►
From YouTube: 2020-12-02 meeting
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
A
A
B
A
C
A
A
A
I
did
add
some
some
of
the
some
of
the
stuff,
which
I
thought.
Probably
we
can
dis.
We
have
a
regular
discussion
on
logging,
prototype,
etw,
tracer
exporter
pr
and
the
pr
issues
would
probably
be
feel
free
to
add
more
more
in
the
agenda
list
which
you
can.
A
A
A
A
D
Okay,
so
the
pr
that
is
linked
in
the
agenda
is
it's
been
approved
by
leila
and
josh,
but
before
merging,
there
were
some
other
unresolved
comments
from
johannes
about
implementing
a
max
a
default
timeout
for
the
shutdown
functions.
D
So,
if
I
don't
think
johannes
is
here
today,
but
if
anyone
would
like
to
take
a
look
at
that
before
it's
merged,
we
also
linked.
We
also
created
an
issue
for
implementing
a
recordable
interface
that
was
related
to
that
pr.
D
So
something
similar
should
be
implemented
for
logs
like
for
traces,
but
because
it
was
a
sizeable
amount
of
work
that
you
know
due
to
the
time
constraints
of
mark
and
I's
like
internship,
we
might
not
be
able
to
complete.
We
filed
an
issue
for
that
and
another
issue
that
we
were
thinking
of
filing
was,
as
layla
suggested.
D
We
could
either
have
support
for
multiple
exporters
per
log
processor,
or
we
could
just
do
what
traces
did
like
in
the
spec
and
have
one
processor
per
exporter,
and
I
am
I
think
this
this
should
be
up
for
discussion,
so
I
will
fight.
D
I
will
like
open
up
an
issue
about
it
as
well
later,
but
I
think
that
having
one
exporter
per
processor
might
be
better,
for,
I
guess
simplicity,
because
having
multiple
exporters
per
processor
could
I
I
think,
because
the
processor
has
to
return
whether
it
was
shut
down
successfully
and
force
flush
successfully
it's
hard
to
return
that
status
when
multiple
exporters
are
attached
to
it,
and
also,
if
you
know
each
exporter
needs
each
export
needs
to
complete.
D
Before
the
processor
is
able
to
create
the
next
batch
of
logs,
then
it
could
increase
the
export
time
per
batch.
So
my
initial
thoughts
are
having
one
process
per
exporter
would
be
all
right,
but
I'm
not
sure
whether
there
are
any
other
opinions
or
considerations
that
I
haven't
thought
of
there.
So
yeah.
There
will
be
an
issue
for
that
later
or
if
anyone
has
any
thoughts
on
that
right
now,
yeah.
A
Okay,
so
I
mean
just
just
to
understand:
I
mean
once
this
the
first
pr
is
closed.
Do
are
we
expecting
any
more
prs?
I
mean
I
I
know
we
are.
There
are
two
open
issues
here.
One
is
I
mean
whether
we
need
a
recordable
interface
or
we
need
a
sharepointer
or.
B
Yes,
I'm
working
on
another
pr
right
now,
which
it'll
come
after
the
simple
processor
gets
merged,
where
it's
making
a
bunch
of
changes.
The
api
and
sdk
like
adding
more
api
overloads
and
like
adding
the
get
name
method
back
into
the
api,
and
I'm
also
making
the
change
where
we're
using
a
shared
pointer
which
we're
going
to
pass
the
processor
instead
of
using
the
unique
pointer
like
we're
doing
right
now.
A
B
And
I
doing
it
like
this,
in
my
opinion,
it's
better
because
then,
if
there's
multiple
processors,
then
they
all
use
the
same
shared
pointer.
So,
instead
of
having
to
duplicate
a
record
like
end
times,
if
there's
n
processors
attached,
then
you
can
just
use
this
same.
Shared
pointer
and
I
know
josh,
you
had
some
concerns
about
doing
it.
That
way.
E
Yeah
so
effectively
you
you're
you're
killing
the
cpu
cache.
That's
well,
not
quite,
but
it's
so
so.
My
main
problem
is
not
actually
performance
like
overall
speed.
It's
actually
about
bounding,
the
performance
of
telemetry,
where
we
want
to
make
sure
that
the
overhead
that
we
make
use
stuff
in
terms
of
memory
and
the
overhead
we
make
use
of
in
terms
of
cpu
is
bounded
for
this
entire
thing
and
so
shared
planner.
E
The
problem
with
it
is
you
don't
know
when
it's
going
to
get
gc'd
at
all,
so
you're,
putting
pressure
on
the
actual
overall,
like
memory
allocation
of
the
allocation
system
that
you
don't
know
when
it's
going
to
occur.
E
The
second
thing
is:
if
there's
any
way,
we
can
kind
of
keep
the
I
take
off
of
this
queue
of
logs
and
shove
it
out
in
an
exporter
as
close
to
as
close
in
cpu
l
cache
as
possible
the
more
efficient
it
will
be
like
if
we're
starting
to
share
these
things
across
multiple
exporters
possible
threads.
It
gets
weird
right,
so
I
don't
know
how
much
like.
Yes,
it's
just
literally,
you
know
unique
point
of
share
pointer
and
it's
simple.
E
It's
just
there's
concerns
that
I
have
in
the
overall
of
like.
Can
we
even
make
it
be
efficient
later
or
found
it?
The
bounded
bit's?
The
most
important
thing
to
me
of
I
want
to
know
when
I'm
paying
for
allocations
and
the
allocations
and
know
when
I'm
dealing
with
you
know
threading
stuff,
like
if
that's
limited
and
very
clear,
where
that's
happening,
it's
better
shared
pointer.
Unfortunately,
in
servers,
specifically,
you
don't
know
when
the
the
allocation
happens,
and
so
it
can
lead
to
inconsistent
performance
as
opposed
to
the
bounded
performance.
E
That's
that's
my
main
concern
with
that.
We
can.
We
can
talk
about
that
more
later,
if
you
want,
but
I
think
really.
B
B
A
A
We
have
events,
we
have
links
attributes
at
the
level
of
span
and
attributes
at
the
level
of
links,
but
probably
we
can
discuss
in
the
ticket
whether
it's
the
right
approach
for
the
log
logging
api
or
not,
and
we
also
have
to
discuss
whether,
as
of
now,
I
don't
see
the
logging
api
supporting
multiple
processors.
D
Okay,
so
for
some
context,
I
think
that
some
of
the
exporter
and
processor
issue
that
could
be
a
potential
blocker
for
the
issue
337..
That
was
the
that
includes
like
a
batch
processor
implementation,
as
well
as
like
multiple
exporters,
there's
like
elasticsearch
otlp
and
also
one
to
console.
D
So
I'm
wondering
if
we
could
get
the
exporter
and
processor
like
sort
of
decision
like
or
make
a
preliminary
decision
on
that
to
unblock
this.
Would
that
be
you
know
in
the
near
future?
Would
that
be
possible.
A
Yeah
we
definitely
I
mean
I
totally
agree.
You
definitely
need
to
have
that
decision
as
soon
as
possible.
Let's,
let's
try
to
in
that
pr
property.
Let's
try
to
tag
your
hands.
I
mean
if
he
also
had
some
concerns
over
there
and
and
let's,
let's
take
it
forward
from
there
but
yeah.
Definitely
I
agree.
We
need
to
decide.
I
mean
my
personal
preference
would
be
having
a
one-to-one
mapping
between
processor
and
exporter
and
having
multiple
processors.
The
same
way,
the
way
we
are
doing
in
tracer
api.
A
A
Okay,
yeah
and
then
there
was
one
task
which
was
created
by
rayleigh.
I
think
dog
feeding
I
mean
he
didn't
create
any
task,
but
he
did
talk
about
dog
feeding
and
user
research
and
me
and
tom.
We
both
agreed
on
doing
this
test
this
research
book.
A
Basically
this
this
really
talks
about
I'll,
just
open
this
one.
So
it
talks
about
try
to
instrument
using
your
c
plus
plus
I
mean
our
c
plus
plus
open
telemetry
api,
try
to
instrument
a
web
framework
http
web
framework
and
try
to
see
how,
as
an
end
user,
how
it
really
works.
A
I
mean
I'm
not
sure
I
definitely.
I
didn't
get
much
of
time
to
really
work
on
that
and
I
did
start
doing
analysis.
I
mean
there
was
definitely
a
fight
to
really
choose
what
kind
of
web
framework
you
can
use.
I
mean
it's
easy
to
decide
on
the
web
frameworks
for
node.js,
ruby,
python
and
other
language.
Definitely
it's
it's
something
difficult
to
really
when
it
comes
for
c
plus,
but
definitely
won't
want
to
hear.
If
somebody
has
any
opinion
and
tom
did
you
really
had
anything
any
update
on
that.
F
A
No,
I
don't
see
any
time
and
deadline
for
that.
I
mean
good
good
to
have
it,
as
at
least
I
mean
whenever
we
find
it's
doable
and
let's
try
to
see
it.
I
see
some
of
the
teams
have
done
it.
I
created
a
ticket,
I
created
a
a
github
ticket
and
I
just
put
some
reference
of
how
other
teams
have
done.
I
saw
ruby
and
probably
php,
I
mean
the
teams
there.
They
have
definitely
done
some
controversy.
A
F
F
A
Our
github
tickets-
here
I
mean
we
can
move
so
I
just
created
this
one
block
pudding
and
probably
I
had
reference
how
the
other
teams
have
been
doing
it.
So
probably
we
can
see
more
information
from
this.
F
A
Thanks
yeah,
that's
all
from
this,
and
probably
the
etw
tracer
exporter,
pr
update,
probably
max
or
mitchell.
I
mean
you
can
just
give
update
on
what
what's
happening.
It
is.
Is
it
ready
for
review?
I
mean,
because
I
see
it's
still
in
active
state
but
and
lots
of
check-ins
are.
C
Yeah
so
microphones,
it's
really
etw
exported.
This
is
really
full
review
and
michelle
can
comment
on
what
we
are
still
missing
is
the
logging
provider,
but
this
work
is
happening
in
parallel
done
by
karen
and
mark,
so
michelle
is
going
to
send
it
another
update,
a
separate
vr,
which
will
also
add
the
login
export.
C
So
right
now,
there's
only
trace
of
export.
I
don't
know,
please
guys,
take
a
look.
There
is
another
dependency
which
I
think
we
need
to
contribute
to
cantriba
ribo
the
listener.
Sorry
like
sidecar
the
agent
for
that
fool.
This
will
be
done
as
a
simple
c
sharp
application,
because
it's
just
easier
to
listen
for
incoming
hw
events
than
and
will
print
the
incoming
events
to
stdl
decoded,
like
in
whatever
structured
format,
most
likely
in
json.
C
Sorry,
okay,
so
I
already
gave
some
tooling
to
commercial
just
to
keep
up
the
flow
to
actually
see
that
events
are.
His
change
should
be
mergeable
in
my
opinion,
but
hours
improved.
If
you
guys
can
take
a
look,
we
definitely
need
a
second
pair
of
eyes
to
go
over
there.
F
Okay,
the
next
item
is
from
me
and
I
think
this
issue
was
reported
by
mark
and
karen
and
in
their
prototype
of
the
logging
support
and
currently
in
in
the
armed
force
flash
and
the
shutdown
api
we
put
on
forge
for
trees
for
trees
and
the
log.
We
don't
provide
time
out
support,
so
I'm
going
to
add
that
currently,
for
the
time,
timeout
value
which
is
set
to
the
micro
microseconds
are
zero,
which
the
comment
says
that
represents
no
timeout,
since
that
is
that
that's
for
infinite.
F
But
since
I
I'm
thinking
about
maybe
zero
is
not
a
good
time.
Timeout
value
for
infinite
right,
I'm
asking
suggesting
here
and
I'm
I'm
thinking
about
minus
one
for
say,
plus
plus
20.
The
there
is
there's
a
static
member
method,
coronal
duration
class,
which
is
max,
which
I
think
that
can
be
used
for
to
represent
infinite
but,
as
our
minimum
requirement
is
say,
say,
plus
plus
11.
I
think
we
can't
use
that.
F
C
This
minus
one
is
this:
a
class
that
we
can
backport
like.
Is
this
something
that
is
available,
for
example,
in
upsell?
It's
a
good
reference.
So
if
it's
already
already
available,
if
it's
a
plus
plus
11
in
abseil,
then
maybe
this
is
something
that
we
can
add
to
non-standard
classes.
C
F
I
haven't
looked
at
how
how
it
was
down
in
c
flat
size
20..
I
guess
it
is.
It
is
just
a
constant
in
a
static
maximum
method,
yeah,
and
I
think
we
just
need
one.
We
can
prepare
twice
right
here,
don't
know
std.
C
Name
space
that,
because
we
already
do
it
everywhere,
for
the
other
things,
like
stream
view
and
a
few
other
features,
this
seems
to
be
a
relative
reminder.
James.
I
can
take
a
look
if,
if
you
guys
want-
because
I
still
need
to
refresh
my
other.
F
You
mentioned
the
library
abc
all
right,
yeah
yeah,
absolutely
yes,
it's
google's
blackboard
yeah
yeah,
I
heard
about
it
is
it
say,
plus,
plus
20
compatible
or
with
it's.
D
C
F
F
C
C
Upsell
there
now
I
had
the
conversation
with
a
mail
before
that
in
some
cases,
maybe
we
can
even
borrow
if
it's
friendly
licensed.
We
can
just
borrow
bits:
okay,.
F
F
F
Good,
okay,
we'll
see
whether
it
is
back
portable
or
if
it
count,
then
I
think
we
can
just
use
minus
one
right
just
to
provides
two
options.
The
first
part
yeah
preference
is
maybe
backward.
You
did
currently
then
minus
one,
maybe
just.
C
A
constant
again
because
then,
if
we
assume
that
it's
a
constant,
we
don't
exactly
just
like
minus
one
right,
it
is
yeah,
but
we
need.
C
The
usage
of
a
constant
instead
of
it.
So
then,
if
something
changes
we
can
then
analyze
it
to
the
actual,
better
type
of
plus
plus
20.
C
F
C
Yeah
yeah-
maybe
I'm
just
thinking
here
on
the
scenario-
is
there
any
way
for
us
to
eventually
configure
this?
Because
when
we
configure
this
in
some
sort
of
system
through
api,
like
options,
configurations
for
the
tracer
or
for
blogger
provider,
then
I
think
it's
not
going
to
be
like
http,
chrome,
microseconds,
minus
one,
it's
going
to
be
like
max
timeout
or
some
constant
payment,
then
yeah.
F
A
A
D
Yeah
all
right
just
to
add
to
that
from
a
spec
perspective,
it
says
that
the
these
two
functions
should
not
block
indefinitely
so
would
having
that
infinite
or
that,
like
not
non
lack
of,
implement
like
lack
of
a
constant
timeout
like
would
that
be.
You
know
like
acceptable.
F
This
is
a
good
question
because
I
I
look
at
the
net
implementation
if
you
use
use
this
time
out,
not
infinite
yeah,
but
if
then,
which
means
we
have
to
choose
some
specific
specific
value
of
time
out,
not
infinite
or
I
haven't
looked
at
the
spec
in
detail.
This
part.
D
Currently,
I've
set
it
to
the
max
like
time,
which
is
a
constant
value.
The
return
for
largest
duration
possible
would
that
be
okay.
F
D
A
Okay,
one
point
I
just
wanted
to
bring
was
that
I
mean
do
we
see?
I
mean
the.
If
I
see
a
38
days
I
just
wanted
to
see
like
should
we
have
meetings?
I
know
30th
is
going
to
be
holiday,
so
definitely
we
won't
have
a
meeting
on
that
day,
hover
21st
and
16th.
I
mean
there
was
a
vacation
plan
of
each
of
us,
which
was
some
more
in
the
next
in
the
last
meeting
it
was
given,
but
I
didn't
find
for
everyone.
A
I
know
karen,
you
guys
will
not
be
there
after
15th
right,
the
24th.
D
A
Yeah
make
sense,
yeah
just
want
to
understand
from
the
others.
A
F
A
A
A
A
A
Okay,
probably
issues
just
created:
one
talk:
folding
user
research.
Probably
we
can
to
have
a
discussion
in
that,
that's
something
which
we
were
talking
about
in
the
agenda
and
then
there
is
implementing
implementing
a
recordable
interface
for
locks,
that's
something
which
karen
has
created.
A
We
already
had
a
discussion
about
it
now
and
I
just
created
one
self
diagnostic,
probably
josh
had
a
good
comment
coming
as
part
of
the
peer
review.
We
are
missing
as
of
now
a
consistent
way
of
exposing
the
errors
and
diagnostic
information
information
like
if
the
tracer
api
itself
fails.
How
do
we
handle
that?
How
the
end
user
should
be
communicated
about
those
errors,
so
I
just
created
a
ticket
for
that,
mentioning
what
the
specs
talks
about
that
and
how
the
dotnet
and
go
link
are
really
handling
that
scenario.
A
So
probably
just
go
through
that
and
let's
start
discussing
how
to
handle
this
scenario,
I
mean
how
we
definitely
need
some
support
for
diagnostic
and
error
handling.
So,
let's
see,
how
can
we
do
it?
Probably
let's,
let's
start
discussing
here
for
that.
A
F
A
A
And
yeah
and
make
version
available
for
runtime
yeah.
This
was
created
by
tom.
This
is
assigned
to
me
and
I
have
raised
the
pr
for
that.
So
yeah,
I
think
the
rest
are
the
old
ones
we
can
just
ignore
them.
A
C
I
have
submitted
the
scripts,
which
I
love
to
build
all
through
docker
and
I
think
linux
I
get.
Linux
allows
to
run
docker
and
docker.
C
So,
yes,
we
should
be
able
to
use
one
of
the
github
action.
The
ubuntu
runners,
for
example,
204
to
iteratively
4-h
run.
I
need
to
see
if
we
hit
any
constraints
like
execution
constraints,
time,
timing
constraints,
usually
it's
like
60
minutes
for
the
entire
run
or
something
but
I'll.
Take
a
look.
A
C
C
C
A
C
A
Yeah,
okay,
yeah
and
then
there
was
one
which
I
created
to
enable
the
http
ssl
verification,
vocal.
Our
current
implementation
for
curl
I
mean
for
http
client
based
on
curl.
A
Does
it
just
ignores
the
ssl
server
certificate,
verification
which
is
not
a
good
thing
to
have?
So
probably
I
need
to
look
into.
How
can
we
enable
the
ssl
verification
we
need
to
provide?
We
need
to
have
a
way
to
provide
the
ca
certificates
and
which
can
be
used
by
the
curl
client
to
do
the
verification.
A
So
probably
I
just
need
to
look
into
that
how
to
do
that.
As
of
now,
it
works
fine
if
we
disable.
If
we
ignore
the
ssl
verification
in
the
code
which
we
have
done
as
of
now,
we
don't
enable
the
ssl
verification,
so
everything
goes
seamlessly.
Only
thing
is
that
we
may
we
may
be
talking
with
the
wrong
server
and
we
may
not
realize
it.
So
that's,
that's
that's
the
issue
and
yeah.
I
think
restaurant
old
one
we
can
ignore.
A
F
A
Yeah
and
prs,
I
think
we
had
discussion
for
most
of
them.
I
just
created
one
pr
for
updating
the
pr
guidelines.
Probably
we
can
just
quickly
discuss
whether
it's
good
to
have
it
or
not.
The
change
is
basically
the
major
change
which
is
in
this
is
basically
we
are
I'm
trying
to.
A
We
should
be
good
tom
had
one
comment
that
will
keep
two
approvers,
but
at
least
one
should
be
from
different
company
as
a
pr
owner.
So
I
mean
I'm
good
with
that.
Just
wanted
to
get
the
what
you
guys
think.
Is
it
okay
to
have
one
approver
so
just.
E
F
Another
thing
I'm
thinking
about
is,
after
the
last
update
of
the
pr
for
non-trivia
one
I'm
talking
about.
Do
we
need
to
see
some
minimum
time
to
get
in
the
merge
like
or
one
or
two
days
to
give
it
give
enough
visibility
to
the
community.
F
A
A
Okay
for
non-trivial
changes.
We
should
explicitly
mention
that
we
have
to
wait
for
one
day
before
really
going
ahead
for
the
well.
E
F
I
just
want
to
give
give
enough
visibility
for
the
pr
yeah,
because,
because
we
emerged
too
quickly,
maybe
some
some
reviewers
may
have
been
saying.
We
haven't
got
a
chance
to
to
review
the
change
yet.
A
Okay,
so
we
already
have
that
open
for
one
working
day
for
any
kind
of
any
kind
of
I
mean
that
basically
would
be
for
non-trivial
to
wait
for
at
least
one
working
day,
so
so
you're
saying
that
in
case
in
case
we
have
some
comments
coming
from
from
the
reviewers
for
trivial.
For
for
non-trivial,
we
should
wait
for
some
more
for
at
least
one
day
before
really
merging
the
pr.
F
Yes,
yeah
or
slightly
different
than
this
row,
but
yeah.
I
think
the
this
three
is
fine
yeah.
A
F
Sounds
great
put
up
the
comments
there
yeah.
A
Okay,
we
already
discussed
about
log,
processor
and
log
exporter.
I
think
we
should
be
good
to
merge.
It.
A
Already
I
mean
we
already
have
the
two
tickets
for
decide
on
the
interface
between
exporter
and
processor
and
the
timeout
thing,
so
I
think
those
those
those
changes
can
follow
this
pr
once
this
is
merged,
so
I
mean,
if
you
all
agree,
probably
we
can
just
merge
this
pr
and
go
ahead
with
the
subsequent
changes
coming
afterwards.
A
A
I
did
create
one
pr
for
versioning
information
because
that
ticket
which
tom
has
created
so
this
is
approved
by
josh
and
by
tom
I
mean
max.
If
you
just
want
to
have
a
look
into
that.
I
mean.
C
I
have
a
look,
I'm
I'm
mostly
happy.
There
are
other
things
that
I
was
thinking
like.
Should
we
keep
most
of
scripts
under
the
tools
directory?
Okay,
a
reasoning
is
if
we
don't,
this
may
become
messy
in
the
root
folder.
If
we
can
keep
everything
under
slash
tools,
okay,
sure
that
is
clear
and
that'd
be
similar
to
how
other
projects
or
how
others
should
but
bigger
size,
simplest
plus
projects
are
highlighted.
C
C
I'll
refresh
this
don't
look
yet
I
will
finish
it
because
it's
gonna
shrink
twice
at
two
times,
because
I
need
to
rebase
and
include
the
tooling.
E
Of
those
actually
the
htw
and
the
standard
library,
there
was
a
call
for
help
for
bazel
support.
Now
that
we've
upgraded
bazel.
Thank
you
to
whoever
did
that.
I
didn't
pay
attention
we
can.
Actually.
I
can
help
so
just
bring
me
back
whenever
you
need
it.
Okay,.
C
Cool,
so
just
a
quick
one
in
the
standard,
non-standard
library,
I
will
also
submit
a
document
that
explains
for
those
who
use
visual
studio,
2019
latest.
It
already
has
decent
support
for
cma
like
built-in
support
for
semi.
Before
in
older
versions,
we
have
to
generate
visual
studio
project
from
cmac,
then
open
that
visual
studio
generated
project.
It
was
ugly
because
you
couldn't
get
the
entire
view
of
all
files.
C
You
had
to
regenerate
this
whenever
you
change
this
image.
Now,
it's
all
streamlined.
You
can
just
open
a
directory
in
visual
studio
which
has
cmac
lists.txt.
It
shows
you
whatever
layout
options
for
the
cma.
It
has
a
transparent
support
for
a
c
test
like
a
google
test
in
there,
and
you
can
actually
have
multiple
configurations.
C
So
this
is
an
addition
how
they
handle
it.
They
have
the
adjacent
panel,
which
lists
configuration,
configuration
name,
destination
directory
and
see
make
options
that,
like
slash
d
with
our
tuning
tees
on
and
off
what
I
was
thinking,
I'm
gonna
document
how
to
use
that
this
is
not
a
required
process,
but
it
is
really
nice
to
have
a
neat
process
and
convenient
process
for
those
who
use
visual
studio.
I'll,
have
a
write-up
on
this
and
I'll
explain
how
you
can
cook
the
sdk
with
non-standard
classes
and
with
standard
classes
using
that
approach
again.
C
A
Okay,
thanks
max,
I
think
definitely
documentation.
Documentation
will
help
in
that
perspective.
Here
yeah,
I
think
that's
all
this
one
is
already
being
taken
care
by
your
hands,
so
I
think
we
can
just
ignore
it.
For
now
I
mean
probably
let
your
hand
comment
on
that
and
I
think
that's
all
we
have
and.
A
Yeah,
I
think
we
are
done
with
whatever
engine
agenda
item
we
had.
If
we
have
something
else
to
discuss,
I
think
we
can
do
it.
Otherwise
we
can
we'll
get
15
minutes
back.