►
From YouTube: IETF102-HTTPBIS-20180718-1330
Description
HTTPBIS meeting session at IETF102
2018/07/18 1330
https://datatracker.ietf.org/meeting/102/proceedings/
A
B
C
So
yeah
we're
definitely
need
to
start.
It
is
just
after
the
start
time.
This
is
httpd.
If
you're
not
here
for
HTTP,
we
have
an
agenda
today.
Two
things
made
sense
on
the
agenda,
so,
first
of
all,
no
well
no!
Well,
if
you
haven't
seen
this
already,
you
can
find
it
by
searching
for
ITF
note.
Well,
these
are
the
terms
under
which
we.
C
Things
like
your
behavior
and
intellectual
property.
So
if
you're
not
aware
of
this,
please
do
read
it
because
you
are
bound
by
these
guidelines.
Well,
the
guidelines
are
rules
anyway,
best
practices,
I,
guess
there
you
go
codes.
Today
we
have
a
scribe.
Thank
you
very
much.
The
blue
sheets
can
now
be
circulated.
Thank
you
very
much.
We
have
a
job
really.
Thank
you
very
much.
So
today
all
we're
gonna
do
is.
C
F
So
HDPE
there's
a
product,
the
quick
working,
not
the
HDPE
working
group
through
interesting
machinations.
Yes,
and
the
quick
it
is
Charlie
was
trying
to
describe
mappings
between
HTTP
number
normal
semantics,
using
quick
and
accommodate
htv-2
extension
mechanisms
in
HIV
over
quick.
So
the
process
of
doing
that
we
have
adopted
some
changes
to
how
HCP
is
mapped
that
might
be
beneficial
in
the
HIV
to
world
as
well,
and
we've
also
kicked
some
things
out
as
that's
HTP
visits
job,
not
ours,
so
I'm
here
to
make
sure
that
everyone
is
on
the
same
page.
F
F
Some
people
will
use
that
to
build
their
priority
tree
and
hope
that
the
server
doesn't
actually
clean
up.
Those
particular
extremes
that
they've
skipped
it's
clean,
but
servers
don't
technically
know
which
close
streams
are
safe
to
clean
up
and
which
ones
aren't
so.
Plants
have
to
make
some
assumptions
and
servers
have
to
make
some
assumptions
about
which
ones
they
can
scavenge,
and
it
can
lead
to
the
priority
tree
being
inconsistent
between
clients
and
servers.
It's
a
little
bit
ablate.
It
should
be
over
quick.
F
F
F
F
G
H
So
I'm
a
big
fan
of
this,
like
in
the
quick
scheme,
I,
think
this
does
a
nice
job
of
preserving
immersive.
The
flexibility
we've
had
while
reigning
in
state
problems
and
the
other
issues
about,
besides
not
being
sure
of
what
the
state
on
the
other
side
is
as
well
as
the
asylum.
The
overall
you
know,
requirements
I,
do
quick
as
h3
in
most
ways
and
I
never
feel
the
need
to
like
back
for
another
feature.
Unless
there's
like
a
critical
urgent
need
right,
and
so
my
implementer
hat
would
be.
H
H
C
I'll
insert
myself
in
queue
and
also
avoid
the
TV
personally
I,
think
I
agree
with
Patrick
I
think
the
only
place
where
this
really
matters
is
when
it
affects
some
someone's
implement,
not
not
the
implementation
but
but
someone
using
the
protocol.
So
if
this
was
surfaced
in
API
somewhere
either
on
the
server
side
of
the
client
side
on,
you
know
the
more
different
abstractions
we
have,
whether
incompatible
the
more
it
causes
angst
and
on
the
client
side,
I
think
it's
pretty
much
a
clean
slate.
J
Just
embrace
anything
that
could
help
simplify
this
priority
scheme.
H2
will
be
very
welcome
if
this
helps
rename
the
state
explosion.
Power
mature,
but
I'm
intrigued
by
Patrick's
comment.
So
with
that
clarification
backwards,
I'm
wondering
you
know,
we
have
a
regular
session
and
the
fallback
position
is
always
there
actually
per
charter.
So
we
may
end
up
with
h2
when
we
think
we're
going
to
for
quite
a
whiie
but
capable
quick
with
h2.
H
Now,
in
this
scenario,
my
h2
is
pH
2,
but
my
I
think
that's
you
know
dead.
Stop
so
Mike's
asking
do
we
want.
You
know
an
h2
extension
here
which
we
can
add
right.
So
do
you
like
h2
and
speaker
new
extension,
with
a
new
set
of
priorities
right?
Is
that
amount
of
machinery
worth
marijuana
protocol
that
you
know
we
expect
to
be
substantially
replaced
by
you
know
by
HQ
right
to
me,
that
seems
like
a
whole
different
ecosystem,
for
you
know,
hopefully
diminishing
returns
as
time
goes
on
right.
J
H
F
D
L
Decision
later
I
air
canary
Apple,
as
someone
tasked
with
providing
api's
that
allow
people
who
express
stuff
I
think
with
priorities.
It's
not
that
big
of
a
difference,
and
here
personally
I,
would
be
perfectly
content
to
leave
things
alone
and
deal
with
it
later.
But,
as
we
start
talking
kind
of
about
the
future
and
where
things
diverge
further,
if
it
diverges
a
lot,
that
starts
to
become
a
little
a
little
more
tricky.
L
C
I'll
just
insert
that
I
think
one
of
the
factors
here
in
terms
of
us
coming
up
with
something
better
in
this
community
is,
is
that
because
of
the
overlap
in
the
communities
you
know,
this
is
gonna,
be
catered
on.
The
availability
of
implementers
to
put
effort
into
something
better
and
right
now,
they're
all
very
busy,
especially
on
the
client
side,
which
is
where
we
really
need
it.
So
I'm
not
surprised
that
we
have
this
so
cut
Roy
and
then
Adam.
A
F
B
C
F
F
F
F
And
lastly,
I
think
is
last
it
GP
over
quick
has
decided
that,
along
the
same
principles
as
TLS
greasing
a
lot
of
good
points
and
quick
greasing
a
lot
of
the
public
pieces
anywhere
where
a
value
should
be
ignored,
if
not
unknown.
It
defines
a
couple
of
frame
types,
a
couple
of
settings,
values
and
says:
if
you
see
these,
these
had
no
semantic
value
whatsoever.
You
don't
understand
them.
Keep
moving
and
I
understand
that
recently,
a
an
implementation
was
encountered
which
did
not
tolerate
an
unknown
setting
type.
F
O
C
C
O
H
O
H
Right
now
so
I
mean
I
think
the
driver,
the
brothers
many
things
brought
it
to
a
head
for.
Tell
us
one
three
and
four
quick.
When
sort
of
the
strategy
was
really,
you
know
innovated
and
it's
been
great
initiating
this
industry
development
and
h2
didn't
have
this
philosophy
right.
So
this
is
really
what
ever
considering
can
you
can
you
back
pork
grease
and
if
your
joint
is
already
arrested,
the
grease
isn't
gonna
help
if
you
slap
it
on
the
top
right,
that's
kind
of
what
we're
concerned
about.
H
So
my
draft
identified
how
to
do
connects
tunnels
in
h2
for
bootstrapping
WebSockets
was
deployed
in
by
a
very
major
Syrian
and
in
that
draft
he
set
out
a
settings
from
the
server
side
towards
the
towards
the
pier
mana
turned
out
a
fairly
commonly
deployed
to
open-source
implementations
upon
receiving
the
settings.
Value
close
the
connection.
H
Like
the
almost
immediately,
but
of
course
you
have,
the
issue
of
you
know,
making
those
updates
in
the
field
I
mean
so,
let's
charter
driving
this
as
a
motivator.
So
if
there
are
other
people
who
be,
you
know
willing
to
jump
into
the
breach
with
me
to
test
these
greasing
points
that
always
makes
a
lot
easier.
If
you're,
not
the
only
one
who's
willing
to
sacrifice
some
connections
for
the
for
the
purpose
of
data
gathering,
so
that'd
be
great.
If
you
could
approach
me,
it
sounds.
F
C
Yeah
I
you
something
it's
been
a
hotel
for
sure,
so
any
any
other
clients
or
especially
servers,
who
would
be
interested
in
trying
to
run
the
similar
experiment.
I
don't
have
to
oh
okay
cool.
So
maybe,
let's
wait
to
hear
back
from
you
folks
on
the
list
to
make
sure
it's
something
saying
to
do,
and
then,
if
so,
we
can
write
a
draft
and
do
the
right
things
and
the
registries
and
so
forth.
Okay,
all
right!
Thank
you!
Thank
you,
Mike!
C
H
P
I
started
looking
at
the
RFC
router,
we
have
27
left
which
have
that
there
are
actually
27
which
are
not
rejected,
13
of
which
verifies
nine
head
for
document
update
five
reported
of
these
27
we
have
already
addressed
18.
Nine
are
still
open.
All
of
them
have
been
for
all
of
these.
We
have
opened
a
table
shoes
and
we
have
got
a
summary
page.
P
Okay-
and
this
is
a
bit
that
I
wanted
to
get
feedback
on
some
of
the
zoning
issues
that
we
had,
the
errata
actually
taxed,
and
the
HTTP
a
DMF,
so
I
really
think
to
get
things
addressed
properly.
We
want
to
have
test
cases
and
test
suite,
and
if
somebody
can
supply
RFC
52:34
confines
a
DNF
rebel
rouser.
That
would
be
great,
otherwise,
we'll
have
to
figure
something
out.
P
One
other
thing
I
was
thinking
about
is
that
you
can't
be
through
everything
was
single
collected,
a
B
and
F,
and
that
doesn't
really
make
a
lot
of
sense,
because
having
messages
is
really
different
from
following
the
comments
of
field
value,
so
I
was
planning
to.
We
arranged
that
in
to
label
the
a
B
and
F
fragments
that
we
have
to
be
in
two
different
buckets,
essentially,
I
probably
should
automatically
for
that
other
discussion
and
that's
all
I
have
before
we
go
on
with
the
tickets.
P
Finished
this
year,
but
it
doesn't
seem
it
gets
finished
before
us.
So
maybe
we
also
need
to
coordinate
this
with
my
quest
at
that
spec.
Actually,
we'll
put
it
on
top
of
new
prospects,
so
the
relationship
shift
to
these
other
specs
that
have
something
to
do
with
HTTP.
If
we
should
discuss
at
some
point
of
time.
C
So
Julian
I
have
an
open
issue
in
pcp
56
best
to
reference.
The
new
documents
and
I
think
we're
gonna,
keep
it
up
and
from
that
from
the
minute
for
those
from
the
discussion
the
other
day,
I
think
maybe
what
we
could
do
is
go
through
the
cookies
dock
and
give
Mike
and
John
some
issues.
If
we
see
anything
where
there's
a
mismatch
or
reference
yeah,
not
sure
about
quick,
I,
guess
I,
guess
they
wait
when
we
started
this
work,
the
idea
was
that
if
these
were
done
in
time
it
would
reference
them
so.
F
C
P
N
H
N
D
H
N
C
Mostly,
what
I
wanted
to
do
those
highlight
to
folks
to
make
sure
they
understood
that
the
real
good
happened
and
that
it's
relatively
stable,
although
I
think
in
our
discussions
we've
said,
we
know
that
we
still
have
some
text
to
write
in
fill
in
in
some
places
and
some
connections
to
make
and
so
forth,
but
I
think
the
overall
structure
is
relatively
solid.
All
the
way
might
tweak
it
as
we
go
along.
Do
you
see
that
it's
accurate
running?
Yes,
okay,.
C
You
really
should
join
github
today,
other
than
it
labels
or
just
hit
that
discuss
when
that'll
get
them
all.
So
we
we've
highlighted
a
bunch
of
issues
that
we
felt
like
it'll,
be
nice
to
get
some
input
from
from
the
broader
group.
I
help
us
move
forward
I'm.
So,
let's
go
through
those
Julian
I'm,
just
gonna
go
through
them
in
the
order
of
the
of
the
issues
list
so
issue.
One.
Two
three
Julian
include
status,
42.
P
Raptor
free
made
a
very
generic
different
definition
of
dead
end
and
it
is
really
widely
used
outside
vector
and
the
one
thing
that
seems
to
keep
a
few
people
from
actually
using
is.
It
is
the
argument
atoms
in
the
backup,
spec,
so
I
think
that's
one.
Paragraph
status
call
definition
that
we
should
actually
pull
into
the
car
drives
aspects.
C
P
O
H
C
Our
headers
are
always
defined
with
a
DNF,
and
so
we
are
when
we
were
discussing
with
structured
headers.
This
came
up
well,
semi
230,
says
they're
set
of
must
are
generating
multiple
header
fields
of
the
same
field,
name
in
a
message
unless
other
the
entire
field
value
that
had
a
field
is
defined
as
it
comma
separator
list,
especially
after
forth,
based
syntax
that
we
come
up
with
in
sections
of
them
or
the
heli-field
lawyers.
Well,
none
exception
is
not
abalone,
which
means
basically
cookies.
C
C
Also,
you
know
at
times
specifics
index
handling,
so
in
other
words,
when
you
look
at
that
a
DNF
that
tells
you
how
to
combine
two
headers,
how
white-spaces
handle
you
know
empty
header
fields,
that
sort
of
thing
and
how
you
can
authorize
those
effectively
and
that's
tied
again
to
that
up
to
thwip
rule.
So
you
know
if
we
can
define
heteros
without
using
a
be
enough.
The
question
is:
are
those
requirements
there's
others
behaviors
to
require
Bhanumathi
headers,
or
is
that
specific
times
the
funnies
in
that
syntax?
C
This
becomes
important
for
structured
heathers,
because
you
know
they're
defined.
Primarily,
you
know
they're.
The
normal
definition
is
is
with
a
bunch
of
algorithms,
and
so
we
talked
through
this
in
the
issue
and
I.
Think
I
made
a
proposal
down
here
which
is
I,
replaced
that
text
with
the
center
must
large
area.
Mental
health
fields
are
the
same
field
name
and
a
message
in
rust.
That
fields
definition
allows
this.
For
example,
it
uses
the
comma
separated
list
in
a
behavior,
so
doctors
are
values
where
the.
C
C
P
Of
course,
I
just
recalled
that
we
actually
have
our
related
slightly
related
issue
to
that
which
I
opens
when
we
did
our
service,
where
we
actually
have
lists
and
tags,
but
the
whole
header
field
is
not
really
defined.
As
of
this
production,
because
we
have
a
special
I
think
Tia
Peralta,
so
we
already
have
a
different
ticket
push
which
is
slightly
related
to
this.
P
P
C
No,
it's
a
little
more
subtle
than
that.
I
think
the
requirement
back
in
an
issue
that
one
refers
to
is
about
sender
behavior.
It's
always
saying
what
it
can
generate:
multiple
header
fields.
It
doesn't
place
any
requirement
on
when
a
recipient
can
or
cannot
combine
headers,
because
to
do
so
would
require
a
recipient
to
understand
the
syntax
of
every
possible
header
field
and
it
becomes
deployment
problem.
C
But
there's
the
implicit
requirement
which
maybe
we
need
to
be
explicit-
is
the
recipient.
When
those
recipient
receives
two
instances
of
any
header
field,
no
matter
what
it
is
and
header
field
should
be
defined
in
such
a
way
that
they
were
bused
in
the
face
of
that
and
now
what
I
have
something
to
say.
H
H
N
C
C
K
N
H
C
O
Sometimes
my
understanding
is
that
no
one
effectively
uses
this
and
those
that
attempt
to
find
that
they
get
results,
that
they
might
not
want
asking
for
something
and
then
getting
it,
and
you
might
not
actually
get
what
you
it's
not
very
clear.
What
are
you
in
particularly
since
media
types?
Often
don't
signal
charges
says
very
clearly
right.
H
D
O
I
think
this
is
this
is
not
a
case
of
throwing
it
out
of
the
document.
It's
it
would
retain
the
description
of
the
header
field,
maybe
maybe
maybe
added
in
the
sense
that
this
one
existed
for
the
purposes
of
negotiating
character,
sets
but
turns
out.
It's
not
a
good
idea
and
we're
obsoleting
editors,
herself
I,
think
that's
that's
a
reasonable
thing
to
do.
I.
O
N
O
C
This
is
our
radius
of
responses
and
I'll.
Just
read
out
the
issue.
The
specification
of
cash
control
and
more
store
on
the
response
side
can
be
a
Wed
to
allow
the
response
to
be
reused
for
another
concurrent
request
since
its
specified
in
terms
of
non-volatile
memory.
What
was
other
cache
directors
or
Sunderland
such
we
use?
So
this
is
so
close.
It's
very
common.
Now,
for
at
least
we
know,
media
cache
is
maybe
the
browser
side.
C
It's
called
request,
collapsed
or
collapsing
request,
forwarding
they're
a
bunch
of
names
for
it
it's
become
quite
common
clarification
as
to
whether
such
we
used
for
temporary
concurrent
requests
would
be
helpful
on
a
special
interest
or
I
expect,
for
other
questions
were
possibly
to
this
came
up.
I
was
a
little
surprised
when
I
heard
someone
rated
this
right
because
in
in
my
mind
and
all
those
questions
we've
had
previous
in
the
working
group,
you
know
the
the
rough
semantics
of
most
oil
that
this
bypasses
the
cache
completely.
C
It's
like
you're
riding
around
the
cache,
and
so
the
things
that
you
can
vary
use
it
for
multiple
requests
is
a
little
surprising,
so
I
just
thought.
It'd
be
good
to
clarify
that
one
way
or
the
other,
and
though
just
to
read,
while
INRI
wasn't
in
our
weather,
the
definition
of
a
no
story,
the
most
or
respond
selectively
indicates
the
cash
must
not
store
any
part
of
either
than
immediate
request.
A
response.
This
directive
applies
to
both
province.
Your
caches
must
up
not
store.
C
O
H
So
so
the
push
I
was
gonna
add,
so
we
might
introduce
the
concept
of
single
use
resources
right,
which
is
what
we
mean
in
a
lot
of
these
cases,
so
the
other
way
that
browser
uses
these
and
I
think
it
may
have
had
to
do
with
the
original
North
Shore
language
a
little
bit.
No
stores
are
stored
like
in
memory,
so
you
can
go.
C
O
Soon
I
just
got
the
crepes
again,
maybe
maybe
this
we
need
to
decide
which
one
of
these
things
no
store
is,
and
then
we
have
some
separate
work
to
do
on
a
new
directive
to
fill
the
other
mate.
Because
there's
this
notion
of
collapsing
one
collapsed
on
on
that's
a
new
directive.
Isn't
it
I,
don't
think
so?
No,
it's
done.
C
H
G
D
H
C
Q
H
Q
From
the
pharmacy
Dean's
perspective,
when
we
see
multiple
results,
multiple
requests
come
in.
At
the
same
time,
we
tend
to
treat
those
as
though
they
were.
You
know
if
they're
really
close
together,
we
think
you
were
going
to
it's
just
like
they
were
the
same
request,
and
so,
if
we
really
do
want
it
to
be
separate,
we
should
be
specific
about
that,
because
I
actually
would
have
assumed
the
exact
opposite.
I.
C
C
C
I
C
O
C
P
O
Yeah,
that
was
that
was
my
sort
of
assumption
here.
Is
it
that
it
would
be
good,
because
it's
such
a
small
thing,
because
it's
it
affects
the
the
meaning
of
those
header
fields
pretty
pretty.
Fundamentally,
it
would
make
sense
to
roll
that
entire
thing
in
would
be
what
an
extra
paragraph
and
the
definition
of
each
theme
at
most
yeah
I
mean
this.
H
C
C
D
G
H
C
Okay,
just
speaking
personally
I
don't
have
a
strong
opinion.
I,
don't
personally
have
an
immediate
use
case
for
this,
but
acne
found
one
for
which
is
kind
of
indicative
and
I.
Think
I
have
a
suspicion
in
which
TP
and
OPI
folks
would
be
interested
in
this
from
time
to
time.
For
the
same
reasons,
I
could.
H
C
C
This
may
be
too
aggressive
in
a
sense
of
what
people
thought
are
just
observing
that
people
tend
to
use
the
star
value
and
they
accept
headers
or
not,
and
often
it's
not
terribly
helpful
to
have
it
in
there.
Awhile
and
I
was
wondering
if
we
could
have
some
advisory
text
around
it
or
carve
some
of
it
off
completely
or
discourteous,
or
something
because
if
the
Q
value
of
your
star
value
is
higher
than
any
other
value,
the
semantics
are
basically
send
anything
in
preference
to
something
specific
which
isn't
useful.
C
I,
don't
know
of
any
practical
use
case
for
that
I'd
love
to
hear
if
Cindy
has
one,
if
the
Q
value
of
two
the
star
value
is
no
even
then
you
know
it
says
some
are
more
specific
or
in
preference
to
something
else,
but
the
default
semantics
of
the
except
headers
are
already
the
case
that
you
can
send
anything
you
want,
because
it's
an
optional
to
support
header.
You
know
you
can
ignore
the
header.
C
So
what
doesn't
really
make
sense
and
in
media
types
specifically,
you
know
slash,
store
I
personally
feel
like
there's
a
lot
of
public
protocol
design
going
on
here.
I,
don't
think
it
actually
gives
us
server
information.
That's
practically
useful,
I,
see
browsers
you
sending
a
lot
of
store
this
and
story
that
I
don't
know.
If
actual
use
cases
for
that
I'd
love
to
hear.
If
people
have
them,
then
we're
enjoying
I
think
might
have
feelings
about
this.
C
D
C
P
Guess
it
could
be
interesting
if
you
have,
as
you
should,
multiple
media
types
for
margin
for
different
things
and
your
REST
API,
and
also
have
support
for
both
Jason
and
something
else's
your
API,
so
the
client
could
say
I
want
the
Jason
variant
of
all
the
responses
that
you
can
produce.
Now
you
will
say
there
should
be
only
one
format
anyway.
So
syntax
I
was
just
saying
that
this
has
been
asked
quite
a
few
times
on
Stack
Overflow,
whether
the
video
range
production
allows
that
use
which
it
doesn't.
C
C
N
H
H
Has
some
implementation
experience
was
just
positive
impacts
of
the
key
values,
even
without
the
store
and
with
a
relative
complexity,
to
the
benefit
they
give,
and
this
is
a
question
that
any
time
you
do
a
priority
scheme
right,
if
we
just
built
something
that
is,
you
know
more
on
really
than
it
is
beneficial
so
I'm.
This
might
be
a
good
time
to
get
a
feel
for
how
fundamental
this
is
to
our
system
other
than
it
being.
You
know,
document
in
different
years,
right.
C
Well,
I,
trying
to
suspect
that
you
know
the
most
we
could
do
here
is
to
just
write
some
words
saying
that
you
know
don't
think
you're
getting
what
you're
not
getting
from
this
and
cautioning
misuse
of
it.
I'm
curious
when
you're
talking
about
HTTP
API
use.
Are
you
just
talking
about
content,
negotiation
and
and
then
perhaps
you
values?
Are
you
talking
about
actually
using
the
local.
N
Have
but
if
they
weren't
I'm,
not
sure
if
they
were
in
the
subject,
you
know
sub
sub
of
subtype,
star
values
or
if
it
was
just
in
the
version
field.
So
I
can't
really
I
can't
remember
if
it
was
a
either
one
of
those
but
I
have
seen
it
used,
but
not
I.
Don't
think
in
the
way
that
we
formally
approved
okay.
C
N
I
should
note
that
the
original
use
of
this
was
when
browsers
had
an
external
gift
viewer,
an
external
image
viewer.
They
would
actually
say
image
slash
star
because
they
didn't
know
what
the
properties
of
the
image
viewer
was
were.
They
would
just
send
the
raw
data
to
the
image
viewer
and
whether
or
not
it
supported
a
given
image
type
was
you
know
who
care.
B
O
C
C
C
D
C
C
O
This
is
a
difficult
question,
because
from
many
ways
what
we
are
talking
about,
yet
the
semantic
layer
in
these
documents
and
focusing
our
efforts
there
and
we're
not
really
trying
to
do
HTTP
tube
is
that's
that
constrains
what
we
can
talk
about
really
and
in
this
context,
a
little
bit
other
than
simply
explaining
more
about
what
the
intent
was
there
when
we
actually
get
to
the
point,
you're,
actually
defining
new
things
or
putting
new
constraints
on
it.
That's
where
I
get
nervous
about
that
I.
C
C
H
Good
thing,
I
think
that's
probably
a
little
bit
more
thinking
to
do
on
this
front
right
and
the
way
this
issue
is
described.
It's
sort
of
pointed
at
the
specifically
pH
extension
of
its
notion
of
extensions
is
one
may
actually
be
an
interesting
thing
to
define
as
well
like
that's
not
I.
Think
7
appears
in
poor
anymore,
like
how
to
extend
this,
like,
sometimes
are
defined
as
being
generic,
but
you
know
the
thought
of
you
know
what
an
extension
means
in
a
core
concept
versus
a
transport
concept
might
be
an
interesting
thing
to
to
write.
O
C
C
P
Yeah,
that's
what
I
wanted
to
go
and
it's
to
be
essentially
after
story
that,
if
both
sides
agree
on
an
extension,
the
extension
can
essentially
modify
anything
and
that
causes
confusion.
If
you
want
to
compose
different
extensions
right
because
they
might
be
in
conflict
about
what
they
modified,
I
think
what
Hetrick
did
for
connected
WebSockets
is
okay,
because
that
was
a
very
specific
thing
to
get
to
their
web.
Sockets
knows
and
there's
no
other
thing
that
you
want
would
want
to
combine
it
this,
but
that's
probably
not
for
other
types
of
extensions.
B
H
H
You
see
in
between
implementations
is
that
there
was
no
guns,
and
then
some
of
the
deviation
is
because
you
know
that's
what
I
meant
for
there
for
the
application
specific
behavior,
but
some
guidance
would
be
helpful.
We've
had
a
million
bugs
around
us
over
the
years
and
doing
a
better
job.
Now
the
notion
of
what
to
copy
and
what
not
to
copy
is
I
would
appreciate,
taking
a
pass
out
to
see
if
we
can
get
consensus
on
I,
don't
know
if
how
Brides
feeling
over
on
the
chrome
team
this
one.
It's
not
a
pinpoint.
C
C
C
C
It
would
be
much
better
for
the
HTTP
wonder
if
we
just
had
a
separate
HTTP
had
a
registry
I
think
that
would
be
incredibly
value
and
actually
get
us
better
participation
in
their
industry.
So
I
have
a
draft
that
I
put
out
a
while
back
anything
really
sorry
I
had
a
draft
a
little
while
back
that
I
Kim
the
approvals.
Sorry.
Can
you
prove
this.
G
G
C
C
D
O
The
if
the
registration
rollers
are
sufficiently
different,
and
so
one
thing
we
might
consider
bilious
is
making
the
process
by
which
new
entries
are
entered
into
this
table
a
little
smoother,
because
one
of
the
things
that
I
find
is
that
this
is
a
very
small
not
particularly
well
represented
subset
of
the
things
that
are
actually
out
there.
Wikipedia.
O
C
Is
the
direct
that
I
put
out
is
not
just
separate
registries?
It's
also
talking
about
procedure
and
I'm
happy
to
just
say:
that's
a
draft
or
groups
consideration
I'm
happy
to
put
energy
into
it.
Can
we
just
consider
adoption
and
see
where
we
go,
knowing
that
the
adoption
does
not
mean
a
separate?
Obviously
it
means
voting
this
approach
into
core
ready.
That
said,
that's
boarding,
I.
B
B
K
B
H
C
C
C
B
C
H
What
really
should
be
proof
light
judgment
calls
so
I
appreciate
that
they're
doing
all
the
heavy
lifting
but
I
do
want
to
make
sure
this
thing
gets
its
broad
and
the
way
I
will
judge
that
is
in
discussion
on
issues
as
they
come
up
on
the
list
and
people
that
nervous
new
things
in
the
text.
So
please
to
make
that
happen
if
we
want
to
get
this
done,
but
thank
you
thank
you
and
I.
Think
that's
gonna
wrap
us
up
LaDonna.
Yes,.