
to
get
a
multiple
transports
like
issue.
It's
just
a
transport
instead
of
climbing,
but
anyway
the
initial
thing.
We
never
connect
anything
we're
just
treating
a
server
that
be
for
fun
and
then
retain.
Can
local.
S
Like
my
flex
ice
presenting
a
while
ago,
once
you
get
in
a
server
flight
contaminate,
you
say
get
retain
that
one
slide.
So
this
is
the
mind-blowing
one.
Basically,
you
listen
to
an
event
and
the
event
would
have
perhaps
named
hash.
The
load
is
you
name
a
fragment
and
then
you
take
that
thing
and
send
it
off
to
some
lookup
and
then
that
would
give
you
back
that
I
pressures
that
you
need
to.
A
S
S
This
is
a
full
example.
This
is
the
whole
shebang
put
together
that
happens
in
Venn
diagram
device
print
work,
creating
like
a
server,
listen
first
server,
reflexive
candidate,
retain
it
listen
from
incoming
I
strike,
get
the
hashed
remote
users
in
fragment,
use
that
to
do
a
look,
Oh
work,
the
ISIS
server
then
do
if
next
a
little
inside
and
then
do
a
new
detail.
Aside,
penetrator
detail:
it's
messy
SPB
transport,
based
on
the
other
information
that
you
didn't
look.
A
S
S
E
K
S
V
S
S
W
S
To
a
server,
and
then
it
sends
you
what
you
open
up
here,
your
local
court,
and
then
you
send
your
candidate
to
a
server
the
server
just
sensing
checks
and
puts
data
in
the
new
friend
field.
And
then
you
can
receive
basically
wrong.
Evp
packets,
with
some
enough
training,
really
I'm
certain.
How
does
the
actually
prevent.
S
Because
the
news
well,
if
you're,
worried
about
very
small,
not
to
be
missed,
like
everything,
then
actually
we
already
have
an
issue.
I
says
is
such
a
go
back
to
slide.
I
was
concerned
enough
about
this,
that
I
thought
through
it
and
what's
already
possible
than
through
this
thing,
nice
implementations
and
you
can
already
get
some
amount
of
data
both
from
server
to
client
and
clients.
Orders
like
gaming
data
entry,
defrag
or
into
resilient
placement
groups.
Oh
in
the
address
field
of
my
name's
box
and
I
could
have
prospective.
V
S
Yeah,
so
how
hard
is
this
to
implement
so
I
went
through
the
different
pieces
and
I
thought?
Okay,
he
order
to
see
stylized
transport
that
sees
the
on
received
check.
That's
just
at
an
event.
Check
comes
in
hashtag,
okay,
so
long
live
candidates
is
probably
easy.
Just
keep
sending
the
son
checks
the
freestanding
dqi,
some
sttd
transport
and
it's
kind
of
moderate.
It's
part
of
weather,
to
see
me
not
to
buy
the
I
start
working.
A
V
W
S
A
So
we
did
this
an
RTC
mine.
We
did
not
do
it
because
the
use
cases
you
you
don't
need
working,
really
drinks
watch
any
kind
of
use
cases
and
the
HTS
weren't
particularly
high,
and
that
particular
browser
thing
but
I
want
you
see,
look
there
were
gaming
people
and
who
are
interested
in
it.
The
main
complexity
there
is
is
that
we
decided
to
have
the
separate
ice
gatherer
object,
because
it
was
easier
to
have
all
the
candidates
within
an
object
rather
than
having
ice
transports.
A
The
complexity
came
out
of
realizing
that
there
were
other
things
that
needed
to
support
working
as
well
like.
Initially,
we
didn't
do
support
for
community
gos
transport
and
we
realized.
Oh
we've
got
a
cast
or
certificate
in
there
because
everybody
every
TT
allows
transfer
when
you
four
could
all
be
started
from
the
same
cert
and
then
that's
what
Gary
the
whole
idea,
particularly
if
you
wanted
to
put
him
under
than
threads.
You
have
to
have
a
post
usage
thing
where
you
started
moving,
you
know
being
able
to
give
the
certificate
teacher.
A
These
things
were
going
to
create
so
anyway,
that
was
so.
It
took
a
lot
of
rework
to
figure
all
that
stuff
out
and
in
addition,
there
were
additional
ice
optimizations.
You
probably
wanted
to
do
that.
I
think
irrelevant.
Equivalent
of
things
like
Cheney
was
nomination
or
recombination
or
all
of
that
stuff.
I,
don't
know
whether
I
don't
think
the
ice
gather
really.
Was
that
helpful
in
those
things,
those
were
the
most
a
that
guys
transcriptions,
not
gathering
things,
but
the
other
thing
is
if
it
was
helpful
to
have
a
nice
gather.
A
If
you
wanted
to
have
a
lot
of
control
about
what
candidates
were
so
like
you
know,
sometimes
people
would
only
want
to
gather
or
I
give
you
six
or
you
know,
sir,
from
certain
interfaces
and
stuff
like
that.
So
I
wouldn't
say:
overall,
it
wasn't
enormous.
It
was.
We
had
to
do
a
lot
of
rewriting
of
covenant,
but
that
was
because
we
didn't
understand
it.