►
From YouTube: OCI Weekly Discussion - 2020-09-09
Description
Weekly OCI developer's call from 9 Sept 2020. Agenda/notes here: https://hackmd.io/El8Dd2xrTlCaCG59ns5cwg#September-9-2020
B
C
C
That
yeah
yeah,
I
could
talk
about
linux
problems
for
a
long
time
anyway.
What
I
was
going
to
say
was
what
happened.
Was
I
watched
the
video
where
darren
shepherd
talked
about
his
issues
with
multi-axe,
stuff
and
yeah?
It
kind
of
made
a
lot
of
sense
to
me
and
I
kind
of
approached
it
from
the
point
of
view
of
writing
a
registry
implementation
and
what
we
could
do
to
help
now.
I
did
like
write
a
proposal
and
I
think
I've
linked
out
there.
C
Okay,
so
it's
like
what
could
I
hack
together,
like
in
an
existing
registry,
if
you
like,
whereas
a
better
solution
would
be
to
address
it,
you
know
within
the
standard
and
the
client
a
bit
better,
but
that
being
said,
we
can
quickly
go
through
it.
Is
it
worth
me
sharing
my
screen
and
open
the
document
there
or
yeah
sure,
and
is
somebody
banging
a
bowl
in
the
background?
I'm.
C
Okay,
so
can
you
see
she
might,
can
you
see
the
screen
yep.
A
A
C
So
hopefully
I
can
just
do
this
yeah
yeah
yep
cool
yeah,
so
it's
actually
a
pretty
simple
idea.
All
is,
is
that
I
have
I
mean
I
wrote
this
up
very
quickly,
so
I
came
with
a
stupid
path.
Name,
obviously,
better
idea
to
use
something
different
than
manju,
but
basically
idea
is
rather
than
the
client
being
responsible
for
building
the
oci
index
or
the
manifest
list
or
whatever
you
want
to
call
it.
We
can
place
a
burden
on
the
registry,
so
all
is
is
we
have
a
certain
path?
C
And
that
includes
the
platform
and
the
os-
and
I
also
you
know
this
afternoon-
I
was
thinking.
Can
we
get
that
automatically
from
somewhere
else?
C
But
I
don't
suppose
we
can
because
there's
not
no
shame
manifest,
but
anyway,
so
the
idea
was,
you
do
docker
push
to
something
like
amd
64
linux,
then
actually,
the
name
of
the
image
you're
interested
in
and
the
registry
itself
will
take
care
of
bill's
nosia
index
and
placing
that
test
test
multi
and
if
one
already
exists,
it'll
update
it
for
this
given
architecture
or
add
a
new
one,
depending
on.
C
B
Yeah
it
does
it's
there's
a
couple
of
interesting
things
about
that.
Come
up,
I
mean.
There's
lots
of
challenges
too,
but
what's
interesting
is
the
you've
accounted
for
like
how
do
you
deal
with
the
path
that
I
want
it
to
be
under,
but
I
want
this
tooling
to
be
underneath
it
that
says
you
know
the
mundra
tool.
Take
the
amd64
and
update
that
manifest
so
makes
sense.
C
Yeah
I
mean
it
is,
is
frankly
a
bit
hacky,
it's
just
a
proposal
yeah.
I
don't
really
have
too
much
more
to
say,
but
it's
really
quite
quick,
yeah
one
thing
we
could
do
is
try
and
make
it.
One
thing
that
is
a
bit
frustrating
is
that
we
don't
have
the
platform
and
os,
and
I
guess
actually
maybe
actually
part
of
like
the
should
that
be
somewhere
else
as
well.
This
must
have
been
discussed
other
oci
stuff,
like
the
manifest
and
stuff.
C
A
A
You
have
to
look
you
actually
you
have
to
you
have
to
you
have
to
fetch
the
so.
If
you
get
an
image
index,
manifest
list
it'll
point
to
an
object.
That's
a
generic
manifest!
Like
you
know,
we
we,
I
guess
we
could
add
an
annotation
there
or
something
but
right
now.
If
you
want
to
figure
out
what
its
architecture
is,
you
actually
have
to
go
fetch
the
config
json
object
and
then
marshal
it
out.
Yeah.
B
C
C
C
I
was
about
to
say
it's
really
nice
being
explicit,
because
if
you're
pushing
to
this
path,
you
clearly
want
to
build
a
multi-arc
image.
So
if
this
doesn't
match
a
thing
in
the
config,
it
would
give
you
an
error.
I
don't
know,
and
the
other
thing
I
was
going
to
say
was
I
guess
we
could
build.
You
know
I
could
build
a
docker
plug-in
to
make
it
a
bit
nicer.
Potentially.
B
Is
phil
here
the
classic
problem
that
we
have
in
addition
to
spec
and
other
ones
is
there
are
lots
of
they're
different
clients,
but
there's
not
too
many
different
clients
and
we
all
use
them
for
various
different
reasons.
Obviously
docker
client
container
decline,
builder,
buildex
and
so
forth.
The
the
bigger
problem
you
have
with
a
change
like
this
is
it's
most.
Registries
have
different,
have
unique
implementations
for
various
reasons,
so
you
basically
can
propose
a
spec.
You
could
even
provide
some
amount
of
code.
B
If
you
you
know,
did
go,
you
get
some
amount
of
registries
python,
maybe
another's
whatever,
and
then
it
then
it
really
comes
back
to
which
registries
support
it,
which
I'm
going
to
turn
to
vincent
over
here
and
say
where's
the
capabilities
thing
that
you
call
it
somewhere
else.
You
call
it
because
we
really
really
really
need
that.
So
the
registry
is
considered.
A
B
B
Because
I
totally
agree
that
having
this
on
the
registry
certainly
simplifies
things
because
I
think
it
was
darren
was
bringing
up
that
you
have
different
build
systems
that
are
building
different
platforms
and
architectures
at
different
types,
and
it's
really
hard
to
get
a
good
quorum
across
all
of
those
to
know.
When
am
I
done?
When
do
I
build
the
index
and
then
there's
you
know
various
race
conditions
and
other
things
that
wind
up
happening
when
you're
updating
an
index
from
outside
so
putting
it
in
the
registry
certainly
helps
consolidate
that
in
one
place.
B
The
problem
is
that
you
need
to
be
a
registry
to
implement
it
and
certain
features
registry
as
well,
because
they,
you
know,
there's
a
business
and
need
around
it
for,
for
various
reasons
and
others
like
yeah.
That's
a
good,
tooling
thing
whatever
and
then
there's
the
other
side
of
what
phil's
been
doing
is
basically
doing
better
client
tooling.
B
That
says
that
as
long
as
registry
support
index,
which
we're
seeing
more
and
more
do,
then
it
just
works,
but
then
you're
back
to
the
other
problems
of
how
do
you
know
when
all
the
images
are
built?
B
I
think
the
the
larger
question
we
have
is
this
is
coming
up
in
some
of
the
notary.
Conversations
is,
do
we
have
the
right
model
like
is
there's
something
else
that
should
be
done
like.
If
I
push
the
same
tag,
I
don't
know
I'm
just
totally.
I
hadn't
even
thought
about
this.
If
I
push
the
same
tagged
reference
to
it's
diff,
it
happens
to
be
different,
architectures
to
the
same
repo
as
an
index
automatically
created.
B
I
don't
even
know
what
that
means
honestly,
but
it's
like
we're
dancing
around
these
two
problems
of
registry
implementations
versus
client,
tooling.
You
know
if
we
were
to
crack
this
because
there's
no
easy
answer.
What
would
the
what
would
it
look
like?
Because
I
think
you
were
trying
to
say
a
dream
like?
How
do
I
do
this
without
changing
anything?
It's
like
okay,
yeah.
C
C
There
we
go
yeah,
so
that
was
about
it.
I
don't
know.
C
So
what
were
you
saying
about
a
better
representation?
Are
we
considering
replacing
those
oci
index
or
is
that
a
potential
consideration.
B
Well,
anything's
a
potential
but
yeah
there's.
I
don't
think
momentum
around
that.
There's
we
are
on
discussing
enhancing
index
so
that
we
have
the
config
object
on
it.
So
we
can
figure
out
what
is
the
index
represent?
Is
it
a
multi-arc
manifester?
Is
it
a
a
signature?
Object,
for
instance,
for
the
notably
two
work
or
is
it
a
c
nav
for
things
that
you
know
that
they're
doing?
B
I
think
the
the
challenge
that
we
always
face
is
how
do
we
enhance
without
breaking
the
ecosystem
so
like
one
of
the
we
were
talking
about
signatures
are
basically
just
another
manifest
that
you
want
to
reference
another
another
artifact
in
the
registry.
Well,
that's
exactly
what
indexes
do
so,
that's
why
we
went
down
a
path.
Maybe
I
had
a
manifest
that
knows
how
to
reference
another
manifest,
but
that's
a
pretty
substantial
change
that
would
break
existing
kind
of
lines.
B
B
I
guess
what
I,
what
do
others
think
like
there's
some
people
that
do
represent
other
registries.
You
know
we
would
would
love
to
see
the
enhancement,
but
as
we've
also
seen
that
unless
something
is
more
available
across
multiple
registries,
you
know
it's
hard
to
get
good
adoption,
because
people
are
getting
getting
images
from
different
registries
and
moving
around
for
all
kinds
of
best
practices.
B
So
we
kind
of
need
some
consistency,
or
at
least
the
ability
to
drive
consistency.
That's
why
I
really
like
vincent's
proposal
is
like
here
is
the
enhancements
that
this
registry
supports
and
when
somebody
comes
up
and
says
hey,
why
doesn't
registry
foo
support
this
feature?
Because
all
the
other
registries
do-
or
I
have
this
customer
that
needs
it
whatever
motivates
that
company
to
do
it
then
there's
a
spec.
They
could
point
at
and
a
capability
they
can
add
and
then
there's
custard
to
them
to
drive
it.
A
A
Proposal
there
and
now
I'm
trying
to
remember
while
we
that
last
call
that
I
had
with
like
jimmy
and
josh
and
j
derek
that
we
thought
about
scrapping
that
extensions
proposal
entirely
was
just
I
forget
the
details.
I'll
have
to
look
back.
I
thought
I'd
commented
on
the
issue,
but
I
do
there
is
a
space
for
those
kind
of
feature
flags.
A
Oh,
but
I
don't
want
to
get
sidetracked
too
much.
No!
No!
No!
That's
good,
though,
because
it's
the
same
kind
of
stuff
as
far
as
like,
if
we
fold
notary
in
and
I
have
been
absent
from
all
the
notary-
progress
progress
in
the
last
months,
but
that
was
the
main
use
case
for
it,
but
this
would
be
interesting
as
well
yeah
or
anything.
B
B
But
the
next
question
is
great:
we
can
push
holistic
catalog
in
something
registries
are
so
much
more
than
that,
but
we
don't
really
have
anything
that
says
what
those
other
things
are.
So
if
there
was
this
list
of
extensions
and
then
registries
could
light
up,
hey
yeah,
I
do
support
this
and
whether
it's
peer
pressure
or
customer
driven
or
just
awareness
to
know
what's
available.
B
It
would
be
great
and
that's
you
know
I
had.
I
actually
had
sorry,
I'm
gonna
take
another
change.
The
the
interesting
thing
about
multi-arc
index
is
trying
to
figure
out.
Where
is
the
need
like
how
like,
if
you're,
building,
java.net,
debian,
sorry
java.net,
you
have
some
runtime
images
or
you're
shipping,
some
software?
I
have
an
accounting
software
package
for
some
reason
my
customers
want
to
run
around
linux
or
windows.
B
B
That
seems
like
a
pretty
small
percentage
of
overall
customers
that
are
building
images,
and
you
know
we
tend
to
build
tooling
for
the
masses
and
and
there's
always
that
there's
a
detail
of
ecosystem
is
there
some?
Is
there
a
larger
pool
of
this
that
I'm
missing
of?
Why
we
want
to
you
know:
how
can
we
get
better
tooling
for
multi-arc
images,
or
do
we
have
enough
and
the
community
that's
building
them
is
doing
what
they
need.
C
I
think
harlem
is
going
to
get
more
and
more
important.
I
think
that's
the
main
one.
B
B
B
F
I
would
say
that
the
embedded
and
iot
community
would
definitely
be
interested
in
building
multi-arch.
I
mean
they've
already
started
looking
at
containers
to
run
on
raspberry
pi's
and
things.
B
F
C
Not
really,
basically,
I
mean
I
did
send
it
to
darren
he's
like
you
know
he
said
yeah.
It
actually
sounds
pretty
much
like
what
he
was
thinking,
but
I
didn't
really
hear
much
back
after
that
I
was
considering.
I
mean
it
should
be
fairly
simple.
I
was
considering
just
doing
like
you
know,
adding
a
proof
of
concept
endpoint
to
trial,
to
register
I've
been
working
with,
but
I
mean
nobody
really
uses
trout,
so
I
don't
know
how
much
feedback
I
get
on
it.
G
Yeah
this
is
this
is
not
dissimilar
from
the
idea
I
had
of
which
is
basically
the
idea
that
steve
just
pitched
to
just
have
something.
My
idea
was
a
proxy
that
sits
in
between
you
and
the
registry.
You
want
to
push
to
when
it
when
you
push
an
image
to
it,
it
checks
the
architecture
and
updates
a
manifest
list
on
the
actual
on
the
actual
registry.
But
the
hard
part
there
is
that
the
image
config
doesn't
have
enough
metadata,
like
things
like
variant,
are
missing
from
the
image
config.
G
Yeah,
so
I
love
this
idea,
and
actually
I
I
would
make
it
bi-directional,
because
there
are
some
clients
for
which
specifying
the
platform
that
you
want
from
the
manifest
list
is
difficult
or
impossible,
like
even
in
docker.
Specifying
the
platform
that
you
want
to
pull
explicitly
is
an
experimental
feature.
It
requires
the
experimental
flag
enabled
on
your
dr
damon
today.
G
Right
yeah,
so
it
in
in
official
images,
especially
we've.
Initially,
it
was
an
implementation
detail
that
we
were
pushing
each
architecture
to
its
own
organizational
was
terrible.
It
repository
up
being
a
useful
feature
for
our
users
to
be
able
to
explicitly
override
the
daemon
and
say
I
need
this
particular
architecture's
image.
G
So
my
feedback
on
this
proposal
would
be.
I
love
it.
I
would
implement
it
as
a
registry
proxy
and
I
would
change
the
platform
specifiers
to
match
the
the
platform
specifiers,
that
container
d
uses
where
it's
os
art
with
an
optional
slash
variant.
F
So
I
have
a
question
about
that.
If
we
were
to
choose
to,
you
know,
have
have
different
paths
for
different
platforms.
Would
that
be
something
in
the
spec
like
how
to
resolve
those
paths.
F
Right,
I'm
sorry
yeah
the
name
spaces
so
right
now
I
am
trying
to
figure
out
how
to
implement
the
namespace
resolution
from
image
and
tag
to
the
actual
paths
and
they're
different
for
different
registries.
F
This
would
complicate
that
if
registries
get
to
say
like
okay,
our
our
whatever
x86
path
is
here
and
our
arm
path
is
there
and
we
have
our
own
special
resolver.
You
can
pick
and
choose.
Would
yours
with
the
oc
aspect,
then
include
like
a
spec
around
pat
namespace
resolution.
B
So
having
gone
through
this
with
acrmcr.net
and
our
integration
with
docker
up,
I
can
tell
you
that
if,
because
we've
seen
people
do
this,
including
some
things,
the
kubernetes
images
that
have
they
build
different
architectures
and
different
namespaces
paths
to
go.
That
path
basically
says
we
give
up
on
index
multi-arc
index
and
it
does
get
complex
because
different
registries
do
implement
different
pathing
capabilities
like
in
acr.
We
actually
don't
limit
how
many
slashes
you
put
in
it.
The
docker
client
does
a
256,
which
ones
are
being
really
long.
It
turns
out.
B
B
The
index
problem
then
try
to
exacerbate
it
with
multiple
name
spaces,
to
try
to
figure
out
the
architectures
that
that
would
be
my
personal
opinion
like
because
we
have
done
this
like
there
are
microsoft
ships,
a
number
of
multi-arc
image
types,
whether
it
be
net
or
java,
or
even
windows,
with
the
what
it
does
its
variants,
and
it
is
a
tool
and
painting,
but
it
does
work
and
the
that's
kind
of
why
it's
going
down
the
tooling
experiences
and
where
the
investment
for
the
consumer,
the
clients
just
work
generally,
there
is
this.
B
There
is
some
problems
where
the
a
single
docker
host
can
handle
multiple
architectures.
I'm
hoping
continuity
is
going
to
make
that
better,
but
the
consumer,
tooling
of
all
this
is,
is
in
a
good
space.
The
problem
is
the
authoring,
tooling,
isn't
quite
up
to
par,
which
you
know
a
lot
of
times.
That's
just
the
burden
of
the
mass
problem.
B
So
anyway,
that
would
be
my
two
cents.
Is
that,
let's
not
before
we
give
up
on
multi-arc
index
and
go
down
a
pathing
place,
a
pathing
name,
space
thing?
What
can
we
do
on
that?
And
for
the
sake
of
time,
what
I
was
going
to
suggest,
since
this
is
the
wonderful
world
of
open
source
community
is
there
is
no
it's
not
like
you're,
making
a
pitch
to
a
particular
company
named
after
something
that
would
just
convince
them
to
invest
in
all
the
tooling
that's
kind
of
evolved.
B
That's
kind
of
the
purpose
of
the
open
container
initiative
is
it's
vendor.
Neutrality
is,
would
love
to
see
a
group
of
people
that
are
really
passionate
about
the
space
and
keep
on
championing
it
and
bring
a
proposal
forward
so
like
adrian,
if
you
can
get,
I
know,
phil
has
been
invested
in
it.
He's
got
some
tooling
around
it
and
darren.
B
C
Yeah
I
mean
to
be
honest.
I
see
this
proposal
very
much
as
a
as
a
stop
gap.
I'm
not
sure
what
the
right
solution
is.
I
think
it
comes
back
to
your
point
about
indexes
and
stuff.
B
The
group
has
been
really
good
around
acknowledging
what
the
gaps
are
and
like
we
went
through
this
with
artifacts
like
we
want
to
be
using
the
config
object
and
we
went
around
a
lot
of
different
places
and
actually
it
kind
of
works
where
there's
other
stuff
with
index.
But
we
want
to
add
the
config
object.
That
is
a
change,
but
we're
figuring
out
how
to
make
the
change
at
a
in
a
minimally
an
impactful
way
with
a
higher
value.
B
So
I
would,
as
opposed
to
like
there
was
an
easy
answer,
just
use
annotations,
which
to
me
was
a
total
hack
and
I
hated
that
idea.
But
details
are
not
important,
so
I
guess
my
point
is
I'd,
encourage
you
to
come
up
with
what
you
think
the
right
model
is
think
about
what
the
implications
are
for
the
existing
clients
that
would
be
interacting
with
registries
or
pulling
artifacts
from
a
registry,
and
I'm
happy
to
work
with
you
on.
You
know
some
of
the
stuff
that
we
went
through,
but
what?
B
C
Yeah
I
mean
one
thing
you
could
do
is
just
to
make
it
a
bit
cleaner
is,
have
you
know
past
if
you
want
to
pass
it
back
and
get
the
config?
Have
it
somewhere
else?
So
you'd
have
like
a
you
know
in
a
post,
have
the
os
and
arc
details
or
have
it
in
a.
I
know.
A
header
wouldn't
be
very
good,
but
it's
not
very
nice,
putting
it
in
the
the
url
right
yeah.
You
sort
of
need
a
new
api
here.
Yeah.
B
And
we've
been
doing
some
stuff
around
using
headers
and
you'll,
see
some
subconscious
into
the
distribution,
pr
notary
that
we're
going
to
pass
the
media
type
in
as
in
the
header,
because
the
encoding
doesn't
really
work
anymore.
So
there
are
some
good
places
to
do
that.
B
B
All
right
so
awesome,
I
think
yeah
we're
definitely
interested.
I
think
the
big
question
thanks.
Let's
I'm
happy
to
work
with
you
offline
with
you
and
darren
and,
however
you
want
to
do
it.
I
think
the
big
question
you
have
is:
is
it
a
client
tooling,
which
is
what
phil's
been
doing
or
is
there
a
registry
side
that
you
know
both
of
them
have
their
pros
and
cons?
Yeah.
B
Yeah
all
right,
I
think
then
who's
next
here.
I
Hello,
oh
yes,
I
added
this
under
agenda.
Can
you
hear
me
welcome
yeah
cool?
Yes,
because
I'm
trying
to
add
a
new
feature
in
fancy
and
in
crm
time
spec.
So
I
put
this
document
describing.
I
I
just
started
work
on
that,
so
this
is
about
psycho
and
in
there
is
a
new
feature
in
the
linux
kernel
to
instead
of
this
heading
statically,
to
accept
our
object,
a
system
call
is
to
defer
the
decision
to
to
another
process,
and
so
it's
on
the
message
to
another
process,
an
admin
that
will
make
that
is
about
to
make
the
system
call
on
the
bear
for
the
process
in
the
container
and
that's
something
that
nxt
is
doing.
I
They
are
pushing
the
code
in
the
kernel
to
support
that
and
now
nxt
has
this
support
as
well.
So
what
I
want
to
do
is
to
do
the
same
in
run
c
and
it
needs
some
change.
Some
updates
in
the
oci
spec
for
that.
I
What
I
would
like
to
do
is
is
not
tied
to
a
specific
feature.
Yet
there
is
different
use
case
for
that,
for
example
for
lxd.
What
they
do
is,
for
example,
support
months.
I
Usually
when
containers
are
not
privileged,
they
cannot
mount
anything,
but
there
are
use
cases
where
they
want
to
be
able
to
mount
some
fine
file
system
like
tmpfs
or
park
fs
or
whatever,
without
giving
more
privilege,
and
they
use
a
seg
comp
to
to
catch
the
mount
system
color
and
send
it
to
the
adjust
outside
of
the
container
that
will
perform
the
mod
on
the
continuous
behalf,
and
there
are
other
use
case
with
a
ppf
system
call
on.
I
A
A
B
Okay,
so
alvin
just
it
looks
super
interesting.
B
It
helps
so
what
we've
been
doing
with
the
weekly
calls,
because
they're
weekly
and
there's
just
completely
random
content
from
week
to
week,
meaning
that
just
it's
just
different
groups,
so
we've
been
trying
to
make
sure
that
we
get
the
agenda
up
beforehand
and
I
I
try
to
send
them
out
by
monday
at
the
latest
and
amy
reminds
me
and
says
I'll
be
having
this
video
or
not,
so
that
people
will
know
to
actually
come
to
the
meeting
this
week.
B
So
like
I
didn't,
I
don't
know
when
you
posted,
I
missed
it
until
I
just
saw
it
now.
There's
a
dev,
alias
it's
a
debit
opening
containers.
I
just
typed
devon,
auto
complete.
A
B
If
you
just
send
this
information
there,
and
maybe
let's
say
two
weeks,
you
know
come
back,
you
might
be
able
to
get
the
folks
that
really
dig
into
this
spot
to
kind
of
call
and
have
the
open
discussion.
Because,
like
I
said
I
was
looking-
and
I
don't
know
mike
and
I'm
trying
to
think
of
who
else
here
would
actually.
D
H
On,
if
you
go
look
at
github.com
container
d,
slash
nri,
there's
a
a
new
interface
service,
that's
being
promoted
as
an
alternative
to
the
hooks
process
in
run
c,
the
you
could
probably
do
an
sec
cop,
you
know
plug-in
in
nri
and
it
would
be
fairly
interesting
as
a
way
to
model
how
to
do
this.
I
think
very
similar
to
what
you
have,
but
you
wouldn't
have
to
use
autolog
right.
H
Yeah,
it's
just
container
d.
Slash
nri
is
the
is
the
brand
michael
crosby's
working
on
it
and
I'm
sure
he'd
be
very
welcome
to
have
more
people
testing
it
out
right,
adding
any
new
types
of
plug-ins,
just
just
a
thought,
just
a
thought
there,
because
we
had
talked
about
not
not
just
a
few
months
ago
for
the
cod
stuff,
where
we're
going
to
add
additional
device
support,
we
were
going
to
do
it
via
hooks
process.
H
That's
bernal
and
morale
p
and
a
bunch
of
the
intel
guys
and
nvidia,
and
all
the
other
kubernetes
cod
guys
and
michael
sat
back
a
little
bit
and
said:
wait
a
minute.
Let
me
do
an
interface
okay,
so
you
would
take
a
look
at
an
nri.
That's
the
node
resource
interface
and
see
what
you
think.
Okay!
Well,
you
know
I'll
put
it
in
the
chat.
B
B
A
Also,
we,
you
know
it's
it's
it's
worth
noting
that
this
recording
will
end
up
on
a
youtube
channel.
So
we
can
point
to
this
the
point
in
time
that
you've
already
given
this
overview,
and
they
could
hear
that
you
know
the
10
minute
pitch
as
well.
If
they
don't
just
read
the
notes
and
whatever
other
supporting
docs.
B
A
Yeah,
it's
it's!
It's
kind
of
neat
and
I've
wished
that
google
youtube
would
get
better
at
the
transcriptions,
but
yeah
it
does
help.
Sometimes.
B
Cool
any
other
freeform
topics
that
anybody
wanted
to
share.
B
As
a
teacher,
some
things
that
come
the
website
making
some
pretty
good
progress
on
the
notary
work,
we'll
have
some
more
distribution.
Spec
proposal
changes,
including
the
refactoring
work
around
the
various
manifests
and
go
libraries.
We've
gotten
some
interesting
conversations
about
what
should
be
moved
and
so
forth,
and
the
metadata
api
stuff
that
we've
been
toying
around
with
as
a
conversation
for
a
long
time,
is
coming
up
again.
There's.
B
Basically
I
so
I've
got
some
proposals,
I'm
trying
to
get
something
written
down
for
people
to
review,
but
that
I'm
hoping
to
make
some
progress
on
as
well,
which
I'm
going
to
remind
vincent
again.
If
you
don't
come
up
with
something
better,
I'm
going
to
propose
an
a
dictionary
list
with
a
clearing
house
of
names
of
like
registries
can
just
say:
yes,
no
to
a
list
of
strings
or
something
so.
A
A
Philosophical,
but
the
the
reason
is
that
we
were
thinking
that
there
would
be
some
top
level,
like
extension,
so
that,
if
you're,
adding
actual
api
endpoints
like
rest
api
endpoints,
everything
would
be
nested
under
like
a
slash
ext
but
effectively.
That
negates
or
flies
against
the
acls
of
any
other
repository.
Where
we,
you
know,
you
might
say,
vbats,
my
app
is
a
private
app
and
so
all
the
other
api
endpoints
are
nested
under
vbat.
A
Slash
my
app
slash
whatever
else,
and
so
it
became
of
the
mind
that,
like
effectively,
one
of
the
goals
of
every
registry
owner
is
to
manage
acls
and
if
you
have
something
like
signatures
or
whatever
else,
and
you
have
all
kinds
of
like
information,
exposure
or
even
just
flush,
flush
attacks
or
whatever,
that
would
should
be
nested
under
the
acl.
So
now
you
have
this
extension
that
describes
endpoints
that
just
go
back
into
the
base
base
api.
J
A
A
Some
like
basic
json
array
or
dictionary
that
says
those
you
know
like
like.
If
you
do
system
ctl
system
cuddle,
dash
dash
version,
you
see
like
plus
fips
plus
ima,
plus
se
linux-
that's
what
that's
what
I
was
hoping
for,
but
the
fact
that
it
would
push
those
things
back
into
the
main
name
space
or
base
main
endpoints
was
less
than
ideal.
So
I
I
wish
I
could
rethink
that,
but
maybe
there's
a.
B
Incremental
step,
because
we
will
hook
that
with
notably
two
I
forgot
about
that
part
of
it.
I
was
thinking
just
what
is
a
registry
support
from
capabilities
like
you
know,
for
instance,
artifacts
all
it
really
requires.
Do
you
have
an
index
well
actually
not
index.
Do
you
support
additional
media
types?
That's
all
artifacts
really
requires
a
registry
operator
right,
but
so
there's
no
api
change,
but
to
your
point.
A
B
It's
fair,
that's
fair
and
I
think
mike
somebody
had
given
big
feedback
on
one
of
the
apis
that
we're
proposing.
But
the
point
is
that
you
would
kind
of
take
it
to
the
next
step
of
saying
great
if
you're
going
to
have
an
extensibility
on
the
registry
and
you
need
an
api.
Where
should
you
put
it-
and
I
I've
forgotten
around
that
aspect-
that
we
will
hit
that
because
there
are
two
different
styles
there's:
a
registry
domain
is
broken
up
amongst
people
through
a
route
name
space
and
then
there's
the
other
registry
pattern.
D
A
So
that's
something
to
consider,
but
on
the
fundamental
I
do
like
saying
here's
this
concept
like
multi-arc,
like
signatures
that
like
yeah,
I
support
that
go
here
for
more
information
on
how
we
implemented
it
and
even
if
that's
like
jason,
schema
or
something,
then
I
don't
know
that
I
could
get
really
creative
really
quickly,
but
that
that
part
I
like
the
managing
how
it
handles
with
you
know:
acr.
D
A
Docker
hub
clay
and
and
not
making
a
security
nightmare
is,
is
a
the
next
hairball.
B
All
right
well
we're
going
to
walk
right
into
that
problem,
because
we
are
introducing
some
new
apis
for
getting
a
referral
list
and
notary
stuff,
so
we're
going
to
have
to
break
through
that
problem
regardless.
So,
let's,
if
we
can
figure
it
out
there,
then
maybe
we
can
figure
out
how
we
can
define
that
for
new
extensions
to
add
apis.
B
E
I
was
just
gonna
point
out
that
we
did
get
the
reorganization
merged,
so
there's
still
a
bit
of
work
to
do
before
we
kind
of
release,
candidates
and
peter-
and
I
are
working
on
that
so-
but
a
lot
of
great
progress
and
thanks
to
the
maintainers
for
helping
push
that
through.
B
B
B
A
B
Some
frog
tour
or
something
I
forgot,
what
it
was
called.
It
was
some
weird
name
anyway.
Just
I
think
the
probably
the
best
thing
I
would
say
is
just
declare
when
you're
done
so
people
know
to
update
all
their
links,
so
we
have
to
update
them.
E
A
A
Yeah,
but
I
just
no
no,
but
the
topic
hasn't
particularly
been
surfaced
as
a
change
to
maine.
I
know
that
can
folk
we've
already
started
renaming
repointing
some
of
our
branches
to
maine,
but
that
that
was
one
repo
low-hanging
fruit
repos
that
didn't
have
open
pr's.
So.
B
No,
that's
fair.
I
I
I
actually
don't
know
if
github
is
trying
to
do
something
where,
if
you
open
something
and
it's
mastered,
you
change
the
name.
Just
redirect
so
that'd
be
an
obvious
cool
thing
to
add
all
right,
I'll
leave
that
one
alone.
So
where
are
we
since
we
do
have
a
couple
of
minutes?
Where
are
we
in
declaring
I
wanna?
A
Yeah,
fair,
okay,
josh
wants
me
to
say
yesterday,
at
least
by
the
end
of
the
month
now
the
they're
they're
I
I
have
not
compiled
that
as
a
timeline
in
my
head.
Just
now,
a
lot
was
going
into
the
mini
updates
that
were
broken
down,
that
josh
just
alluded
to
or
were
merged.
A
I
think
there's
a
few
cleanup.
There
was
some
hope
around
the
extension
proposal,
but
that
was
post
100..
There
is.
A
A
That
that
that
that
that
is
something
that
I've
had
sitting
in
my
inbox
for
a
little
while
and
unrelated
that
also
runtime
spec
1.0.3,
but
the
the
distribution
spec
1.0
is
obviously
got
eyes
on
it,
and
I
don't
have
a
estimation
on
that.
A
E
A
I
should
just
close
that
sorry,
given
the
fact
that,
unlike
some
of
the
other
ones,
we
were
kind
of
working
from
something
that's
fully
in
in
industry
adoption.
We
were
just
making
it.
We
had
good
iterative
conversation,
but
we
also
were
basically
working
with
cleaning
up
what
was
already
fully
adopted.
A
I
don't,
I
don't
foresee
that
we're
going
to
go
through
the
same
kind
of
rc
cycles
as
we
did
for
the
image
spec
and
runtime
spec.
I
think
that's
been
long
agreed
and
that's
part
of
why
the
incremental
rewrite
that
josh
and
peter
just
went
through
was
so
meticulous.
Was
that
making
sure
we
didn't
lose
anything
there,
but
it's
basically
what
already
exists
so
probably
we
haven't.
I
could
imagine,
having
like
an
rc1
rc2
and
then
go,
but
rc1
in
the
next
couple
of
weeks
would
be
super.
B
H
We
did
two
answers
right,
one
would
be
to
say
this
is
from
some
other
place.
Some
of
the
some
of
the
text
was
built.
So
take
it
back
three
answers.
Some
some
were
code
built
some
some
were
you
know
this
is
cited,
we're
not
you
know
no
edits
here.
Please
and,
and
others
were
yeah.
We
probably
couldn't
delete
that
right.
So
there
were
definitely
three
answers.
H
H
B
Yeah,
in
this
case,
what
we're
just
saying
is
do
as
part
of
the
cleanup
do
we
lift
them
up
into
artifacts
and
start
using
that
as
a
focal
point
for
where
these
live,
because
they
get
used
in
multiple
places,
but
instead
of
being
duped,
and
do
we
want
to
put
that
as
part
as
the
windows
spec
of
distribution,
and
do
we
want
to
add
config
to
the
one
of
those
spec
of
distribution?
B
So
that's
that's
a
conversation
and
I'm
fine
either
way.
I
think
there's
some
motivations
for
one
or
the
other,
but
we
should
figure
out.
I
I
just
think
we
believe
that
signatures
is
a
pretty
big
thing,
that
people
are
looking
we're,
seeing
more
and
more
and
more
pressure
on
a
on
a
weekly
basis.
So
well
I'll
tell
you
what
I'll
put
that
up
for
next
week's
conversation?
B
B
B
We've
been
getting
better,
we've
been
getting
better
about
it
and
I
really
like
the
idea
of
let's:
let's
keep
notes
of
the
agenda
items
at
this
minute
into
the
the
call.
So
this
way
it'll
align
with
the
videos
amy
does
the
video,
I
guess,
does
the
video
start
on
the
hour
or
the
start
when
the
first
person.
A
B
B
Yeah,
okay,
so
what
I
was
thinking
we
would
do
was
in
our
notes
for
next
to
each
agenda
item
we'll
note
down
what
time
that
topic
started
this
way,
somebody
can
scroll
forward
to
that
point
in
time
in
the
video.
The
problem
is:
is
that
if
the
video
started
10
minutes
before
the
hour,
then
we're
going
to
be
off
by
10
minutes?
If
I
just
note
the
time,
that's
why
I'm
wondering
if
there's
somebody
to
correlated
time
time
coding.
D
I
would
actually
suggest
being
able
to
have
people
look
through
the
auto
transcript
of
youtube
and
find
like
the
point
which
they're
actually
looking
for,
because
that's
a
really
easy
way
to
scrub.
Okay,.
B
All
right
all
right,
let's
see
if
we
can
I'll,
take
a
look
at
what
we
have
and
see
if
there's
something
other
than
this
thing,
or
maybe
maybe
the
transcripts
are
good
or
not
better
cool
all
right.
Well,
thank
you.
We'll
see
everybody
next
week.