►
From YouTube: DASH Behavioral Model WG Oct 27 2022
Description
Chris Sommers provided overview of structured annotations, P4 runtime spec
Virtualized ReDis SyncD addition to BMV2 GitHub project
Closed PRs 268, 264, 263, 232
A
Coming
to
the
behavioral
model,
work
group,
meeting,
October,
27th
and
I-
wanted
to
number
one
make
sure
you
guys
can
see
my
screen
yeah,
yes,
okay
and
I
wanted
to
cover
bring
up
what
we
covered
last
time
the
week
before
ocp
and
I
pretty
much
put
in
red
here,
the
the
most
important
things,
because
we
went
through
a
lot
of
items
last
time,
and
so
last
time
we
talked
about
Mariano
Nvidia
decided
to
it
was
making
changes
to
the
PSI
apis
and
those
took
priority
over
changes
to
bmv2,
and
then
he
was
going
to
go
back
and
do
p4c
p4c
changes.
A
So
we
had
that
going
on
and
we
talked
a
lot
about
this
pull
request
here
and
then,
at
the
end
of
the
day,
Chris
over
here
Chris
was
going
to
do
an
experiment
here
and
leave
a
comment
on
a
PR
and
then
at
the
very
end,
Vincent
was
gonna,
see
if
we
needed
in
to
open
an
issue
for
prefixes
in
the
Apple,
so
I
guess.
If
let
me
check
and
see
if
Marion's
on
the
call
he
is
not
okay,
so
maybe
we
could
start
with
Chris.
Then
here.
B
Yes
and
in
fact,
I
I
made
a
kind
of
a
test
branch
and
shared
shared
with
Marion
and
for
those
who
are
interested
in
P4
annotations,
I,
just
pasted
a
link.
So
you
can
look
at
this
scratch
Branch,
where
I
just
shows
some
examples,
structured
annotations
and
it's
just
an
alternative
way
so
that
the
metadata
is
right
in
the
P4
source
code
explicitly
Christina.
Do
you
want
to
click
on
that
link?
Just
so,
people
don't
have
to.
B
So
if
you
just
look
at
that
highlighted
line
at
the
top
I,
don't
know
if
you
can,
if
you
can
magnify
a
little
bit
sure
I
can
thank
you
so
just
to
re
remind
people
what
the
discussion
was
about.
I
raised
a
concern
in
a
PR
where
we
we're
extracting
P4
table
names
and
side
table
names
using
some
kind
of
heuristic
in
the
code
generator
saying:
okay,
if
The
annotation
or
the
P4
table
name
is
such
and
such
we
will
construct
the
table
name
and
the
API
names
from
that.
B
Annotate
from
that
and
I
was
concerned
that
there's
like
hidden
magic
in
a
code,
generator
That,
You,
Don't
See
in
the
P4
source
code
and
and
I
generally,
don't
like
that
kind
of
buried
magic.
So
I
suggested
just
bringing
it
right
out
as
an
annotation
in
the
code
and
I
explained
that
there's
there's
different
ways
of
annotating
and
then
you
can
get
that
from
the
people
or
info.
B
It's
not
obvious
from
reading
the
code
and
so
I
said,
look,
let's
just
make
an
annotation
and
call
it
right
out,
and
then
we
could
replace
line
38
with
39
as
a
for
instance,
and
so
I
just
wanted
to
show
that
and
then
I
extracted
the
P4
info
from
that
and
showed
it
to
Marion.
So
he's
digesting
this,
so
that
was
that
was
the
concept
and
someone
else
has
an
alternative
idea.
B
I'm
all
you
know
we're
all
ears,
but
not
everyone's
familiar
with
a
structured,
annotation
format
inside
square
brackets,
where
you
you
can
actually
put
in
different
kinds
of
expressions
in
there
and
the
P4
compiler
will
actually
parse
it
for
you
and
create
small
data
structures
like
maps
in
the
P4
info.
So
it
actually
does
a
lot
of
work
for
you,
yeah.
That's
a
newer
annotation
scheme
that
may
be
introduced
a
couple
years
ago
by
me
actually
for
this
kind
of
purpose
and
it's
in
the
P4
language
and
the
P4
compiler.
C
Yeah
I
think
for
this
particular
PR
that
we
talked
about
the
need
was
to
have
annotation
on
the
parameters
to
let's
say
action
functions,
oh
right,.
B
There's
there's
probably
yeah,
you
know
what
you're
right,
there's
another
yeah,
there's
multiple
annotations
in
this
file,
I'm
trying
to
remember
where
it
was.
Thank
you
for
reminding
me.
It's
been
a
few
weeks
hasn't
it
say.
Let
me
find
it
I
think.
If
I
search
for
the
word
Psy
I'll
find
it.
B
You
can
put
anything
you
want
and
we
just
said:
oh,
let's
just
make
it
a
Boolean,
so
you
can
see
it's
mandatory
and
there's
no
magic,
there's
no
heuristic
in
the
code,
generator
which
hides
all
that
you
know
decision
making
it's
right
there
in
the
code
and
it's
it's
really
obvious
what
it
says.
So
we
can
extend
this
with
any
kind
of
key
values
and
the
P4
compiler
parches
it
for
you.
B
Yes,
it's
the
p4c
attracted
I
I
compiled
this.
You
can
actually
go
to
this
branch
and
build
it
just
right.
Just
you
make
P4
and
you'll
immediately
see
the
output
not
three
seconds
and
encourage
people
exploring
this
technique,
I
think
it's
a
little
better
than
some
of
the
other
methods
that
have
you
know
arbitrary
parsing
schemes
and
and
formats,
and
you
can
put
different
expressions
in
those
square
brackets.
If
you
look
at
the
P4
standard,
it
spells
out
several
examples.
B
Yeah
sure
I'll
I
don't
want
to
make
this
whole
I,
don't
want
to
take
over
the
medium
with
this
one
thing,
but
I'll
I'll
try
to
find
that.
Oh
later
is
fine
just
to
yeah
yeah,
so
you
know
we
can
adopt
any
kind
of
convention
we
want
again.
This
is
where
you'd
establish
a
convention
by
saying
I
hereby
declare
this
Psy,
you
know
annotation
in
use
right.
That's
all
you
have
to
do
is
just
start
using
it.
E
Yeah,
that
can
be
any
keyword
or
the
compiler
doesn't
care
right.
So.
B
A
B
So
here's
an
app
here's,
The
annotation
mixed
expression
list
here:
I've
got
a
bunch
of
things
right.
I've
got
a
comma
separated
list
with
even
some
key
values
and
you'll
see
how
the
parser
will.
The
P4
info
will
have
this
and
here's
the
name
of
it
and
you
can
you
tell
this,
can
be
expressed
in
adjacent
to
the
output,
can
be
Json
or
probably
about.
This
is
a
protobuf
type
output
right,
but
you
just
see
how
it
just
tells
you
here's
a
key
value
list
and
it
gives
you
these
KV
pairs
all
right.
B
B
It's
also
in
the
P4
standard
in
a
little
different
format.
More,
like
you
know,
Samantha
compiler
notation.
You
know
showing
how
syntax
is
expressed
and
I'll
just
I'll
paste
this
here
in
the
chat.
So
people
can
take
a
look
but,
to
my
knowledge,
no
one's
using
the
structure
format.
Yet,
even
though
it's
been
in
the
compiler
a
couple
years,.
B
F
F
Is
great
yeah,
this
is
great
Chris
I.
Think
I
also
didn't
like
the
the
way
we
kept.
Some
of
this
information
in
the
code
generation
or
basically
the
script
that
generated
the
the
PSI
apis
and
so
forth
and
I
think
this
is
annotation
is
the
right
way
to
go
to
ensure
that
we
don't
have
that
magic
or
that
extra
knowledge
that
essentially
can
deviate
us
from
you
know.
If
you
change
something
you
know,
other
things
basically
get
changed
as
well
right.
So
right.
B
Okay,
so
that
closes
that
action
item,
at
least
in
terms
of
sharing
the
idea
and
as
far
as
implementing
you
know,
but
at
some
I
guess
Mario's
plate
for
now.
But
anyone
who
wants
to
take
this
bar
scene
on
and
use
it
they're
welcome
and
I.
Think
if
we
start
adopting
this
approach,
then
we'll
probably
write
a
readme
document.
That
says
here
are
the
conventions
we've
adopted
for
this
Dash
behavior
model
P4
code
right.
We
want
to
spell
it
out,
because
otherwise
we
have
another
heuristic
in
the
code.
A
Yep
yep
now
it
looks
like
we
don't
have
Vincent
on
the
call,
but
I
don't
know
if
bud
can
speak
to
Vincent's
item,
but
we
can
wait
till
next
week,
if
not
I.
G
Don't
remember,
I,
remember
what
he
was
talking
about.
I,
don't
have
the
no
big
deal
the
the
issue
in
front
of
me.
Let
me
take
a
couple
minutes
in
the
background
and
I'll.
A
A
A
Let
me
see
if
I
can
distill
it
down
to
the
very
like
the
shortest
version,
but
we
we
were
having
a
conversation
around
SCI
Challenger,
and
this
is
just
for
the
people
who
weren't
on
the
call
yesterday
and
basically,
at
the
end
of
the
day,
we're
missing
some
linkage
between
having
a
virtualized
I.
Guess,
the
sync
D
and
the
redis
in
Sonic
to
be
able
to
run
testing
without
actual
hardware
and
doing
a
virtualized
fashion
am
I,
as
am
I
summarizing
this
correctly
Chris.
A
Yes,
yes,
so,
and
and
just
at
the
end
of
the
day,
I
went
ahead
and
created
a
work
item
inside
of
I
didn't
know
exactly
where.
To
put
it
so
I
put
it
in
a
behavioral
Model
A
GitHub
project
here
and
I
threw
it
in
the
bottom
of
the
backlog,
where
we
I
said
virtualize,
sync,
D
and
Cyrus
for
Dash
bmv2
and
the
the
reason
is
to
exercise
Dash.
B
Thanks
Christina
I'll,
just
sort
of
put
in
a
slightly
different
context.
The
media
issue
was:
we
want
to
contribute
this
side
Challenger
based
framework
for
dash.
We
enhanced
it
for
Dash
and
Gohan's
reservation
is
well.
We
don't
have
any
like
even
POC
to
show
that
the
cyratis
connection
goes
all
the
way
down
to
the
bmv2
model
and
what's
missing
is
a
sync
D.
You
know
demon
that
compiles
the
BMV.
B
Do
lib
PSI,
along
with
the
sync
D
source
code
in
Sonic,
to
prove
this
whole
connection,
and
so
it
you
know
it's
a
valid
concern,
and
so
even
stepping
back
several
months
ago,
I
I
think
I
put
out
to
the
community
hey.
Do
we
need
a
sync
D?
Are
we
going
to
have
a
whole
virtualized
Dash
Sonic
stack
bmv2,
and
it
was
a
rhetorical
question
that
never
really
got
answered
and
I
think
I.
Think
this
topic
now
is
putting
it
back
on
the
table.
It
looks
like
without
at
least
sync
D
with
bmv2.
B
You
know
we
don't
have
a
simulation
of
the
redis
connection
going
to
a
dash
pipeline.
So
you
know
the
question
is:
do
we
need
that?
And
you
know
who
would
do
the
work
I'm
actually
in
talks
with
peel
Vision
in
the
background
trying
to
understand
the
scope
of
the
problem,
we're
all
wrestling
with
that
right
now,.
A
Yeah
so
I
kind
of
just
wanted
to
update
everyone
on
this
call,
specifically
Andy
I
thought
he
might
find
this
interesting
and
that's
why
I
brought
it
up
but
yeah.
So
I
added
that
here
any
comments
on
this
or
anyone
want
to
talk
about
this
for
a
minute.
A
Yeah
yeah,
okay,
cool
thanks.
So
so
those
are
all
the
action
items
I
had
on
the
table
for
today.
I
don't
have
a
I,
don't
have
Mario
in
here.
So
oh
sh
is
shy
here.
Shy.
Do
you
have
a
an
update
regarding
Marion's
action
item
from
last
time
around
making
doing
the
p4c
changes
and
making
the
sci
API
Sonic
changes.
A
Think
so
open
Forum,
then
I
guess
any
anyone
else
have
anything
or
I
can
give
time
back
today.
H
F
You
know
this
intersects
with
both
and
behavior
model.
We
we
talked
about
in
the
let's
recall
about
this
one,
which
is
to
say
that
today
you
know
the
proposal
that
AMD
pensando
had
presented.
F
It
actually
defines
the
the
PSI
API
for
manually
right.
However,
we
did
discuss
in
in
some
of
the
some
of
the
meetings
in
that.
How
could
we,
you
know,
have
some
ways
to
to
either
Auto
generate
much
like
how
we
do
it
for
other
cases,
within
the
you
know,
create
an
P4
model,
and
then
you
know
Auto
generate
the
apis.
However,
there
are
challenges,
as
we
all
know,.
F
Or
not,
we
can
actually
represent
it
in
P4
and
then
be
able
to
generate
the
the
apis.
But
this
is
this
work
that
at
least
we
should
investigate
right
and
then
I
think
we
we
agreed
to
that
in
that
you
know
one
of
the
meetings
that
we
should.
We
should
investigate
where
you
know
what
is
the
right
way
to
really
Define
whether
in
you
know,
have
a
P4
model,
or
perhaps
maybe
you
know
some.
E
F
Or
whatever
the
way
it
is
done
under
this
eye
to
at
least
Define
some
logical
pipeline-
and
that
essentially
is
becomes
the
source
of
Truth
through
which
the
apas
are
generated
or
apis
are
basically
can
be
referred
to
right,
so
I
think
I.
You
know
we
should
basically
have
some
some
work
item
to
investigate
how
we
we
Define
the
source
of
Truth
for
the
HSI
apis.
B
I
remember
this
discussion,
but
try
to
clarify,
for
me:
are
these
apis
that
actually
describe
or
the
pipeline
Behavior
or
are
they
more
administrative
apis
like
setting
oneip
address
to
do
such
and
such
feature?
You
know
it's
not
really
a
pipeline
operation
as
much
as
maybe
an
endpoint
or
device.
You
know
configuration.
B
F
E
F
F
Today,
what
happens
is
that
any
API?
We
have
a
lot
of
other
apis
inside
today,
whether
it's
underlay
or
overlay
side,
which
are
you
know,
which
have
some
source
of
Truth,
and
they
also
are
like
what
you're
describing
some
administrative
apis
in
which
we,
you
know,
set
certain
values.
But,
however,
each
and
every
of
those
things
you
can
refer
back
to
some
logical
Pipelines,
which
says
that
okay,
you
have
defined
certain
objects
or
certain
you
know,
instance,
in
somewhere,
and
then
you
are
doing
this
crud
operation
right
and
discard
one
of
the
current
operations.
F
It
can
be
referred
back
to
that
one.
So,
although
it's
not
so
called,
you
know
elaborate
pipeline,
so
to
speak,
but
at
least
you
know,
some
representation
is
there
which,
which
becomes
that
people
can
refer
to
I.
A
See
so
I
hadn't
really
created
a
GitHub
project
for
yet
I
have
one
for
all
up.
Dash
work
and
I
have
one
for
Behavioral
model
and
I
I
didn't
have
one
for
because
I
wasn't
sure
if
we
were
there
quite
yet
because
we
need
to
do
the
definitions
of
the
protocol,
we're
going
to
use
to
sync
and
like
what
we're
going
to
sync
and
how
often
and
blah
blah
blah.
So
are
you
saying
honey
if
we
should
have
a
project
for
h
a
similar
to
this
and
start
defining
work,
items
or.
F
A
It
does
it
does
they're
all
kind
of
meshing
together.
Okay,
all
right
I
will
think
on
that
for
sure.
F
H
A
H
H
I
think
about
that
yeah.
So
on
the
dash
repo
I
opened
one
issue
related
to
the
IPv6.
So
if
anybody
can
take
a
look
on
this,
so
actually
the
issue
is
related
to
the
under
labor
U6.
So
I
found
information
that
IPv6
for
underlay
should
be
supported
but
looks
like
the
pmv2
pipeline
is
not
expecting
to
have
the
underlay
V6.
So
something
needs
to
be
fixed
or
BMV
to
pipeline
for
the
documentations,
but
because,
as
I
understand
so
underlay
IPv6
should
be
supported
right
or
they
don't
expect.
A
So
we
need
a
volunteer
to
to
craft
this.
A
Laughs
well,
sir,
who
else
should
be
oh,
this
should
have
been
reviewer,
not
assigning
I'm.
Sorry,
Marion
I'll,
try
and
fix
it.
B
B
B
If
you
want
to
look
at
PR's
in
the
closed
I
think
it
would
be
worth
people
just
making
a
comment.
You
can
just
click
on
that
close
right
now,
yeah,
it's
actually
a
link
there
I
found.
Finally,
the
let's
see
if
I
can
find
it,
the
264
some
with
the
awesome
handle
of
pterosaur
from
I,
think
they
from
micro,
C
from
Microsoft
I,
believe
I,
don't
know
the
person
but
a
z
gun.
B
We
still
don't
test
them
yet
and
I'm
not
sure
if
they
actually
operate
I
think
there
still
might
be
some.
You
know
issues
but
we're
getting
close
to
be
able
to
actually
test
ACL
rules
in
the
pipeline
and
that's
important
and
then
263
hunt
updated,
and
we
have
a
working
B6
test
case
now,
where
we
set
up
Enis
and
we
use
a
V6
underlay
if
I'm
not
mistaken,
IPv6.
F
B
Overlay,
oh
yeah,
overlay,
yeah,
of
course,
getting
my
words
mixed
up.
V6
overlay.
You
know
it's
kind
of
a
collaboration
with
several
of
us
pitched
in
we
figured
out.
The
problem
is
really.
The
problem
was
just
simply
kind
of
a
tear
down
tear
down
instruction
in
the
PTF
test
case.
It
was
easy
to
fix
once
someone
pointed
it
out.
A
A
C
So
I
had
pointed
out
that
this
actually
more
issues
in
ankle
that
raised
in
the
issue
a
month
back
or
something
so
to
go
to
the
bottom
yeah,
okay,
32.
C
C
Actually
I
did
take
a
look
at
the
changes
for
this
one
yeah
I
can
talk
about
it
a
bit,
but
yeah
changes
are
not
very
simple,
so
I
wanted
to
talk
about
it,
okay,
so
basically
what
we
need
to
so.
The
problem
here
is
that
in
the
pipeline,
the
medical
table
is
split
into
the
multiple
stages
right,
so
each
is
a
different
table
in
the
P4.
C
It
all
becomes
separate
instances
like
each
day.
Each
each
stage
is
a
separate
instance
of
the
actual
table,
whereas
in
this
I
API
itself,
it's
a
single
API
that
takes
the
stage
is
just
a
parameter
and
then
we'll
have
to
populate
the
stage
correct
stage
table
in
the
in
the
pipeline.
So
the
BMV
to
implementation
right
now
is
broken
because
it
doesn't
do
that.
C
C
C
So
how
is
he
going
to
solve
this.
I
Yeah
there
should
be
a
kind
of
multiplexing
done
to
derive
from
attribute
name
derive
the
right
table
to
to
put
all
the
all
the
entries
in.
B
I
C
See
yeah
the
replication.
The
problem
that
I
ran
into
with
replication
was
that,
because
our
ACL
table
is
inside
the
the
different
pipelines
right,
so
the
action
actually
is
different.
It's
not
the
same
action
ID,
the
the
P4
compiler
when
it
compiles
the
stages.
So,
for
example,
the
inbound
and
the
outbound
end
up
with
two
different
action.
Ids.
C
G
C
C
C
So
that's
the
issue.
I
page
I
was
able
to
get
it
replicated,
but
the
problem
was
with
this
because
of
the
action
ID.
It
was
failing.
So
that's
where
I
stopped,
so
one
option
would
be
to
combine
the.
A
C
Lookup
into
the
bring
it
into
this
common
Dash
pipeline,
rather
than
keep
it
as
a
separate
inbound
around
bond
yeah,
but
that
needs
some
work
like
this,
but
if
there's
an
alternate
way
of
selectively
provisioning
it,
then
you
can
probably
can
look
into
that
I,
don't
know
if
this
annotation
that
Chris
was
talking
about.
Would
that
help
here?
D
C
So
it's
it's
a
single
table,
but
it's
actually
invoked
from
multiple
places
in
the
pipeline.
The
apply
is
called
from
multiple
places,
so
it
ends
up
as
multiple
instances
in
the
actual
pipeline.
D
D
C
E
D
C
But
yeah
each
of
these
have
the
same
set
of
parameters
and
actions,
and
everything
is
identical,
so
I
think
we
modeled
it
as
one
PSI
API
or
the
control
plane.
Api
is
the
same
and
there
are
like
multiple
so
there's
a
stage
attribute
that
goes
into
the
control
plane,
API,
which
differentiates
differentiates
which
actual
table,
for
instance
in
P4.
The
row
should
go
into.
D
E
D
C
Yeah
yeah,
that's
what
we're
talking
about,
but
with
that
the
issues
that
the
invoke
the
the
tables
are
getting
invoked
in
different
input.
The
pipeline
itself
is
different,
so
the
action
IDs
are
that
is
getting
generated
are
different
between
the
two
you.
D
D
D
The
built-in
P4
info
ID
attribute,
which
must
be
unique
across
all
different
tables
and
actions,
you're
going
to
be
climbing
uphill
in
the
P4
runtime
spec
to
try
to
make
those
identical
and
it
probably
won't
work.
They
need
to
be
unique
if
they're
different
named
P4
tables,
but
you
can
create
a
new
site.
Id
attribute
that
could
be
identical
across
multiple
tables.
If
you
want
no
trouble.
C
So
Martin
is,
is
the
other
engineer
going
to
look
at
this.
B
I
Think
he
is
planning
to
to
have
this
part
of
implementation,
I'm,
not
sure.
For
now.
We
just
discussed
similar
problems.
A
B
You
know
just
well
Martin
and
I
actually
had
quite
a
bit
of
dialogue.
You
know
weeks
ago,
when
I
created
this,
so
he's
he's
he's
got
everything
he
needs.
Maybe
he
could
give
us
his
take
on
things
so
Mariana
just
prized
people
did
a
little
tutorial
on
these
structured
annotations
and
shared
the
brand.
So
did
you
did
you
find
that
useful?
Are
you
guys
going
to
use
that.
I
Well,
yeah,
definitely,
structures
structured
annotations
is
something
that
we'll
need
to
move
to.
Eventually,
it's
much
more
straightforward:
it
will
reduce
the
auto
generation
to
pretty
much
nothing.
E
B
That's
great
I
think
what
we
want
to
do.
As
we
start,
adding
them
in
is
just
open
up
a
make.
A
readme
document
that
just
lists
the
conventions
we're
using.
So
people
have
to
learn
it
from
reading
code.
It's
just
spelled
out
so
there's
like
a
guide,
a
style
guide
or
something
for
this
project.
I
think
that
would
be
useful.
A
So
I
guess
the
last
thing
what
we
started
the
meeting
with
was
we
were
covering
what
we
talked
about
prior
to
going
to
ocp,
and
so
we
talked
about
this
Marian
about
making
the
PSI
API
changes
and
going
back
to
p4c.
If
you
wanted
to
tell
us
what's
going
on
with
that,
and
then
we
could
probably
end
the
call.
I
I'm
still
working
on
Psy,
there
are
issues
with
the
apis
that
we
found
because
we're
working
with
Microsoft
on
the
York
agent
integration,
so
it
would
work
at
least
virtual
switch,
but
we
found
still
some
issues
in
the
site.
Guys
there
are
minor,
IP
range
list
of
the
wrong
type
should
be
prefixes,
so
I'm
fixing
that,
and
then
there
is
a
draft
PR
for
starting
the
integration
of
the
of
the
what
it's
called
the
connection
tracking
support
in
the
compiler.
However,
it
needs
to
be
fixed.
A
little
bit.
I
I
saw
Chris's
comments,
I
just
didn't
get
to
them,
so
the
idea
is
that
I
will
integrate
the
compiler
first,
so
we
can
compile
before
code
with
connection
tracking
next
would
be.
There
are
some.
There
is
also
bmv2
support
for
that.
I
I
I
I
A
Oh
thanks,
Marion
and
bud.
Did
you
say
you
had
just
comment
on
Vincent's
and
again
we
don't
have
to
cover
it.
If
you
don't
know.
G
G
Yeah
and
I
can
share
my
screen.
Yeah.
E
G
Okay,
so
this
is
the
PSI
experimental
Dash
ACL
that
each
generated
file.
Now
this
is
from
a
workspace.
That's
a
couple
weeks
old.
So
hopefully
this
hasn't
changed,
but
this
is
what
we're
the
question
was
regarding
so
the
attributes
for
the
rule,
you
can
specify
the
different
things
you
can
match
on
in
the
packet
like
the
source,
Port
desk,
Port
protocol,
Sip
and
dip.
G
It
was
our
understanding
that
the
IP
addresses
that
you
can
match
on
the
matching.
Rules
are
actually
IP
prefixes,
not
IP
addresses.
So
we
were
expecting
to
see
like
a
prefix
length
being
able
to
be
specified
here
as
well.
I
It's
not
only
one
IP,
it's
a
list
of
ips
by
the
way,
I
think
the
API
generated
here
is
done
with
D4
code
without
Dash
match
defined.
If
you
look
at
the
ACL
file,
there
is,
if
def
Dash
match,
which
which
switches
from
single
values
that
are
supported
in
default,
bmv2
to
lists
which
is
not
supported
in
in
default,
bmv2,
which
we
need
to
add
support
to
as
well,
but
yeah.
It
should
be
in
this
single
representation.
I
It
should
be
prefix
and
PSI
has
a
special
type
for
its
IE
IP,
prefix
T
and
in
in
the
version
that
is
actually
kind
of
the
the
official
version
of
Dash
apis
It's.
A
list
of
prefixes
instead
of
list
of
IP
addresses
okay.
So
that's
what
I'm
changing.
A
A
Oh
and
Marion
I
also
added
a
work
item
to
I,
didn't
know
where
to
put
it
exactly,
but
the
in
the
behavioral
model
project
for
the
virtualized
whole
conversation
we
had
yesterday
around
site.
Challenger
I
can
move
it
around,
but
I
did
add
it
for
tracking.
So
just
so
we
remembered
it
and
that's
all
I
have
for
today.
Anyone
else
have
anything.