►
From YouTube: IETF102-CBOR-20180717-1330
Description
CBOR meeting session at IETF102
2018/07/17 1330
https://datatracker.ietf.org/meeting/102/proceedings/
A
B
You
know
he's
on
me
deck
or
we
have
a
few.
We
have
a
couple
of
other
people
in
jabber,
but
it
looks
like
some
like
well.
Klaus
is
in
both
we
have
matt
is
in
the
room
yeah.
We
have
Jeff,
yes,
Jeff
PC,
but
I.
Don't
Jeff.
Are
you
on
audio
as
well?
Let's
see
if
he
responds
I.
Occasionally
we're
on
this
slide
info
would
be
nice,
but
not
required.
Okay,
so
Jeff
is
hearing
the
audio.
B
D
C
Great
so
you've
seen
it
a
couple
of
times,
but
if
you
haven't,
please
take
a
moment
to
read
it
as
a
reminder:
minutes
are
taking.
Thank
you.
This
meeting
is
recorded
and
your
present
it's
logged.
There
are
the
blue
sheets
over
there.
You
can
start
putting
in
your
name.
This
is
the
agenda
for
today.
First
of
all,
introduction
on
the
status
by
us
and
then
custom
will
take
on
to
talk
about
what
is
left
in
CDL
and
in
seaboard,
and
then
we
have
a
little
bit
of
flex
time.
C
What
is
left
of
the
review
comments
that
we're
done
for
the
working
group
last
call
before
last
meeting
and
what
are
the
open
issues
left
in
the
key
table
and
then
for
the
1749
this
draft
it
was
last
updated
before
IDF
101,
so
it
wasn't
updated
for
this
meeting
and
I
will
see
what's
the
state
test
and
also
what's
the
status
of
the
discussions
in
the
main
list
and
in
the
github
about
it.
So
I
think
you
can
take
it
Gaston,
while
I
change
the
presentation.
E
B
A
B
A
So
we
are
already
late
and
we
don't
want
to
be
infinitely
late
and
a
third
thing
that
we
are
supposed
to
do,
but
only
in
a
limited
way.
So
far
we
were
a
tree
charter
at
some
point
is
look
at
see
what
tags
tags
are
one
of
the
two
extension
mechanisms,
extensibility
mechanisms
of
people.
There
is
another
one,
the
simple
data
items
that
really
hasn't
you
been
exercising
yet
in
the
last
five
year.
So
that's
not
where
the
focuses
most
people
want
to
define
tags
and
that's
something
that
we
also
are
looking
at
so
yeah.
A
Let's
do
this
in
three
steps
so
that
the
tags
part
I
have
fits
into
the
flex
time
part
on
the
agenda.
So
let's
first
talk
about
CD,
DL
and
I
need
to
issue
an
apology
for
Hank,
who
is
at
the
trusted
execution
environments
meeting.
So
it's
really
hard
to
get
him
because
he's
usually
involved
in
another
work
group
at
the
same
time,
so
I'm
the
messenger
here
today
again
so
at
ITF,
one
or
two,
we
had
a
nice
idea
of
what
would
be
done
until
May
and
then
we
would
be
finished.
A
What
actually
happened
is
that
directly
after
the
ITF
I
got
sick
and,
and
that
delayed
the
process,
why
do
I?
So
it
hasn't
all
happened.
The
the
over
three
that
essentially
takes
the
London
results
is
two
weeks
old
and
let
me
quickly
go
through
what
we
did
there
and
then
through
what
we
missed
doing.
A
Unfortunately,
because
of
course
this
was
one
of
ten
internet
drafts
that
had
to
get
ready
for
this
meeting,
so
they
have
been
a
number
of
really
useful
editorial
comments
and
I
hope
I
see
a
few
more
editorial
comments
and
in
the
next
couple
of
weeks,
and
so
we
clarified
the
definition
of
group
entry
is
clarified
how
their
words
behave.
We
fixed
the
inheritance
example
to
actually
be
legal
and
had
a
few
typos,
so
that's
really
useful.
A
We
know
we
noticed
that
we
that
the
usage
of
C
bar
out
there
had
drawn
a
thicker
line
between
integers
and
floating
point
values.
Then
we
thought
we
what
five
years
ago,
so
in
a
few
places
we
made
this
explicit
now.
Also,
for
instance,
a
one
doesn't
match
a
one
point:
zero
and
a
one
point:
zero
doesn't
one.
A
A
So
it's
really
useful
and
finally
I
just
went
ahead
and
added
straumann
for
a
control
operator
registry
for
a
yenna
registry
for
a
control
of
our
operators
without
trying
to
also
define
a
policy-
and
that's
one
of
my
next
slides-
whether
we
actually
want
to
do
this
and
that's
how
there
is
an
editor's
copy
on
the
github
side,
github.com
/c
board,
WG
I
should
have
put
that
on
the
slide
and
there
are
a
few
more
things
in
there.
We
didn't
define
what
a
byte
is.
A
There
are
still
people
out
there
that
think
we
have
to
use
20th
century
world
octet
because
byte
is
not
defined,
so
we
just
defined
what
about
this
and
the
predefined?
What
the
target
types
are
that
the
size
control
operator
actually
can
control.
So
it's
for
byte
strings.
Think
for
text
rings.
I
didn't
remembered
right
now,
but
it's
written
up
and
for
unsigned
integer.
So
that's
all
there
is
defined
for
and
if
you
think
it
should
be
defined
for
something
else,
then
we
should
act
quickly.
A
Unfortunately,
there
was
one
review
that
that
was
so
good
that
we
sing
at
it
out
and
and
didn't
put
it
into
issues,
and
you
always
didn't
do
something
like
that,
and
there
was
Jen's
video
so
that
there
are
like
10
things
in
gems
review
that
we
still
have
to
act
on
and
I
mean
we
all
have
been
there.
Not
I
mean
not
turning
something
into
issues,
always
fights
you
okay.
So
there
are
still
a
few
issues
open
on
github
and
some
of
them
just
aren't
going
to
be
done
in
the
current
version
of
CTD
air.
A
So
there
was
a
request
for
co-occurrence
constraints.
It
actually
turns
out
that
many
co-occurrence
constraints
fit
naturally
with
a
grammar
foundation
of
obsidian.
So
if
you
want
to
say,
one
map
key
always
occurs
together
with
another
map.
Here
they
both
don't
occur,
that's
easy
to
say,
but
more
complicated
ones
like
there
need
to
be
two
timestamps
in
there.
That
starts
from
and
the
ends
by
timestamp
and
the
second
time
stamp
has
to
be
larger
than
the
first
time
step.
A
We
cannot
say
that
inside
yet
because
it's
a
grammar,
it's
not
an
executable
language
where
you
would
be
able
to
put
in
predicates
like
that.
So
that's
something
I
want
to
get
back
to
later
on
when
we
are
done
with
the
current
spec,
but
that
we
will
cannot
do
today.
I
think
there
also
was
an
editorial
comment
that
we
should
be
more
explicit
about
the
history
of
the
document
and
its
contributors
and
I.
Think
that
that's
a
very.
E
A
A
There
is
one
issue
about
specifying
group
matching
I.
Think
the
current
spec
does
this
in
Appendix
C,
but
maybe
that
has
to
be
improved
and
that's
one
of
the
issues
that
is
tagged
as
had
wanted,
because
somebody
has
to
read
this,
who
doesn't
already
know
how
it
works,
because
for
me
it's
it's
perfectly
obvious
how
it
works,
but
I
happen
to
know
how
it
works.
So
we
want
you
don't
want
to
write
signage
documents
as
trapdoor
functions
that
you
only
can
understand
if
you
already
know
what
what's
in
there.
A
G
Can
you
hear
me
yes,
yes,
I
I
think
this
could
be
handled
editorially
if
everyone
agrees
that
that
parse
expression,
grammars
are
kind
of
the
basis,
but
I
think
it's
it's
kind
of
more
important
more
than
just
out
of
brush
it
off
for
next
version,
because
there's
no
definition
of
what
parse
expression
grammars
mean
in
the
spec
and
there's
no
reference
to
them
either.
So,
if
someone
doesn't
doesn't
know
what
they
are,
this
there's
no
standard
definition.
A
A
Think
no
think
knows
what
a
pass
expression
is
party
you
can
canal.
He
cannot
get
me
okay,
so
yeah.
Maybe
we
can
take
that
as
one
item
for
the
notes
that
we
find
a
good
reference
for
that
there
needs
to
be
a
textbook
somewhere
and
if
we
need
to
recite
the
dragon
book
that
you
recite
the
dragon
book
that
certainly
can
be
done
and
and.
G
D
A
A
A
The
control
operator
takes
one
type,
the
target
type
and
the
control
type
and
generates
type
according
to
the
way
the
control
operator
has
defined,
and
we
had
a
recent
discussion
on
the
mailing
list
and
I.
Think
one
observation
was:
whenever
you
have
these,
these
registered
things,
it's
really
hard
to
collect
things
together
and
and
know
what
is
the
current
state
so
I
have
this
specification
here
and
is
this
a
city
a
specification?
A
Yes
and
it
needs
this
extension
and
that
extension
and
a
third
one
so
that
of
course,
is
an
operational
problem
in
in
working
with
extensions.
On
the
other
hand,
the
observation
is,
we
probably
are
not
going
to
do
all
useful
control
operators
of
this
kind
here,
because
there
are
other
SDOs
that
have
very
specific
things
like
their
own
legacy,
binary
syntax,
and
they
want
to
transport
this
in
in
SIBO
data
structure.
A
A
H
In
general,
my
understanding,
their
designated
expert
or
specification
required
is
exactly
the
type
when
you
cannot
algorithmically
prescribe
how
you
will
reject
or
accept
registration,
so
I
think
defining
the
box
and
defining
eliminations
and
what
expert
is
allowed
and
not
allowed
explicitly
to
do.
I
think
it's
it's
it's
right
right
to
go
about
it.
So.
I
B
J
Sean
leonard,
so
I
guess
I
don't
see
the
need
to
define
a
registry
for
control
operators.
Now
I
think,
there's
a
lot
of
issues
here
that
will
just
delay
the
release
of
this
specification
and
I'm
not
sure
actually
how
much
people
are
using
or
will
use
control
operators
when
defining
specifications
and
not
regret
putting
the
control
operator
there
in
the
first
place,
because
in
a
revision
of
their
protocol,
they
may
want
to
change
what
the
restrictions
are.
A
Yeah
one
problem
with
specification
required
is
also
that
many
people
do
not
read
RFC
681,
whatever
and
and
don't
know
what
it
actually
means
and
think
if
I
draw
up
a
piece
of
paper
and
tend
that
to
the
listing
and
expert,
then
I'm
done
and
that's
not
what
is
minute
so
that
the
the
expert
actually
has
the
prerogative
to
do
some
quality
control
on
that
specification.
So
if
the
specification
is
not
good
enough
for
insurability,
it's
not
a
specification.
Is
that
that's
how
I
read
81?
What's
the
number
26
I.
K
L
However,
I
will
also
give
the
caveat
that
we
are
going
to
choose
the
wrong
way
to
do
this.
Every
working
group
gets
it
wrong.
You
know
the
I
can't
think
of
a
single
instance
where
someone
said:
oh,
we're,
gonna
do
the
registry
this
way
this
way,
and
everyone
was
happy
so
and
I-
think
we're
within
the
realm.
So
I'm
not
saying
oh,
let's,
let's
argue
about
this
longer,
we're
gonna,
get
it
wrong,
but
I
think
let's
get
it
at
least
meaning.
Let's,
let's
set
up
the
registry
in
this
document.
So.
B
L
If
we
have
an
I
on
a
registry
with
an
expert
reviewer
and
the
expert
reviewers,
not
insane,
then
if
someone
comes
to
them
with
a
piece
of
paper
where
it's
written
on
the
expert
reviewer
says:
oh
god
everyone's
going
to
understand
this,
we
don't
need
the
piece
of
paper.
That's
fine,
most
expert
reviewers
would
say
where's
your
draft
where's
your
document
and
it
becomes
the
the
moral
equivalent
of
specification
required,
but
not
exactly
under
the
exact
you
know
terms
or
whatever
it's
more
trusting
a
human
than
trusting
the
process.
H
B
L
Right,
especially
because
it
almost
always
includes
the
most
important
piece
of
information,
which
is
the
email
address
of
the
author
of
the
current
email
address
the
author.
They
always
change,
but
then
somebody
who
wants
to
use
it
if
they
don't
understand
the
piece
of
paper
they
understand,
they
can
call
up
somebody
and
often
it
turns
out
like
no
don't
don't
use
this,
make
up
your
own
gym.
I
So
I
just
sent
this
mail
to
the
mailing
list
earlier
today,
which
is,
if
we're
going
to
do
a
CD,
DL
I'd
rather
hold
off
on
this
registry.
Until
we
say,
we've
got
a
lot
of
these
things
that
we
know
about,
and
we
don't
want
to
standardize
them
or
we're
not
planning
to
publish
another
version
of
the
other
language.
L
Paul
Hoffman,
that's
not
the
way.
I
read
your
message,
so
thank
you
for
clarifying
I,
don't
agree.
Well,
I
mean
we're
trying
to
pick
the
future.
So
that's
bad,
but
from
my
experience
with
IPSec
Ike
v2
I
think
it
is
fine
to
create
a
registry
and
then
let
the
working
group
or
if
the
working
group
has
already
dissolved
the
I.
L
You
know
the
remnants
and
the
ietf
decide
at
what
point
a
v2
could
be
useful
and
it
collects
all
of
the
things
that
are
in
the
registry
and
at
leat
it
either
redefines
them
or
just
points
to
and
says
my
God
look
at
this
registry.
You
know
without
even
knowing
how
you
know
how
much
will
be
in
it
at
the
time
the
person
reads:
it
III
think
us
predicted
trying
to
predict
when
when
or
why
we
would
do
a
CDL
to
is
really
hard.
So,
let's
not
let's
not
make
that
a
gating
factor
was.
A
It
there
are
some
things
in
the
spec:
you
cannot
fix
with
a
control
operator
and
one
is
the
reach
of
cuts
and
I.
Think
that
is
a
pretty
strong
incentive
to
actually
go
ahead
during
yesteryear
to
and
I
also
read
that
bundling
up
things
I
have
a
slide.
The
end
of
my
slide
said
about
bundling
up
things
bundling
up
things
in
in
RC
at
some
point
in
time
after
five
years
or
so
it's
always
a
good
idea.
So
that's
another
reason
to
do
a
2.0,
but
either
way
does.
A
The
point
of
having
a
registry
is
not
internally
facing
its
External
Affairs
right,
so
it's
telling
people
you
can
use
CV
DL,
even
though
we
don't
have
your
favorite
control
operators
yet,
and
do
you
don't
have
to
wait
five
years
for
the
IETF
to
it
will
standardize
it?
You
can
do
it
on
your
own.
That's
really
the
important
part
of
the
message
in
my
point
of
view,
and
that's
also
the
reason
why
I,
like
the
term
specification
required
because
it
packages
in
very
few
words
that
they
do
have
to
do
somewhere.
A
B
A
E
A
N
Also
just
highlighting
you
have
a
bunch
of
clear
vision
of
what
is
to
be
included
in
to
that
all
and
then-
and
it
is
I
think
sometimes
it
will
be
of
the
same
size
to
that
already.
So
there
are
some
interesting
problems
we
want
to
solve
and
also
regarding
specification
required.
I
think
that
I'm
not
understand
alternative
of
having
the
expert
review
and
there's
some
code
of
conduct
that
there's
basically
requiring
you
know,
educate
to
have
that
I'll.
B
Explain
it
so
expert
review,
says
you
have
a
person
or
a
group
of
people
assigned
to
earlier
requests,
and
they
are
given
some
sort
of
guidance
document
about
what
they
should
consider
but
know
nothing
other
than
a
request
for
registration
is
required.
Okay
specification
required
adds
to
that.
The
requirement
that
you
have
a
publicly
available
stable
specification,
okay,
that
the
reviewer
that
the
reviewer
and
other
people
can.
N
E
N
M
It
says
Chris
Comcast,
the
it's
worth,
noting
that
people
are
going
to
extend
it
if
we
don't
give
them
a
registry
to
share
and
put
their
extensions
in
there.
Ten.
What
we
have
is
a
mess,
not
you
know.
So
what
we're
really
talking
about
it?
How
do
we
organize
the
boxes
and
I
hear
a
lot
of
support,
for
everybody
wants
people
to
write
specifications,
but
it's
worth
noting
that
if
we
don't
give
them
a
easy
way
to
contribute
and
share
and
whatnot
they're,
just
gonna
extend
it
anyways
and
who
knows
what
the
result
will
be.
C
O
O
H
A
B
A
P
Just
along
the
same
lines,
Dave,
Walter,
Meyer,
I,
think
I
think
having
a
registry.
It
doesn't
prevent
folks
from
experimenting
and
and
and
you're
playing
around
with
the
language,
but
when
there
are
two
parties
that
want
to
do
the
same
thing,
it
gives
them
a
way
to
agree
on
that
and
that
adds
value.
H
Alexei
Melnikov,
the
other
comment
I'd
like
to
make
is
that
by
picking
one
or
more
experts
you
can
somewhat
control
the
outcome
of
this
I
mean
that
might
be
a
bit
obvious,
but
you
can
specify
actually
in
a
document
what
kind
of
restrictions
or
what
kind
of
encouragement
he
will
make
to
the
experts.
But
you
can
also
select
experts
with
potentially
divergent
views
to
help.
You
know
allow
for
wider
types
of
control,
operators
for
the.
H
M
All
of
the
arguments
for
designated
expert
that
I've
heard
and
I
say
all
okay.
Most
of
the
arguments
for
designated
expert
basically
boil
down
to
well
designated
experts
will
encourage
good
specs,
so
I
I'm,
hearing
a
lot
of
support
for
specifications
and
I
think
I
would
support
specifications
and
if
there
are
people
who
think
that
we
should
do
designated
experts
specifically
so
that
we
can
have
control
operators
that
don't
have
specs
I'll
be
really
interested
in
hearing
the
argument.
Well,.
B
Paul's
argument,
if
I
may
save
them,
the
trouble
of
jumping
up
is
that
there
may
be
registration
requests.
That
really
don't
need
much
of
a
specification
and
simplifying
that
what
it
would
be
helpful
and
the
designated
expert
can
always
say
sorry.
Your
request
is
complicated
enough.
That
I'd
like
you
to
write
up
the
specification
right.
Q
I
I
A
Think
that's
a
good
point
actually
for
the
the
text
in
DNA
considerations
section.
So
some
organization
might
want
to
write
a
control
operator
that
uses
some
information
that
actually
that
is
actually
confidential
and
well.
The
specification
could
be
pretty
short
and
say
this
uses
confidential
specification
so
and
so,
and
then
we're
done
with
it
and
the
text
could
be
explicit
that
this
is
something
that
we
do
want
to
allow
in
cases
that
make
sense
yeah.
O
K
M
H
Having
you
know
allowing
for
secret
specification
references,
well-known
references
to
secret
specifications
but
again
I'm,
just
comparing
this
with
media
types
with
media
types,
a
specification
is
encouraged,
is
not
required,
and
sometimes
people
will
say
an
expert.
A
sort
of
almost
bound
by
NDA
by
but
expert
has
to
look
at
be
able
to
look
at
the
specification
to
decide
whether
it's
it's
a
good
thing.
So
yeah
now.
A
B
A
B
B
A
It's,
of
course,
a
control
operator,
and
there
are
a
number
of
control
operators
that
were
added
pretty
late
in
the
process,
because
we,
when
doing
real-world
specifications
we
found
we
had
a
problem.
We
couldn't
say
that
something
needed
to
be
larger
than
three,
but
on
the
other
end,
so
we
added
less
than
a
greater
equal
greater
than
less
and,
of
course,
usually
you
add
an
equivalent,
not
equal
as
well.
A
The
more
likely
uses
less
than
and
greater
than
and
these
are
defined
for
numeric
types,
I'm,
not
aware
of
another
type
in
CDL.
That
really
could
make
useful
sense
of
this
I
mean
we
could
define
the
string
comparison
for
this,
but
really
nobody
needs
that.
So
we
also
have
regular
expressions
for
most
of
the
cases
where
that
would
be
useful,
so
I
think
it's
very
clear
that
the
inequality
is
less
than
greater
equal
and
so
on
that
these
stay
confined
to
numeric
types.
A
So
then
we
looked
out
there
and
in
front
oops.
People
are
using
them
with
strings
in
particular,
they're,
not
equal.
What
the
equal
one
doesn't
really
make
a
lot
of
sense,
but
they're,
not
the
not
equal
one
is
actually
useful
application,
so
there
are
actually
two
variants
that
you
might
want
to
use.
A
One
is
the
numeric
equality
where
you
have
a
constant
somewhere
that
happens
to
be
a
floating
by
constant,
and
you
still
want
to
constrain
some
integer
values
to
not
take
on
that
particular
floating
point
concern,
so
one
two,
five
and
not
equal
3.0
would
be
one
two
four
five,
because
not
equal
is
defined
on
a
numeric
basis
or
you
could
say
something
like
any
number
that
is
equal
to
3,
which
is
a
short
way
of
saying,
3
or
3.0.
So
you
don't
have
to
define
a
constant
twice
so
I
think
these
are
reasonable,
reasonable
applications.
B
K
E
A
I
A
The
point
we
have
Amadou,
you
need
the
rest
we
have
set
the
base.
Language
makes
a
strict
distinction
and
this
control
operator
is
expressly
defined
not
to
make
it.
So
exactly
is
it's
exactly
the
escape
hatch
you
can
use
when
you
are
unhappy
with
that,
the
ability
of
the
basic
it's
meant
to
be
the
other
way.
A
K
B
L
A
I
read
it
I
mean
this
is
discussing
something
for
a
napkin,
but
I
just
wanted
to
open
up
the
choices.
So
we
can,
at
the
end,
agree
on
something.
So
imagine
that
this
says
something
like
we
have
a
solution
where
we're
any
and
Q
the
numeric
operators
and
then,
of
course
the
question
is:
do
we
actually
need
this
on
the
structural
side
of
the
general
structure
inside
as
well,
and
we
also
have
the
default,
which
is
essentially
dot
and
E,
but
Ria's
used
for
a
lot
of
things
that
are
not
number.
B
A
Don't
you
always
know
the
the
question
is
when
a
specification
has
a
composite
construct
for
a
number
that
does
something
we
protocol
types
are
actually
not
used
to
that.
We
don't
deal
in
numbers,
we
deal
in
integers,
but
Langley
doesn't
even
have
floating-point.
Imagine
that
so
it's
a
little
bit
unusual
for
us,
but
if
you
have
a
composite
construct
that
already
has
a
useful
abstraction
in
the
rest
of
your
specification.
And
then
you
want
to
grab
into
this
construct
and
move
one
numeric
value,
which
may
have
two
representations
three
and
3.0.
That
would
be
okay,.
A
A
Donny
Q
is
very
rarely
used,
I
mean
almost
always
when,
when
you
want
something
to
be
equal
to
something,
you
just
write
the
second
something
so
that
I
had
to
think
a
little
before.
I
came
up
with
the
example
there.
So
the
the
actual
case
where
we
have
to
worry
about
backward
compatible
compatibility
is
not
any.
But
of
course
he
want
these
to
be
symmetric
or
we
violate
the
principle
of
least
surprise.
A
So
what
I
actually
have
seen
in
specifications
excuse
the
thing
that
looks
like
a
period
that
that's
just
a
bullet
label
dot
an
e.
True,
that's
a
very
common
construct,
and
we
probably
want
to
allow
something
like
that,
but
we
may
have
to
for
specification,
like
writers,
to
use
a
different
control
operator
yeah
so
that
that
we
try
to
avoid
that.
But
maybe
here
it
is
unavoidable.
So
we
want
to
keep
an
e
for
the
numeric
case.
You
would
come
up
with
a
new
one
for
the
structure
case
and
maybe
another
reason
for
that
is.
A
G
D
G
A
A
J
Okay,
so
I
I,
pretty
much
agree
with
with
Jeffrey
about
the
points
that
he
made.
I
think
I
just
want
to
point
out.
Maybe
a
broader
concern
I
have,
which
is
that,
as
we
said,
that
EQ
is
there
mostly
for
symmetry,
but
the
actual
utility
of
it
is
very
dubious
in
specification.
Writing
at
this
point
and
I
think
that
that
raises
the
question
of
you
know.
J
Maybe
we're
trying
to
define
too
much
or
trying
to
do
too
much
with
control
operators
in
general
when
they
just
don't
have
a
utility
I,
don't
want
to
pick
on
dot-ne
because
it
does
seem,
people
are
actually
using
it
quite
a
bit,
but
there
are
a
number
of
I
mean
at
least
you
know
for
dot,
EQ
I
wonder
why
do
we
even
put
it
in
there
for
not
sure
when
it
seems
like
nobody's
actually
using
it,
and
similar
arguments
could
be
made
with
a
couple
of
the
other
control
operators?
Do.
A
A
When
designing
protocols,
I
always
have
been
when
I
started
to
make
policy
level
judgement
level
decisions.
Is
this
particularly
useful
or
not?
People
will
find
users
that
you
have
no
idea.
They
exist,
okay
and
as
long
as
it's
a
simple
it
does
here
in
this
case,
I
think
having
a
symmetry
is
part.
Okay,
symmetry
is
always
a
good
argument.
Okay,
thank
you.
A
So
when
you
represent
something
you
you
take
it
down
one
level
on
the
hierarchy
of
different
modeling
mechanisms
that
we
have
so
you
represent
an
information
model
in
a
data
model
represented
at
your
model
in
the
serialization
and
so
on,
so
I
think
a
better
term
would
be
serialization
variance,
because
that's
really
the
issue
that
we
were
having
and
I
think
on
the
mailing
list.
We
discussed
about
serialization
variants
in
protocol
definitions
and
implementations,
which
are
really
two
two
separate
issues
that
just
have
to
work
together
and
a
good
way.
A
A
Allow
to
say
this
verb,
canonicalization,
okay
and
now
I
need
mouthwash.
So
the
idea
behind
consistent
encoding
is
you,
don't
have
serialization
variance
so
for
every
data
model
item
there's
exactly
one
way
to
encode
it,
and
this
is
what
the
chunk
consistent
and
coning
really
tries
to
express
and
it
turns
out
in
a
language
like
a
sieve.
A
Now
there
are
other
places
where
there
are
series,
Asian
variants
and
the
current
document
doesn't
quite
go
so
far
and
say
there
is
a
preferred
encoding
for
this
case,
but
it
it
kind
of
is
written
with
the
subtext
that
there
is,
and
maybe
we
should
make
this
a
little
bit
more
explicit
and
say
that's
just
the
way,
you're
supposed
to
encode
things.
And
yes,
if
you
have
an
implementation
constraint
that
doesn't
allow
you
to
do
that,
then
we
will
have
to
deal
with
that.
A
So
it's
a
little
bit
little
piece
of
posterior
principle
here
be
deliberate.
What
you
accept
be
conservative,
in
what
you
send
and
the
preferred
encoding
will
essentially
be
the
conservative
part,
and
but
we
would
not
stop
it,
saying
be
liberal
in
what
you
accept.
So
we
would
continue
to
accept
the
other
variants,
which
is
fine
as
us
different
from
utf-8
nucleation
was
made
to
not
accept
them,
because
people
found
out
only
after
years
of
using
utf-8
that
they
could
be
used
for
fun,
interesting
attacks.
A
A
Asian
variants,
so
let's
get
a
little
bit
up
the
stack
again
and
see
what
what
are
the
actual
series
Asian
choices
that
are
relevant
in
practice
and
not
just
talking
about
implementation
is
talking
about
specifications.
So
the
specification
for
instance,
might
say
there
are
no
floating-point
numbers
in
this
specification,
and
normally
you
don't
have
to
write
that,
because
your
data
model
specification
already
contains
no
floating-point
numbers.
A
So
you
actually
don't
have
to
make
this
serialization
choice,
and
once
you
say
there
are
no
floating-point
number
of
civilizations,
you
cannot
have
a
data
model
that
allows
that,
because
you
wouldn't
know
how
to
serialize
them.
So
that's
that's
a
choice
that
can
be
handled
entirely
on
the
datum
level,
so
we
actually
can
express
this
in
CDL
not
explicitly,
but
by
just
using
for
it
now.
A
One
other
thing
that
actually
makes
a
lot
of
sense
in
some
embedded
violence
is
to
say
we
disallow
sixty
four-bit
for
any
point
numbers,
because
we
only
have
software
floating-point
and
the
software
protocol
and
library
use
only
supports
32-bit.
So
in
this
case
you
actually
make
a
serialization
choice.
A
If
your
data
model
only
uses
numbers
that
can
be
represented
as
32-bit
integers
and
you
use
preferred
encoding,
then
again
you
can
express
this
at
the
data
modeler.
So
again,
you
don't
need
something
in
CDL
to
express
the
serialization
choice,
because
CDL
is
about
the
data
model.
Anyway,
you
can
do
this
by
by
just
saying
which
numbers
you
use:
support
that
example:
disallowing
64-bit,
integer
string,
lengths
item,
counts
and
tags.
That's
also
a
pretty
obvious
thing
you
might
want
to
do
in
an
embedded
system,
so
that's
also
something
that
you
would
get.
A
A
Q
Jones
Microsoft
the
fighter
2
C
tab
specification,
which
uses
C
bar
for
its
data
representation,
addresses
a
number
of
cases
where
data,
in
particular
for
extensions,
has
to
be
transformed,
from
jason,
to
see
more
and
in
the
other
direction
as
well
from
seaboard
to
Jason,
because
Jason
represents
all
numbers
of
64
bit
floats
that
specification
minimal,
scrupulously
treason.
Yes,
sorry
excuse
me,
javascript
represents
all
numbers.
64-Bit
floats
that
specification
made
the
choice
that,
when
converting
JavaScript
values
to
Seaborn,
we
would
always
use
the
64-bit
floods.
P
P
You
know
the
tooling
problem
in
general
and
this
in
this
space,
I
guess
taking
my
chair
hat
off
I
think
we
want
to
make
decisions
around
serialization
that
lead
to
more
tooling,
and
it
seems
like
the
more
fragmentation
that
we
get
in
serialization,
the
more
fragmentation
that
we're
going
to
get
around
tooling,
and
that
just
seems
like
a
really
bad
idea.
We
should
probably
focus
our
efforts
on
on
allowing
choices
that
can
be
included
in
C
GDL
directly.
P
You
know
like
disallowing
floating-point
numbers
and
avoiding
as
much
as
we
can
the
kinds
of
choices
that
would
be
bound
to
the
specific
serialization
format.
Just
because
again,
that's
just
gonna
complicate
tooling,
you
know
it's
gonna
mean
you're.
Gonna
have
to
do
a
lot
of
special
case
handling,
and
that
just
seems
like
a
really
bad
idea
from
you
know,
from
a
tooling
and
interoperability
perspectives.
N
N
I'm,
calling
this
encoding
civilization
isn't
see
were
specific.
Okay,
so
talk
about
a
seaboard
I
think
there's
a
benefit
to
disallowing
it
in
a
document
in
the
data
definition
document,
because
in
general,
because
first
of
all
it
would
be
noisy
to
do
all
the
other
controls
that
would
disallow
each
value.
And
again,
if
you
have
extension
points,
you
can
still
mess
it
up
with
your
extension.
So
this
would
inherit
into
extensions
and
that's
important
for
the
constraint
device
not
to
be
I.
Don't
know
backstep
bias
to
for
the
extension
and
that's
important,
I.
R
So
Lawrence
lund,
late,
you
know
on
the
in
constrained
the
highly
constrained
environments
and
the
tooling
is
some
comments
there.
I
would
imagine
like
95%
of
all
implementations
are
going
to
be
fine
with
64-bit,
ants
and
floating-point,
and
they
can
use
the
preferred
in
serialization
variant
if
I'm
using
the
right
term
there.
R
That
seems
reasonable
to
me
and
that
you
know
a
lot
of,
and
we
can
have
a
lot
of
good
tooling
around
that
and
that's
fine,
but
I
think
we
still
have
to
have
the
option
for
I
really
would
like
to
see
us
be
able
to.
You
know,
have
good
interoperable
protocols
that
run
on
like
8-bit
CPUs
that
just
have
no
floating-point,
and
you
know
you
know
less
than
a
kilobyte
of
on-board
RAM
or
something
like
that.
So
those
are
those
to
me
like
seem
like
special
case
tooling,
and
it's
it's
only
a
small
market.
R
A
Okay,
so
what
I
was
trying
to
say
is
we
kind
of
have
everything
95%
of
the
applications
need
I
agree
with
what
I
think
somebody
said
you
can
still
mess
up
in
some
control
operators,
or
so
that's
maybe
a
problem,
but
for
95%
of
all
situations
we
have
what
we
need
if
we
make
it
explicit
that
a
specification
of
a
format
might
contract
contain
additional
English
language
that
contain
constraints.
The
series
ation
I
still
think
that
having
a
preferred
encoding
is
a
good
way
to
reduce
the
number
of
variants
that
actually
sprout
from
that.
A
So
I
know
that
there
are
some
high
speed
applications.
For
instance,
they
always
want
to
send
all
floating-point
numbers
as
64
bits.
So
that's
a
very
specific
serialization
constraint
that
that's
not
compatible
with
another
series
that
you
constraint,
which
says:
don't
use
64
bit,
so
we
are
not
going
to
solve
that
that
problem.
There
are
these
implementations,
bold
it
operates,
but
we
can
make
it
easy
for
the
default
choice
or
for
the
default
case
to
actually
interoperate
and
that's
why
I
think
we
could
should
stick
with
what
we
have
at
the
city
level.
For
now.
A
Maybe
add
something
like
a
series
18
control
language
later
and
on
the
seawall
side
put
in
something
that's
called
preferred
encoding.
Almost
all
of
them
is
in
canonical
encoding
or
strict
encoding
at
the
moment,
so
preferred
encoding
probably
would
be
a
very
short
section.
By
just
saying
do
everything
you
would
do
for
distance
according
except
that
for
maps
you
don't
have
to,
and
then
we
would
be
done
with
it
now.
A
Okay,
if
we
are
done
with
this,
then
I
would
like
to
come
back
to
to
Jim's
comments,
because
we
didn't
tell
them
so
I
think
some
of
the
editorial
stuff
we
already
have
done
and
just
didn't
take
off.
One
editorial
comment
was
there
what
could
be
a
much
better
section
numbering?
Unfortunately,
there
are
30
internet
drafts
out
there
that
reference
specific
sections
of
this
draft.
It
would
be
really
yeah
burden
on
them
to
change
that
and
and
people
not
only
don't
like
changing
that
they
get
it
wrong
or
the
child.
A
I
A
A
A
A
1.0
helpful
is
about
the
word
that
Douglas
Adams
has
another
term
Belgium
everywhere
outside
of
Belgium
is
the
worst
inside
here,
which
you
can
utter,
and
so
helpful
is
something
that
people
on
this
read
would
would
have
positive
connotations,
but
when
an
implementation
of
protocols
that's
become
head
for,
you
have
to
run
because
it's
going
to
hurt
you
and
and
yeah.
So
that's
why
you
have
to
have
ways
of
getting
rid
of
that
health
and
silencing
these
warnings.
So
at
some
point
we
will
need
some
pragmatics.
A
A
A
A
A
A
W3C
x,
SD
probably
has
the
lowest
amount
of
damage
at
this
point
in
time
to
people
out
there,
and
we
can
quickly
solve
it
later
on
by
just
registering
specifications,
especially
education
requirement
document
is
not
going
through
down
roof
checks,
so
it
will
be
easy
to
define
another
operator
that
does
it
for
this.
Yet
so
that's
similar.
A
We
have
this
weird
combination
of
a
grammar
that
is
fundamentally
context.
Insensitive.
Okay,
control
operators
change
this
a
little
bit.
But
fundamentally,
this
is
your
Chomsky
Type
2
engine
that,
but
then
the
grammar
I
create
something
that
then
has
to
go
through.
The
funnel
of
the
civilization
mechanism
that
we
are
using
and
maps
are
definitely
extremely
context
sensitive
because
you
cannot
have
a
duplicate
key.
You
have
to
look
at
your
context
to
find
out
whether
you
can
produce
something
or
not.
A
Unfortunately,
there
is
no
way
around
that
and
I'm
surprised
by
how
well
this
works
by
having
context
insensitive
foundation
and
then
funneling
the
results
through
the
context-sensitive
Maps,
and
now
we
are
introducing
cuts
which
put
in
a
sequence.
Dependence
that
funnels
through
through
the
map
specifications
doesn't
make
the
sequence
in
the
match
relevant,
but
it
makes
the
sequence
in
the
specification
relevant.
So
that
is
something
that
is
a
little
bit
against
the
principle
of
least
surprise.
A
So
it's
a
bit
weird
and
it's
a
little
bit
like
steak
with
chocolate,
sauce
or
something
but
I
think
we
can
get
used
to
the
taste
and
it
seems
to
work.
But
I
can
say
you
that
it
does
cause
a
little
bit
of
concern
when
you,
when
you
think
about
it
for
the
first
time,
because
it's
concepts
that
normally
don't
mix.
B
A
A
B
B
B
A
A
B
A
I
think
that
finishes
the
CDDA
part.
I
have
a
couple
of
slides,
maybe
just
to
pick
your
your
interest
a
little
bit
on
what
may
be
next
and
that
there
are
two
areas
in
which
we
could
work
on
GDL.
If
the
isg
lets
us
and
one
is
just
making
the
language
better,
solving
the
same
problem
and
the
other
would
be
adding
scope
to
it
and
allow
it
to
solve
more
problems.
A
So
in
the
first
group
we
have
things
like
specifications
in
the
LA,
so
people
are
actually
starting
to
compose
pretty
complicated
stuff
out
of
City
debt.
So
you
have
a
the
specification
of
cozy
here
and
then
you
have
a
specification
that
uses
coffee.
How
do
you
use
something
that
that
already
provides
some
structure,
but
you
mix
in
your
own
structure
I
think
we
have
to
examine
this
a
little
bit.
So
when
I
look
at
the
suit
spec
I
found
the
first
place
in
the
IDF,
where
it
would
be
good
to
have
this
composition.
A
I
think
it
can
be
handled
with
1.0,
but
we
may
find
things
we
we
want
to
fix,
but
there
are
also
things
like
module
systems
naming.
If
you
merge
two
specifications
together,
somebody
might
have
used
the
real
name
foo
and
in
both
places,
and
then
you
have
a
program
at
the
model
and
the
other
thing
is
controlled,
adding
controls
and
pretty
sure
we
will
have
quite
a
few
controls,
but
yeah
I
think
we
understand
how
to
do
this.
Now.
The
the
other
group
of
work
might
be
case,
insensitive
whatever
I
mentally
I.
A
The
other
thing
is
going
beyond
structure
or
interpret
II
into
the
lens
of
semantic
interoperability
and
yeah.
So
we
could
look
at
semantic,
augmentation,
so
most
schema
languages.
Ctl
is
not
a
schema
language.
Most
schema
languages
have
a
way
to
actually
add
semantics
to
something.
So
you
can
say
this
is
not
only
in
number,
but
this
is
in
heaven
or
in
microkelvin
or
whatever,
and
this
is.
A
The
calibration
data
for
this
information
can
be
found
there
and
there
are
lots
of
semantics
that
that
could
be
added
here
and
once
you
are
able
to
to
argument
a
specification
by
semantics.
It's
also
much
simpler
are
much
more
powerful
to
do
cogeneration
from
it.
So
right
now,
I'm
still
discouraging
people
to
do
cogeneration
from
Cydia.
Although
there's
one
pretty
interesting
idea
for
the
embedded
space
so
doing
cogeneration
from
CBL
right
into
a
parser
that
has
essentially
code
for
every
single
element.
A
That's
pretty
interesting,
but
the
general
idea,
like
people
are
doing
their
their
Java
from
from
WSDL
or
something
like
that.
That
doesn't
work
too
well
with
this
idea.
Second
saying
those
co-occurrence
constraints
and
more
general
predicate
based
extensions,
so
these
usually
mix
nicely
to
grammar
based
foundation,
and
we
already
have
an
example
with
schema,
Tron
and
relax
ng
where
this
has
work
pretty
well.
So
maybe
we
want
to
do
something
similar,
but
we
don't
have
to
redo
schema
track.
A
P
Wall
tracker
I
mean
if
we
already
have
schema
Tron
one
way
and
I
see
you
have
a
bullet,
maybe
against
this,
but
one
way
of
handling
this
is
to
do
some
binding
of
XPath.
You
know
JSON
or
C
bore,
and
then
you
know
using
using
something
like
schema.
Tron
would
be
sort
of
a
natural
outcome
of
doing
that.
I
wouldn't
require
anything
additional
and
C
D
DL.
So.
A
I
wrote
the
word
C
both
paths
there
and
one
difference
between
zero
path
and
XPath
might
be
that
that
we
don't
run
into
the
problem.
That
XPath
has
that
you
essentially
cannot
compile
it,
because
it
kind
of
has
very
tight
coupling
between
the
characters
that
are
in
the
XPath
and
the
characters
that
are
so.
You
can
write,
pathological,
expose
statements
that
cannot
be
compared.
Remember
we
can
do
something,
that's
a
little
bit
simpler,
yet
maybe
only
has
80%
of
the
functionality,
but
actually
can't
yeah.
A
N
Yeah,
this
is
Hank
from
my
experience.
Speaking
from
personal
experience,
there
are
flavors
of
the
scope.
Xpath
is
used
in
so
typically
you'd,
never
use
everything.
Xpath
is
capable
of
and
finding
right
flavor.
That
is
basically
a
subset
of
extra
functional
capabilities
for
a
sea
bore
path.
I
think
that
could
be
an
ongoing
process.
N
This
is
not
a
project
and
maybe
from
lesson
learned.
We
can
make
multiple
path
versions,
some
of
them
more
applicable
to
a
vendor
devices,
and
some
of
them
may
be
more
at
ease
to
that.
Well,
really
powerful,
so
because
there
can
be
a
kind
of
well
anti
constraint,
note,
environment,
behavior,
that
filter
stuff
and
finding
and
finding
stuff,
and
so
we
really
must
suddenly
wonder
why
didn't
want
to
give
people
good
and
this
explicit
guidance
what
to
use
when
and
where
I
think.
P
A
One
one
interesting
question:
I
wrote
this:
there
is
something
like
a
JSON
path,
the
Jason
point.
They
are
different
things
because
one
really
is
meant
to
be
used
in
a
specification
like
environment
and
one
is
really
pointing
into
an
instance.
So
you
actually
can
bind
it
more
closely
to
what's
in
the
instance
right
now,
and
maybe
we
actually
want
to
do
these
two
yeah.
N
P
A
P
A
P
I
mean
I
know
my
use
case
is
I
need
to
be
able
to
select
a
specific
data
point
out
of
a
C
bar.
You
know
data
set
and
then
do
some
kind
of
comparison
against
against
that,
and
so
what
I
need
a
pathing
language
is
the
ability
to
extract
that
information
and
support
that
kind
of
comparison
differently.
I.
B
Hate
to
interrupt
useful
conversation,
but
this
is
not
for
this
document,
so
we
are.
We
are
running
way
late!
Okay,
let's
take
that
to
another
decision.
Sorry
about
that,
but
we
we
are
supposed
to
have
30
minutes
for
this
and
we
only
have
25
minutes
total
left.
So
can
you
do
this
in
about
15,
and
so
we
can
do
some
wrap-up
stuff,
okay,.
A
Thanks,
okay,
so
the
the
other
document
that
we
want
to
do
here
is
the
RFC
1749
this
and
just
to
remind
people.
So
this
was
not
meant
to
reinvent
the
language
and
re-examine
all
the
decisions
and
do
a
second
system
syndrome,
but
really
mostly
to
document
interoperability
or
issues,
make
needed
improvements
in
specification
quality
and
throw
out
things
where
it
turns
out.
Maybe
those
weren't
such
a
good
idea
and
yeah.
A
So,
for
instance,
we
already
have
discussed
the
text
about
canonicalization
or
consistent
encoding
and
decided
to
or
found
that
out,
and
of
course
this
is
also
a
good
time
to
do
insurability
checks
between
implementations-
and
there
is
an
hour
of
c64
10
I
already
have
this
slide
number
64
in
there.
That
defines
how
to
do
this,
so
this
is
the
process
document
that
guides
us
here.
I
still
haven't
heard
about
any
patent
claim
that
would
apply
to
170
49,
which
is
mostly
because
we
are
using
technologies.
170.
E
A
Actually
I
found
the
first
RFC
that
uses
BNF
is
our
C
5.
So
if
you
ever
need
to
explain
why
we
are
using
PMF
in
the
ITF,
it's
our
XE
5,
the
number
of
implementations
are
still
growing.
Of
course,
some
of
these
implementations
are
now
complete,
hurt
projects
to
be
friendly
or
abandoned
projects.
So
we
may
want
to
clean
this
up
at
some
point,
but
that's
not
something
like
this.
Where
does?
But
there
is
a
body
of
code
out
there
that
implements
this
soul.
A
A
There
is
no
urgency
perceived
on
1749,
but
of
course
you
want
to
get
it
done
before
this
burger
business
closed,
but
by
an
impatient
ad,
and
what
happened
here
in
this
case
particularly,
was
that
I
got
sick
after
the
last
ITF
and
when
I
got
sufficiently
recovered.
There
were
so
many
other
things
pulling
on
me
that
happened,
so
we
are
kind
of
pretty
much
where
we
were
when
we
left
London.
A
So
we
we
already
discussed,
maybe
the
most
important
civil
issues
around
theorization,
invariants,
consistent
or
preferred
encoding
and
yeah.
There
is
some
discussion
on
that
still
happening
on
the
mailing
list.
I
think
we
need
to
complete
that
discussion.
There
are
some
age
github
issues
at
the
moment,
I
think
most
of
them
are
won't
fixes,
but
we
have
to
do
the
write
ups
for
those
one
fixes
and
there
are
a
few
things
in
the
ITF
101
minutes
that
are
still
on
the
to-do
list.
A
So,
for
instance,
one
issue
that
was
raised
was
that
the
reg
eggs
tank-
we
are
not
talking
about
the
reg
axis
in
insidious.
We
are
talking
about
the
the
ability
in
a
SIBO
instance
to
say
this
character
string
is
to
be
interpreted
as
reg
X.
This
tank
is
not
very
specific.
It
actually
says
this
is
either
in
PCIe
or
in
ACMA
262
in
JavaScript
syntax.
A
So
there
is
only
a
limited
usefulness
of
the
information
this
text
this
tech
provides.
So
there
was
a
proposal
to
add
a
tag
specifically
for
the
ACMA
script
across
scripts
and
checks,
and
actually
maybe
for
the
current
one
and
not
for
the
1999
other
one,
so
yeah
we
could
do
that,
but
this
is
not
really
an
interval
issue
that
needs
to
be
fixed.
A
We
can
just
go
ahead
and
define
those
tags
if
we
need
them
in
an
application,
but
it's
not
something
that
that
actually
requires
handling
it
as
an
issue
on
this
document,
at
least
that's
my
point
of
view,
we
have
a
little
bit
of
editorial
work
to
do.
We
now
have
a
defined
data
model
and
we
haven't
really
touched
all
the
places
in
the
document
that
would
benefit
from
that
and
we
actually
have
taken
in
some
new
text
recently.
That
makes
less
use
of
it
then,
and
it
could
so
I
think
pull
request.
A
Number
11
needs
some.
Some
tender,
loving
care,
there's
also
a
related
discussion
about
the
processing
behavior
on
in
village
input.
So
many
other
series,
8
informants,
have
elaborate
rules
about
handling
invalid
input
and,
of
course,
the
the
textbook
example
years
HTML,
where
a
browser
derives.
Some
some
market
benefit
from
being
able
to
handle
even
the
most
broken
HTML,
and
that's
not
something
we
want
to
prescribe
in
this
specification.
A
A
So
if
somebody
sends
you
a
tag,
one
with
a
text
string
in
it,
that's
Bellefonte,
but
we
haven't
defined
what
it
means
and
and
Dakota
is
not
required
to
hand
this
to
an
application
and
and
say
oh
I,
gotta,
take
one
with
a
text
ring
in
it,
so
that
those
are
things
that
we
are
careful
not
to
require
a
decoder
to
do,
and
maybe
we
have
to
document
this
a
little
bit
more.
Also
people
that
don't
don't
make
too
many
suggestions
on
how
short
handles
all
right.
R
E
A
It
will
be,
there
are
different
kinds
of
security,
so
in
the
C
language,
in
the
definition
of
the
C
language,
they
have
a
specific
term
called
undefined
behavior,
which
means
if
the
coder
writes
this
code,
the
compiler
is
free
to
make
a
memory
coffee
or
to
explode
or
whatever.
That,
of
course,
is
not
the
undefined
behavior
that
we
are
talking
about.
So,
of
course,
it
has
to
say
in
its
box,
but
it
doesn't
mean
it
has
to
produce
a
consistent
result
that
there
is
always
the
same
thing.
It
just
has
to
say
it
doesn't
work.
G
G
It
has
been
illegal
right
so
so,
if
you
want
to
constrain
all
implementations
to
simply
bail
out
and
not
not
continue
parsing
in
that
case,
that's
also
fine
with
me
like
making
new
things
fatally
ill-formed
is
is
totally
within
the
spirit
of
pull
request.
17
I
just
want
out
of
some
definition
of
some
classification
of
like
the
kinds
of
mistakes
encoders
could
make
and
what
decoders
are
allowed
to
do
in
response.
A
A
G
B
B
B
B
So
we
don't
have
to
have
two
weeks
each
time
and
what
I
thought
we
might
do
is
give
it
three
weeks
from
now
which
gives
us
some
time
to
get
the
Secretariat
to
make
the
first
announcement
so
sometimes
the
week
of
the
6th
of
August
to
have
the
first
conference
call,
and
at
that
point
we
should
that
should
be
about
a
week
after
the
CD
DL
spec
comes
out
and
we're
in
the
middle
of
working
group
last
call.
So
we
can
just
check
on
our
progress
and
that
we've
made
that
and
we're
all
we're
going.
B
B
B
B
B
We
understand
not,
everybody
is
gonna,
be
able
to
make
all
of
them,
but
you
know
if
we
try
to
set
them
up
on
a
regular
basis.
We
figure
most
people
will
make
most
of
them
and
the
deal
here
is
it's:
it's
not
going
to
be
critical
that
everybody
make
every
one.
It's
going
to
be
checkpoints
every
couple
of
weeks
on
where
we're
going,
what
the
issues
are
and
what
the
next
deadline
is
and
I
see,
Paul
nodding
and
Carsten
nodding,
and
so
okay,
I
I
will
set
that
up
with
the
Secretariat.
E
B
C
A
In
the
document
see
what
was
meant
to
be
batteries
included,
and
we
deliberately
decided
to
have
a
number
of
things
that
are
not
part
of
the
basic
machinery,
but
we
pushed
into
tags
because
we
knew
people
were
going
to
need
it.
But
we
didn't
want
to
make
the
basic
notion
any
more
complicated
than
needed.
So
at
one
point
in
the
development
had.
A
I
think
I
fixed
that
one
so
yeah.
There
are
various
things,
and
these
things
are
often
a
little
bit
in
a
prototype
stage.
So,
for
instance,
the
my
message
that
we
originally
defined
was
defined
to
be
a
text
string
which
is
not
very
bright
and
then
somebody
just
registered
another.
My
message
that
is
binary
but
in
general
I
think
the
the
basic
tanks
work
out
pretty
well.
We
will
have
to
fix
one
of
the
tag:
definitions
in
the
SIBO
document
with
respect
to
padding.
A
We
discussed
this
to
IETF,
Sagawa
I
think
I'm,
not
sure
that
this
text
is
actually
in
there.
My
to-do
list,
but,
more
importantly,
it's
very
easy
to
register
your
own
see
what
he'll
use,
there's
even
a
space
that
is
first-come,
first-served
and
we
have
about
a
little
bit
more
than
the
same
number
of
tags
that
are
registered.
For
instance,
Kosir
editors
register,
some
CWU
IDs
sets
binary
mine,
as
I
mentioned,
some
pearls
have
hot
their
language,
takes
strings
and
at
some
corporation
support.
A
So
maybe
it's
worth
when
we
are
through
this
thick
with
the
two
main
documents
to
examine
that
landscape
a
little
bit
and
for
Bangkok
to
have
collected
something
like
my
favorite
egg
document
that
contains
all
those
tags
that
are
not
in
1749
but
are
generally
considered
useful
I
mean
where
they
we
already
have
RFC's.
We
are,
of
course,
only
going
to
point
to
them,
but
there
are
other
specs
that
are
generally
useful,
like
the
binary
mime
tag
and
of
course,
people
can
go
to
the
IANA
registry
and
then
follow
the
link
and
so
on.
A
But
then
they
will
run
into
specifications
that
look
very
different,
have
very
different
levels
of
details
and
just
collecting
this
stuff
in
a
slightly
more
consistent
form
might
be
useful
information.
The
document
and,
of
course
the
target
date
is
October,
which
we
won't
make
because
there's
the
fifth
anniversary
of
the
RFC
Dave.
L
Park,
and
so
the
question
is
what
is
the
value
of
this,
and
much
of
the
value
is
discoverability
by
people
who
are
coming
in
to
see
Bohr
and
think.
Oh
I
need
a
new
tag,
oh
my
god,
and
they
haven't
bothered
to
read
the
Ayana
considerations
in
the
base
spec,
so
they
don't
know
about
it.
I'm
not
convinced
we
need
it.
I,
don't
think
it
would
hurt.
I,
don't
I'm,
not
aware
of
any
other
ITF
working
group.
That
has
had
this
kind
of
easy
extensibility
where
this
has
been
done.
L
A
data
point
is
in
DNS:
we've
had
DNS
extensions
for
well
over
10
years
and
there's
half
a
dozen
like
real
ones
defined,
and
because
there's
no
such
document
and
people
just
due
to
the
way
that
the
Ayane
registry
is
it's
sort
of
hard
to
find.
The
registry
people
do
reinvent
the
same
extension
over
and
over,
and
so
this
might
help
on
it.
L
I
like
like,
even
though
I'm
skeptical
of
the
value
of
this
I,
also
don't
think
it's
gonna
take
a
heck
of
a
lot
of
work
for
us,
so
it
doesn't
have
to
have
a
lot
of
value
and
I.
I
really
do
believe
that
some
people
will
be
pleased
to
read
it
as
in
wow.
This
is
sort
of
cool.
That's
a
good
thing
to
have.
People
read
when
they're
coming
in
to
Seaborg
know.
B
R
Laurance
seems
like
protocol
extensions
for
poppin.
Dns
are
things
that
are
rarely
done,
required
writing
a
lot
of
code
and
we're
see
where
we
expect
people
to
express
all
kinds
of
protocols
in
it.
So
this
is
something
it
would
be
much
more
common
thing
for
people
that
use
extensions
here,
so
I
would
be
in
favor
of
this
reason,.
A
B
P
Walton
we're
back
on
the
wiki
topic,
though
I
mean
how
how
frequently
do
we
see
updating
this
information?
Is
it
mean?
Do
we
do
we
need
a
formal
process
in
order
to
update
an
informal
set
of
information?
If
our
goal
is
really
just
you
know,
availability
of
information,
maybe
we
do
it
as
a
webpage
and
then
sprinkle
references
to
that
all
over
in
various
places
where
people
commonly
seek
sea
bore
information.
L
Paul
Hawken,
we
don't
know
if
it's
enough
and
so
and
we
do
know
people
fall
into
RFC.
You
know
come
to
RFC's
and
so
on
that
level.
I
think
that
that
would
be
good.
It
doesn't
need
to
be
up.
It
doesn't
need
to
be
published
right.
If,
if
the
working
group
does
know
like
we
don't
want
to
publish
this
document,
it's
not
a
big
loss,
but
it
is
a
loss
so
and
by
the
way,
I'm
very
much
against
wikis.