►
From YouTube: IETF-CBOR-20220518-1400
Description
CBOR meeting session at IETF
2022/05/18 1400
https://datatracker.ietf.org/meeting//proceedings/
A
D
B
Hello,
hello,
brandon.
A
Okay,
we
have
two
large
okay,
it's
changed.
We
have
three
and
a
half
fields
of
discussion
plan
for
today,
one
being
the
tag
38,
the
extension
to
problem
details
that
does
the
language
direction
and
the
half
discussion
attached
to
that
is
that
we
might
want
to
talk
a
bit
about
how
we
evolve
types
in
general
and
on
the
topic
of
file
of
working
group
documents.
There's
a
question
of
the
status
of
file,
magic
and
carson.
A
Has
a
few
updates
on
seaboard
packed
originally
would
have
arranged
these
things
according
to
the
slides
that
carson
prepared
but
brandon.
If
you
are
only
here
for
pact,
if
you
want,
we
can
re
shuffle
back
so
that
we
start
with
zebra
pact,
and
I
don't
know
if
you're
interested
in
the
other
topics.
A
A
Okay
yeah,
then,
then,
let's
shuffle
around
and
do
it
that
way
and
in
this
case
either
way
floor
jewels.
I
think
you'll
be
sharing
slides
yep.
C
Package,
so
we
we
just
finished
the
programming
last
call.
We
got
a
few
responses.
We
got
a
lot
of
offline
thingies
coming
up
here
and
there,
so
I'm
not
even
sure
that
the
list
that
is
on
on
that
slide
is
is
even
complete.
So
I
think
the
the
overall
outcome
is
this
can
be
done
as
it
is,
but
there
are
still
lots
of
ideas
of
of
maybe
improving
or
changing
it
in
one
way
or
the
other.
C
For
instance,
I
think
christian
was
you
who
asked
about
deterministic
encoding
impact,
and
I
must
admit
I
haven't
really
thought
about
that
at
all,
and
it's
not
even
clear
whether
we
are
talking
about
deterministic
forms
of
packing
or
packing
of
deterministic
encoding,
both
of
which
are
interesting.
C
The
the
first
one
deterministic
packing
would
be
interesting
because
at
the
moment
we
don't
specify
the
packer.
We
only
specify
the
unpacker,
so
the
the
having
the
pekka
produce,
something
that
is
useful
as
a
signing
input
would
be
a
whole
new
thing
to
think
about.
C
But
of
course
that's
something
that
that
any
document
that
uses
the
the
packed
scheme
as
a
normative
reference
might
might
be
doing.
Oh
marco
was
it
thank
you
and
the
the
other
thing
packing
deterministic
would
be
making
sure
that
our
unpacking
algorithm
actually
leaves
something
behind
that
can
again
be
used
as
a
deterministic
encoding,
and
I
think
we
we
haven't,
really
thought
about
that
that
issue
either.
C
C
E
So
my
take
on
this
is
that
we
should
treat
them
separately.
E
One
of
the
points
of
of
packed
seabor
is
that
it
can
be
used
in
place
so
unpacking
it
and
expecting
to
have
deterministic
encoding
doesn't
make
sense.
To
me,
I'd
say
that
if
you've
signed
something,
then
what
you've
signed
is
what
you
get
trying
to
pack
something
post
signature
makes
no
sense
to
me.
C
So
your
database
says
this.
This
particular
playbook
entry
was
produced
by
john
and
then
you
have
a
signature
that
that
verifies
that
and
to
to
be
able
to
create
designing
input
for
the
signature.
You
would
want
to
use
some
form
of
deterministic
mapping.
E
Right
right-
and
I
I
appreciate
this-
but
what
what
I'm
trying
to
say
is
that
these
are
these.
That
use
case
doesn't
seem
compatible
with
pact
to
me,
there
are
use
cases
that
are
compatible
with
pact.
This
isn't
one
of
them.
So
I
think,
if
you're,
if
you're,
trying
to
construct
things
in
that
way,
packed
seabor
is
probably
not
for
you
that
or
there's
got
to
be
a
fallback
deterministic
packing
algorithm.
One
of
the
two.
C
Yeah
so
far,
we
we
have
discussed
packing
with
respect
to
generic
pekkas,
which
are
really
hard
to
get
deterministic,
but
if
we
have
an
application,
specific
format
such
as,
for
instance,
suit,
we
might
have
very
defined
points
where
the
packing
mechanism
comes
in.
So
I
I
even
think
it
would
be
easier
for
these.
These
application
specific
packed
formats
to
be
useful
in
a
deterministic
manner
as
well.
C
It
depends
on
how
much
packing
you
actually
want.
If
this
is
just
a
high
level
thing
that
avoids
your
format
having
to
define
ways
of
avoiding
growth
redundancy,
then
it
might
be
possible,
but
maybe
yeah.
E
C
Actually,
don't
have
to
solve
it
now,
so
that
that
would
be
one
outcome
of
the
discussion
to
say
yeah.
This
can
be
thought
about,
but
please
think
about
when
you
need
it,
because
your
application
will
shape
what
what
this
is
actually
going
to
be,
and
it's
hard
to
discuss
this
in
a
lecture.
C
Yeah
this
is
at
least
now
something
that
we
understand
the
application
developer
may
want
to
do
at
some
point.
Next
question:
do
how
do
we
handle
sequences
now?
Obviously,
you
can
pack
each
element
of
the
sequence,
but
that
doesn't
really
allow
you
to
to
use
the
full
redundancy
between
the
elements.
So
what
we
really
would
like
to
be
able
to
do
is
pack
a
sequence,
and
how
do
you
do
that?
C
Well,
of
course
you
stash
this
sequence
into
an
array
pack
and
then
unwrap
so
that's
easy,
except
that
we
don't
know
how
to
do
this
unwrapped
thing
and
of
course,
if,
if
you
come
from
the
lisp
world,
then
of
course
you
you
immediately
invent
something
that
would
be
called
unwrapped
splicing,
which
is
a
way
to
unwrap
an
array
and
put
the
elements
of
the
array
in
this
into
the
sequence
where
the
unwrapped
slicing
data
item
was
before.
C
So,
if
you
look
at
this
example
here
with
element,
one,
the
the
tag
thing,
take
n
n,
n
and
element
four.
So
this
this
starts
life
as
a
three
element:
sequence,
and
what
the
unwrapped
splicing
essentially
does
is
when
you
process
it,
it
expands
into
two
more
elements
of
the
sequence.
C
So
this
is
certainly
something
that
that
can
be
done.
It's
probably
something
that
implementations
that
that
operate
on
sequences
really
need
to
expect,
because
it's
not
at
all
something
that
current
would
expect.
C
So
this
this
requires
some
processing,
and
maybe
the
the
first
thing
that
that
is
going
to
be
an
implementation
issue.
Is
that
the
overall
sequence,
once
you
start
reading
it
has
the
wrong
number
of
elements.
C
So
if
you
use
this,
for
instance,
in
in
a
place
where
you
would
distinguish
different
arrays
by
by
different
sequences
by
their
size,
you
would
look
at
the
sequence
before
unpacking
it
and
it
would
have
the
wrong
size
so,
for
instance,
in
in
the
example
I
made
here,
you
would
look
at
it
and
say:
oh,
it
has
three
zebra
data
items
and
only
when
you
actually
look
at
the
zebra
data
item,
number
two
you'll
see
that
it's
it's
unwrapped
splicing.
C
So
again
the
the
application
would
have
to
be
very
warned
about
that,
but
I
think
it
can
be
done
in
cases
where,
where
you
still
would
want
to
use
a
sequence,
because
it's
usually
the
the
more
efficient
way
of
doing
things.
But
if
you
actually
want
to
put
some
packing
in
there
packing
across
elements
of
the
sequence,
this
unwrap
splicing
thing
would
do
it.
C
So
that's
something
we
could
define
in
this
document.
That's
something
we
could
define
later.
So
it's
not
really
necessary
to
define
this
now,
but
I
think
we
should
think
about
whether
we
have
use
cases
that
would
want
to
use
redundancy
across
items
in
a
sequence
or
elements
in
a
sequence
and
whether
this
particular
solution
already
solves
that
problem.
C
C
Yeah,
I
just
wanna,
you
are,
I
think
you
your
specifications,
at
least
you.
You
are
involved
with
specifications
that
that
use
sequences,
and
I
was
trying
to
to
pick
your
brain
here.
D
I
think,
on
the
end
doc
point
I
believe
marco
is
a
better
reference
there,
because
he's
more
actively
working
on
that
implementation.
We
have
so
if
he
has
any
feedback
on
that
point.
F
C
Yeah,
so
this
sounds
a
little
bit
like
like
a
lukewarm.
Maybe
so
the
question
really
is:
should
we
try
to
define
this
thing
now
or
should
we
define
packing
first
and
then
go
ahead
and
maybe
define
this
unreplacing
thing
separately
from
that.
A
C
Good,
so
it
seems
we
have
approximately
the
same
resolution
again
if
somebody
needs
this.
We
we
know
which
slide
to
look
at
and
can
start
working
on
this
so
next
item.
C
So
assuming
that
you
want
to
do
some
some
string
processing
here
and
you
have
three
uis
that
that
use
hd,
oh
typo,
https,
packed
example:
the
s
didn't
vanish
for
a
reason
it
just
vanished
as
a
mistake,
then
you
can
extract
the
common
prefix
https
pick
the
example
put
it
into
the
prefix
table
and
we
even
have
a
common
suffix
here,
dot
html,
which
is
only
common
to
two
of
the
three
but
but
still
may
worth
extracting,
and
then
you
can
apply
the
prefix
to
the
result
of
applying
the
suffix
tofu
and
to
bar
and
then
just
the
prefix
to
n
dot
cbo,
so
that
works
well.
C
What
we
cannot
do
is
is
kind
of
the
the
inverse
situation
where
the
things
on
the
outside
change
and
the
inside
is
is
constant.
So
I
I've
called
this
midfix
here,
which
is
different
from
circumfix,
which
is
again
different.
C
So
we
we
have
the
domain
impact.example
in
all
three
entries
of
of
the
array
or
three
elements
of
the
array,
and
we
we
extract
the
pack.example
into
the
prefix
table,
and
now
we
can
do
something
with
that.
Prefix
table
use
prefix
0
to
build
the
pack.example.html
and
again
use
that
for
bardot,
sibo
and
so
on
and
yeah.
The
example
works
well
for
the.
C
Case
where
we
only
have
a
single
concatenation,
but
if
we
need
two
concatenations,
then
we're
kind
of
missing
a
construct.
So
we
don't
have
a
tag
that
would
allow
us
to
outright
concatenate
two
strings
or
other
things
that
can
be
concatenated.
The
concatenation
is
always
combined
in
the
current
document
with
accessing
the
prefix
table
or
the
suffix
table.
C
Of
course
you,
if
you
would
know
that
you
want
to
do
this,
you
can
simply
throw
the
the
strings
that
you
you
need
to
concatenate
into
the
prefix
table,
https
and
and
co-op
s
and
still
build
what
what
you
need,
but
that's
not
very
useful
for
static
dictionaries.
So
the
the
use
case
where,
where
the
mid
fix
stuff
is
really
missing,
is
a
use
case
where
you
would
use
static,
dictionaries
yeah,
so
the
the
the
status
quo
is
we.
C
We
cannot
do
anything
for
this
use
case,
except
by
by
cheating
like
in
this
case,
which
works
fine
for
purpose
built
dictionaries.
Is
this
something
that
we
we
want
to
look
at.
A
C
A
Which,
which
will
also
kind
of
in
in
attack
that
concatenates,
might
also
help,
with
the
last
case,
about
the
expanding
in
the
inside
the
sequence.
So
maybe
that's
a
topic.
We
can
extract
from
this.
C
E
I
think
that
suit
might
conceivably
make
use
of
this.
It's
really
hard
to
know
without
without
trying
it
out
a
bit
more
but
yeah,
I
think
mid
fix.
E
Certainly
it's
not
like
top
priority
for
for
pack
c
ore,
but
I
could
see
it
being
useful
and
especially
when,
combined
with
with
sequence,
unwrapping.
C
Okay,
I
will
have
another
slide
about
extensibility
and
our
strategy
for
doing
extensibility
here,
but
it
seems
to
me
this
is
again
something
that
is
best
discussed
in
the
context
of
a
specific
application.
And
if
we
have
the
extensibility
in
place
to
make
use
of
this,
then
we
can
do
it
when
we
come
to
that
application.
C
C
These
are
all
invalid,
which
is
fine
because
we
have
defined
error
handling,
but
maybe
there
are
some
cases
where
it
actually
makes
sense
to,
for
instance,
automatically
widen
a
single
data
item
to
an
array,
or
things
like
that.
So
I
think
this.
This
would
be
a
good
time
to
think
about
that.
C
In
this
case,
we
wouldn't
be
defining
a
new
tag
to
enable
this,
but
we
would
be
extending
the
behavior
of
the
existing
prefix
and
suffix
referencing
tag
so
that
that's
a
very
different
type,
the
very
different
kind
of
extension
we
would
make.
C
C
So
when
you
look
at
chris
zipp's
record
type,
then
you
immediately
have
about
10
ideas
of
how
you
could
use
splicing
together,
arrays
and
maps.
I
haven't
actually
defined
any
one
of
these,
but
it
would
be
possible
and
when
you're
thinking
about
combining
arrays
and
strings,
you
immediately
think
about
python
join.
C
So
I
think
that
there
are
a
number
of
combinations
that
are
possible
here.
That
actually
makes
sense
again.
It
would
be
nice
to
have
the
applications
that
that
really
need
that
but
yeah.
When
we
look
at
the
cri
problem
there,
we
have
a
again
a
situation
where
we
actually
would
like
to
combine
things
that
are
structurally
not
quite
the
same,
but
we
haven't
even
started
to
understand
what
we
want
to
do
there.
C
So
I
think
it's
very
much
conceivable
that
you
want
to
do
that,
but
it's
not
like
that.
We
have
a
single
natural
way
for
combining
these
things.
A
A
There
I
think,
you'll
come
back
to
this
later.
There
is
this
point
about
possibly
merging
pre
and
postfix
styles,
and
if
we,
if
we
in
some
way
generalize
the
pre
and
postfix,
maybe
that
joining
operation
will
not
so
much
be
a
property
of
referencing
into
a
prefix
or
postfix
item
and
then
that
this
addition
here
would
be
described
in
in
in
pre
and
postfix
terms.
A
C
Yeah,
so
the
special
behavior
could
be
attached
to
the
referencing
tag,
which
is
relatively
expensive
because
you,
you
need
to
have
a
number
of
those
referencing
tags
and
you
have
to
have
compact
encoding
for
them
to
actually
benefit
from
from
the
packing
or
it
could
be
attached
to
the
table
entries
which
we
have
avoided
so
far,
because
it
means
that
the
table
entries
need
some
additional
data
attached
to
them
or
what
we
have
done
is
we
have
actually
attached
some
of
the
semantics
to
the
table,
so
these
are
approximately
the
three
cases
three
places
where
we
can
do
something.
C
C
C
Of
course,
you
could
also
do
a
static
dictionary
by
by
referencing
some
data
using
ui
or
hash
and
so
on.
But
if
it's
actually
static
and
and
standardized
like
like
what
was
it
three
four
eight
five
was,
then
you
would
just
give
it
a
tag
and
then
it's
there
so
that
that's
an
obvious
extension
point
and
we
can
define
new
reference
tags,
so
we
could
define
reference
tags
for
midfix,
circumfix,
template
expansion
and
and
so
on.
C
But
of
course,
if
we
have
new
reference
tags,
these
either
need
to
use
the
existing
table.
So
they
would,
in
the
current
situation,
use
one
of
the
three
tables
sharing
prefix
and
suffix,
or
they
would
define
the
the
new
table
type,
the
new
table
types
and
then
we
would
have
to
explain
how
the
table
setup
works
so
that
that's
pretty
much.
The
extension
points
that
we
have
at
the
moment,
and
I
think
you
said
that
you
wanted
to
bring
up
the
ability
to
put
something
into
the
tables
beyond
just
putting
the
the
data
item
itself.
C
A
Yeah
yeah,
so
I
think
whether
it's
in
the
table
or
in
the
data
item
in
model-wise
doesn't
matter
so
much.
I
would
view
it
more
in
being
inside
the
table
entry,
but
it's
how
it's
set
up
doesn't
need
to
occur,
needs
to
match
one
by
one
to
to
how
it's
in
the
table.
So,
for
example,
it
there
could
be
a
table
setup
that
still
says
we
have
a
list
of
prefix
things
and
list
of
suffix
things,
but
that
populates
a
single,
a
single
sequence
of
somehow
typed
entries
in
the
table.
A
That,
for
example,
then
contains
three,
the
first
three
items
that
are
prefixes
and
thus
use
string
concatenation
if
there
are
strings
and
every
concatenation,
if
they're
areas
and
some
post-fix
entries
and
then,
if
there's
different,
different
table
setup
or
an
additional
entry
to
the
an
additional
type
of
yeah.
If
the,
if
the
table
is
set
up
differently,
then
that
could
allow
inserting
elements
to
the
into
the
table
that
have
yet
again
different
rules.
That
would
then
be
any
kind
of
the
circle
made
or
other
fixed
entries.
C
How
do
we
do
this?
Now
I
mean
we,
we,
the
the
question,
is:
how
how
do
you
do
we
define
the
extension
point
in
such
a
way
that
we
can
add
this
later
in
a
way
that
that
doesn't
sacrifice
backwards,
compatibility.
C
So,
for
instance,
we
could
say
that
that
one
of
these
tables
actually
is
not
a
data
item
but
a
functional,
and
so
you
apply
the
functional
to
the
argument
of
the
the
the
content
of
the
reference
tag
so
that
that
would
solve
most
of
the
problems
and
we
could
even
hijack
one
of
the
existing
tables
by
by
saying
the
default
situation.
Unless
you
do
this
and
this,
the
the
functional
is,
is
essentially
created
by
by
doing
the
prefix
or
suffix
operation.
C
So
that
would
be
one
way
of
doing
it.
So,
right
now
we
cannot
merge.
We
cannot
prefix
the
suffix
tags.
So
when
you
find
a
tag
in
one
of
the
the
prefix
or
suffix
tables
that
that
could
have
a
different
meaning
than
other
things,.
C
But
that's
maybe
related
to
the
next
item.
So
so
one
question
was:
do
we
really
need
to
have
a
separate
prefix
table
from
a
suffix
table
and
clearly
having
one
less
table
means
that
you
save
a
byte,
because
in
in
the
table
setup
tag
you
no
longer
have
to
reserve
an
array
for
the
prefix
ones
and
the
reserve
an
array
for
the
suffix
ones,
and
it
also
makes
it
easier
to
actually
use
the
same.
C
Piece
of
of
text
or
the
same
piece
of
array
in
both
the
prefix
and
the
suffix
position
right
now,
you
essentially
have
to
use
one
of
the
entries
to
refer
to
the
other
entry
concatenating
it
to
the
empty
string
which
does
work,
but
that's
kind
of
ugly.
Of
course.
The
the
next
question
is:
why
do
we
separate
the
sharing
table
from
the
ethics
table?
C
C
So,
given
that
that
we
have
two
positives
here,
saves
the
bytes
and
makes
it
easier.
C
The
question
is:
wasn't
there
a
downside
and
of
course
there
was
a
downside,
because
when
you
actually
write
down
a
reference
tag
and
have
exceeded
the
the
range
of
of
a
particular
length
or
size
of
reference
tag,
then
you
get
into
the
next
one
and
the
the
way
we
have
structured
these
things
for
prefix
tags.
We
have
this.
This
first
group
of
oneplus
31
or
32
references
which
are
significantly
more
efficient
and
for
suffix
tanks.
We
actually
only
have
eight
of
them.
C
So
as
soon
as
you
get
out
of
that
area,
you
you
lose
a
bite
and
that
that
can
make
a
difference
with
respect
to
to
packing.
If
efficiency.
A
C
Right,
the
the
reason
is
that
it's
just
very
expensive
to
put
that
information
into
the
table
entry.
Of
course,
we
could
invent
less
expensive
ways
of
doing
that,
but.
C
Yeah
these
also
require
some
processing,
then,
and
and
so
on,
yeah.
Of
course
it
would
be
much
nicer
to
have
a
single
range
of
referencing
tags
that
says,
do
whatever
the
table
says
and
the
table
sent
then
says:
I'm
a
prefix,
I'm
a
suffix,
I
am
a
functional
and
then
we
have.
We
don't
need
to
spend
another
space
of
reference
tags
for
the
suffixes.
A
The
current
document
already
makes
a
statement
that
prefix
is
expected
to
be
the
more
common
case,
so
maybe
prefix
could
be
the
default,
and
then
there
would
be
kind
of
one,
but
one
byte
marking
for
for
post
fixes,
for
example,
so
the
the
the
com,
the
most
common
case,
would
still
be
cheaper.
C
E
Is
that
it
probably
makes
more
sense
to
save
a
bit
of
space
for
simple
cases
and
and
most
of
the
time,
I
think
this
will
save
a
lot
of
space
and
effort
in
simple
cases.
It
also
means
that
you
can.
I
mean,
I
think,
you've
made
the
argument
here
already
that
it's
easy
to
share
affixes.
E
But
I
I
don't
know
I
it
would
be
really
nice
if
we
could
get
a
quantification
for
actual
applications
on
the
impact
this
has.
But
I
don't
think
we
have
enough
actual
samples
that
for
for
real
applications
that
we
can
make
a
properly
informed
decision,
so
this
is
going
to
end
up
being
finger
in
the
air,
regardless
of
what
we
do.
Unless
we
do
that.
C
I
mean
this
would
be
a
little
wart
in
in
the
grand
scheme
of
things,
but
it
would
it
easy
to
share
the
same
string
as
a
prefix
and
suffix,
and
it
would
still
keep
the
idea
of
open
that
if
you
put
a
tag
in
there,
then
that
tag
tells
you
what
the
the
actual
functional
is.
That
is
applied
to
the
argument
of
the
tag
of
the
referencing
tag.
E
I
mean
that
makes
a
lot
of
sense
to
me.
I
think,
honestly,
you
know
it
costs
a
bite
or
so
you
save
a
bite
and
sometimes
it
costs
a
bite
for
the
size.
A
C
C
Okay,
I
think
I'm
getting
a
direction
here,
so
I
think
this
this
has
to
be
written
up,
but
I
think
I
will
be
able
to
generate
a
good
request
from
that
and,
of
course,
I
also
want
to
to
change
my
implementation
to
implement
that
as
well.
C
Okay,
as
always,
tags
are
the
one
thing
that
make
things
possible
in
sibo
that
are
not
possible
in
other
environments
great,
so
we
have
14
minutes
left.
Let
me
quickly
run
through
the
the
last
three
slides
here.
There
were
some
editorial
comments,
so
I
think
we
need
terms
like
current
table.
C
C
C
Okay,
next
item
interoperability,
so
I
think
we
we
have
two
natural
profiles
right
now.
Some
people
will
want
to
implement
this
for
sharing
only.
C
Because
thank
you
brendan,
so
there
are
cases
where
the
sharing
really
brings
the
the
large
benefits
and
sharing
has
the
advantage
that
implementation
can
do
this
with
simple
pointer
chasing.
So
you
you
don't
have
to
do
any
string,
processing
and
so
on.
C
So
the
sharing
only
profile
of
pac
will
be
perfect
for
very
constrained
devices,
and
the
the
sharing
plus
fx
situation
is
probably
the
other
profile
where
that
that's
everything,
but
you
could
possibly
split
this
into
a
strings
only
case
and
a
full
implementation
that
does
arrays
and
and
maps
too,
because
I
think
quite
a
bit.
Fewer
cases
will
really
benefit
from
doing
arrays
and
maps,
but
they
are
really
useful
for
some
cases,
so
I
would
like
to
keep
them
in
yeah.
So
do
we
suggest
these
as
profiles?
C
That
was
the
main
question
I
had
on
this
slide
and
the
other
question
was:
do
we
maybe
do
specific
tags,
just
as
we
have
take
51
by
the
way?
I
give
that
thing
a
new
number
in
in
the
next
version,
because
51
is
actually
the
number
three.
So
that's
not
something
we
want
to
spend
here
anyway.
So
should
we
have
separate
tags
for
these
profiles
now
that
we
are
about
to
merge
prefix
and
suffix?
Maybe
the
expense
of
having
a
single
tag
for
both
is
no
longer
as
big
yeah.
C
Okay,
so
we
don't
have
to
run
this
discussion
again.
Anything
else
on
page.
B
So
just
the
quick
thing
of
chris
zip's
thing
of
the
record
tags
for
records.
That's
if,
if
that's
done,
that's
going
to
a
separate
document,
so
there's
nothing
in
here
for
that
right.
B
E
C
Actually
doesn't
use
any
specification
required
tags
anymore,
okay
and,
of
course,
we
could
think
about
how
the
the
new
idea
of
putting
functionals
into
the
tables
relate
to
to
what
he's
doing
so,
there's
a
little
bit
of
new
thinking
that
may
be
coming
out
of
of
this
meeting.
C
So,
let's
use
the
the
last
eight
minutes
to
quickly
talk
about
tank
38,
which
is
being
defined
in
the
core
problem
details
document.
So
we
are
not
concerned
with
the
whole
problem
statement
thing,
but
we
are
concerned
with
the
the
change
to
the
tag
38..
Just
as
a
reminder
this
slide.
C
So
if
you
want
to
say
something
like
shalom,
then
you
have
to
change
the
writing
direction
and
there
is
a
common
view
in
the
internationalization
community
that
this
is
best
done
by
a
separate
argument
separate
from
the
language
tag
itself,
because
a
language
tag
doesn't
really
say
you
tell
you
which
writing
system
you
are
actually
using,
and
you
can
read
the
the
references
down
here
and
we
have
a
proposal
in
in
the
call
draft
that
defines
this.
C
The
the
one
thing
that
came
up
in
the
last
minute
was
whether
we
maybe
need
a
third
value
for
the
direction
which
is
called
auto
in
the
w3c
document.
So
left
to
right
and
right
to
left
are
the
ones
that
are
really
going
to
be
used
in
the
end.
C
But
there
is
also
an
auto
case,
and
of
course
you
could
say
you
get
this
if
you
don't
put
in
this
optional
element,
but
if,
if
this
is
used
to
override
something
that
is
inherited
from
the
context,
you
may
want
to
reset
the
context
back
to
the
default.
So
that's
why
it
may
be
a
good
idea
to
have
a
third
alternative
there
and
more
generally,
that
that
opens
the
question:
do
we
want
to
have
some
extensibility
there?
C
So
if
we
nail
this
down
to
those
two
or
three
values,
what
do
we
do
next
week
when
we
find
out
that
the
direction
this
needs
more
cases
and
that
maybe
leads
into
the
the
agenda
item
that
we
may
not
have
time
for
anymore,
which
is
how
do
we
even
evolve
tags?
So
this
is
the
first
time
we
we
actually
are
changing
the
the.
C
Domain
of
a
tag,
so
the
meaning
of
text
changes
all
the
time
because
the
the
data
that
go
in
there
are
subject
to
various
registries
and
so
on.
But
here
we
are
actually
changing
the
structure
of
of
the
content
of
that
tag
and
so
far
we
have
avoided
that.
C
A
I
think
I
think
I
read
one
of
the
documents
about
this
and
I
think
they
said
that
there
are
there's
an
explicit
list
of
languages
where
it's
recommended
not
to
spa
stated
even
that
when
it's
right
to
left.
But
there
are
languages
where
it
should
be
explicitly
stated
because
it's
necessary,
but
that
that's
all
in
the
in
the
details
of
the
languages.
C
So
the
the
problem
is
that
the
unicode
characters
that
are
in
that
string
do
not
necessarily
all
have
strong
directionality
and
as
soon
as
you
have
characters
with
weak
directionality
in
there,
you
need
the
the
context
how
to
start
setting
up
the
bi-directionality
engine,
and
I
invite
you
to
go
into
the
that
iat
and
thread
where,
where
people
are
pulling
out
old
documents,
they
have
written
15
years
ago,
that
that
show
you
really
have
to
put
this
there
and
they
lament
that
they
are
not.
C
That
bcp-47
did
not
make
this
explicit,
but
this
is
kind
of
oral
tradition
at
the
moment
and
and
they
have
to
fix
that
yeah.
C
Yeah-
and
so
I
I
just
wanted
to
to
use
three
minutes
here-
to
bring
up
completely
different
approaches,
so
the
the
we
could
do
this,
as
as
it's
defined
in
the
current
draft,
we
could
wait
for
bcp-47
to
grow,
something
that
can
indicate
the
direction
in
the
language
tag.
C
Now
that
is
unfortunately
recommended
against
by
w3c
and
also
given
the
the
pace
in
which
internationalization
discussions
are
resolved.
We
are
probably
not
going
to
get
the
outcome
of
that
this
decade.
C
Okay,
so
people
don't
like
it
at
least
some
people,
don't
like
it,
and
even
if
we
like
it,
it
would
take
too
long
so
another
way
to
solve.
This
would
be
to
separate
out
separate
tags
for
direction
only
leaving
tag
38
unchanged
and
wrapping
the
tag
38
by
by
another
tag
that
does
this
base
direction.
C
Setting
that,
of
course,
would
be
a
clean
solution,
except
that
the
the
internationalization
people
tell
me
these
really
should
be
coming
together.
It
should
not
be
separated
out
so
yeah,
but
that
would
be
an
alternative
that
would
make
it
unnecessary
to
change
tag
38,
but
it
would
keep
attack
38
there.
That
is
essentially
much
less
usable
without
the
explicit
direction
information
attached
to
it
and
then,
of
course,
we
could
just
define
as
many
take
38
analogs
as
we
have
directions.
C
So
the
the
thing
we
are
doing
in
many
other
places
as
well,
that
we
encode
some
of
the
information
in
into
the
tag
number
that
that
is
certainly
an
alternative
that
that
we
should
think
about.
But
looking
at
all
this,
I'm
still
happy
with
what
we
have
in
the
document.
But
but
I
wanted
to
hear
whether
people
have
a
strong
preference.
A
A
point
I'd
like
to
make
for
you
for
the
one
we
have
in
the
document
now
is
that
it's
easy
for
implementations
that
don't
that
basically
discard
language
and
or
directionality
information,
because
they
can't
process
it,
because
these
will
just
see
it's
tag.
38.
I
take
element
number
one
in
the
array
and
use
that
and
discard
everything
else,
and
if
we
get
two
or
three
or
an
extensible
number
of
tags,
that's
just
a
list
of
tags
that
I
text
to
basically
to
ignore
and
pick
this
or
that
element
from
the
tag
for
for
those
implementations.
A
I
we
didn't
really
get
to
the
point
of
file
magic
I'll.
Just
take
this
as
a
point
as
take
this
as
being
poked
as
the
shepherd
to
see
where
this
is
stuck
any
further
feedback
on
the
on
the
tags,
this
guy
on
the
packed
and
38
and
if
evolution
evolvability
would
be
much
appreciated
on
list.
B
No
just
just
agreeing
that,
and
thanks
marco,
as
always
for
taking
notes,
appreciate
that.