►
From YouTube: SES Meeting: Realms Debrief
Description
Realms attempted an advance to Stage 6 at TC-39 Plenary the prior week. In this meeting we debrief and discuss next steps toward advancing that proposal, including an important relaxation of the prohibition against host extensions, IO, or mutation side-effects provided that these can extensions can continue to be erased by the SES shim, in anticipation of a future Lockdown proposal.
A
Welcome
to
the
cess
meeting
today
we're
we're
returning
from
a
tc39
plenary
that
was
last
week,
so
we're
going
to
begin
with
a
summary
from
leo
on
what
occurred
followed
with
a
conversation
about
how
to
make
progress
on
the
realms
proposal
before
the
next
meeting
the
and
we
are
expecting
jordan
in
20
minutes
to
join
us
for
that.
Second
conversation
so
leo.
Take
it
away.
B
B
We
showed
that
the
path
that
we
went
ahead
as
for
the
module
graph
and
also
the
web
at
the
globals
that
we
should
be
adding
the
path
that
we
are
proposing
we
actually
proposed
in
in
the
meeting,
was
to
keep
realms
just
adding
the
x
my
script
globals
and
having
requirements
for
the
host.
If
the
hosts
add
globals,
it's
still
possible
to
add,
we
have
we
presented
two
requirements
that
all
the
globals
should
be
configurable
and
another
requirement
to
for
the
global
student,
have
authority,
io
access,
etc.
B
Some
of
the
discussions
from
this
tc39
meeting
raised
an
interesting
part
to
discuss
if
we
should
just
remove
the
second
requirement
and
to
just
keep
the
requirement
of
having
only
configurable,
all
the
all
the
globals
being
configurable.
So
this
way
still
allows
us
to
do
a
proper
configuration
of
the
each
realm
instance.
C
I
want
to
I
want
to
make
one
clarification.
The
requirement
is
not
just
that
the
property
be
configurable,
it's
that
it
actually
be
deletable.
Often
we
assume
that
the
first
means
the
second,
but
it's
that's
not
necessarily
the
case.
The
object
and
variance
require
that
it's
a
global
is
not
the
not
configurable
that
it
not
be
deleted,
but
if
it
is
configurable,
an
exotic
object
may
still
refuse
to
delete
it.
In
this
case,
we
actually
insist
that
the
that
the
properties
be
deletable.
B
Yeah,
I
can
add
text
like
that.
I
just
need
to
proper
when,
when
I
say
a
property
is
actually
configurable,
I'm
just
relating
to
the
object,
description,
descriptor
and
I
always
understand
configurable
as
like
being
deletable,
but
we
can
definitely
expand
the
the
in
this
text
like
what
it
means
to
be
deletable
as
well,
just
to
make
sure
like
we
actually
not
only
connecting
to
the
configurable
as
the
only
way
to
to
make
sure
something
is
actually
right.
C
Well,
I
wouldn't
not.
I
would
still
mention
configurable
because
since
we're
talking
about
the
the
global
object
in
the
browser,
there's
this
big
confusion
about
window
versus
window
proxy.
So
I
would
still
make
it
explicit
that
the
pro
that
the
property
is
configurable
but
the
but
the
heart,
but
the
the
important
requirement
is
that
it's
deletable.
A
C
Yeah
there's
a
the
the
current
status
quo
on
window
proxies
in
the
browser.
The
stat,
the
de
facto
standards
grow
is
actually
quite
disturbing
with
the
window
proxies
sometimes
violate
the
object
and
variants
in
ways
that
the
some
of
the
browser
makers
have
been
grumbling
about
ever
fixing,
so
so
making
making
very
clear
that
this
is
not
a
window
proxy
that
this
is
the
global
object
but
the,
but
I
would
still
I
think
it
should
still
be
explicit
that
the
properties
are
deleted.
E
C
D
B
C
A
B
In
the
same
text
that
we
make
it
that
we
make
the
requirement
for
configurable
and
we
can
not
only
add
deletable,
but
we
can
also
set
a
reminder
like
a
note
that
a
global
object
is
still
an
ordinary
object
and
we
can
make
this
reference
there.
Just
as
a
reminder
like
as
long
as
it's
an
ordinary
object.
Right.
D
B
Yeah
and
to
continue
the
report,
one
of
the
things
that
I
just
want
to
mention
on
as
next
to
pointed
out
is
the
reason
that,
like
people
are
still
showing
a
strong
preferences
over
how
this
api
should
be.
But
most
of
these
preferences
are
conflicting
with
some
requirements
like
here
and
there
most
of
them
like
from
implementers.
B
B
Milestone
like
we,
we
really
needed
to
to
go
through
that
one,
and
I
think
chrome
is
actually
okay
with
this
ahead
in
the
current
proposed
format.
We're
actually
just
doing
some
section
of
the
current
requirements
and
I
think
from
team.
The
chrome
team
will
also
be
on
on
board
with
that.
I'm
actually
gonna
confirm
with
everyone
involved,
implementers
and
etc,
but
just
trying
to
quickly
address
some
strong
preferences,
as
we
have
like
some
strong
preference
from
jordan.
Harbin
jack
works.
B
I
used
to
it's
for
me.
It
still
seems
like
clear
that,
like
with
this
api,
we
can
achieve
the
use
cases
we
are
trying
to
resolve
and
the
the
preference
go
forward
on
what
is
going
to
be
necessary
in
new
zealand
to
provide
configuration
like
setting
an
a
main
brain
framework
that
works
with
it,
setting
configurations
on
the
globals
that
are
presented.
B
So
these
preferences
are
actually
not
really
a
technical
blocker
to
address
the
use
cases
like
it's
still
solvable
in
user
land.
It
might
still
add
an
extra
step
here
and
there,
but
it's
still
possible
to
achieve
to
resolve
the
use
cases.
So
that's
why
I'm
satisfied
with
the
current
proposal
and
that's
why
I'm
I'm
like
willing
to
continue
proposing
this
current
api
in
the
next
meeting
to
advance
to
stage
three.
D
I
want
to
add
a
couple
of
nodes,
so
I
think
after
the
meeting
I
have
a
couple
of
conversations
also
with
some
parts,
including
dominic,
the
it
is
important.
I
think
it
was
very
important
that
mark
was
able
to
articulate
in
the
meeting
something
that
we
didn't
really
think
before
around
the
role
of
realms
in
terms
of
mutations
and
the
foot
gun
that
we
were
trying
to
address.
D
The
concern
for
the
foot
gun
that
crime
team
has
tr
has
been
trying
to
address
with
the
content
proposal,
and
so
on
has
been
about
the
the
identity
problems,
the
the
cross
cross,
object,
graph,
intersection
and
so
on.
It
has
never
been
about
mutations,
and
the
callable
boundary
is
not
a
boundary
for
prevent
to
prevent
mutations.
D
Is
that's
not
the
role
of
that
boundary?
The
role
is
to
prevent
access
to
objects
that
are
not
from
your
realm
and
therefore
eliminate
the
foot
gun
there.
That
we
know
is
real
and-
and
I
think,
having
that
distinction
and
making
sure
that
everyone
understand
that
realms
are
not
going
to
put
in
place
a
boundary
that
prevents
mutation
out
of
the
outer
realm
is
an
important
distinction.
D
When
you
evaluate
something-
and
you
call
it
with
a
a
argument
that
happens
to
be
a
callable
when
they
call
that
callable
that
collaborate
might
produce
side
effects
on
the
outer
realm,
there's
nothing
that
we
can
do
to
prevent
that.
Similarly,
the
import
case
of
dynamic
import
is
the
same
thing,
you're
doing
a
dynamic
input
that
thing
might
or
might
not
have
observable
changes
on
the
outer
realm
in
this
case,
in
the
module
graph
and
the
the
the
caching
of
the
monograph,
which
is
shared
between
the
two
realms.
D
A
A
A
It
is
currently
the
case
that
the
session
replaces
eval
function
and
the
and
the
module
loader
and,
in
fact,
provides
a
compartment
before
lockdown.
That
is
just
a
module
loader
and
provides
no
security
benefits
and,
in
the
same
way
that
it
replaces
the
function
with
a
container
with
a
locked
down
function,
constructor,
it
needs
to
replace
compartment
and
it
by
extension,
it
stands
to
reason.
A
It
will
need
to
also
replace
the
realm
with
a
lockdown,
a
locked
down
version
of
the
realm
in
order
to
enforce
that
containment
and
a
lot
of
the
conversations
we've
been
having
about
the
globals
in
the
realm
have
been
have
been
informed
by
a
requirement
that
we,
as
the
cess
folk,
have
brought
to
the
table
that,
for
example,
they
not
be
able
to
communicate
we've.
We
put
perhaps
an
unnecessarily
restrictive
requirement
on
realm
because
we're
thinking
about
realm.
A
C
C
Is
that
if
you
create
a
restricted
realm,
but
within
that
restricted
realm,
the
realm
api
is
exposed.
You
have
to
ensure
that
any
restriction
that
alice
imposes
on
the
bob
realm
that
bob
cannot
escape
those
restrictions
by
creating
a
new
carol
realm.
That
is
not
subject
to
those
restrictions.
So
there's
this
thing,
we
call
inescapable
restrictions,
restrictions
that
propagate
transitively
over
creating
creating
new
sub-worlds
that
are
like
the
restricted
world.
C
When
you
restrict
the
world,
you
have
to
also
transitively
propagate
those
restrictions
to
worlds
created
from
that
world.
That's
not
to
argue
against
anything,
that's
been
said,
it's
just
that!
That's
something
that
we
need
to
remember
to
be
aware
of,
and
that
some
ways
of
constructing
these
apis
might
make
it
more
or
less
hazardous
to
successfully
propagate
a
shimmed
restriction
in
an
inescapable
manner.
C
When
you
create
a
realm,
you
necessarily
create
the
start,
compartment
of
the
realm,
the
star
compartment,
meaning
the
the
there's,
an
initial
global
object
and
evaluation
context
for
the
new
realm,
but
that
new
start.
So
we
call
that
the
start
compartment
that
start
compartment
does
not
have
its
own
compartment
instance
object
because
it
was
not
created
by
saying
new
compartment.
C
It
was
created
by
saying
new
rel,
so
the
the
realm
instance
has
to
also
provide
somehow
the
functionality
that
we
would
expect
of
a
compartment
instance
with
regard
to
controlling
the
compartment
nature
of
the
start,
compartment
of
the
new
realm
in
a
similar
manner.
When
you
create
a
new
thread
or
by
the
way,
I'm
assuming
thread
means
agent
and
process
means
agent,
cluster.
C
C
C
C
D
The
possibility
to
create
mutations
or
to
produce
a
mutation
on
the
outer
realm,
a
state
mutation
on
the
other
realm
is
that,
in
order
for
you
to
create
a
mutation
on
the
outer
realm,
the
outer
arm
has
to
give
you
something
you
give
you
some
power
by
giving
you
a
callable
while
the
usage
of
import,
dynamic
import
is
not
something
that
the
outer
realm
can
control.
So
when
you
create
a
rom,
you
have
access
to
that
syntax.
Therefore,
you
can
do
dynamic.
D
Import
and
the
dynamic
input
produces
a
mutation
on
the
module
map
and
the
cache,
and
it's
a
second
cache
and
so
on
on
the
outer
realm,
which
is
what
they
have
been
complaining
about,
that
this
mutation
is
happening.
So
that's
that
was
the
the
foundation
of
of
the
discussion
that
we
have
on
one
of
the
threads.
So
I
think
that's
obviously
correct.
D
Like
yes
in
in
one
case,
you
already
have
access
to
make
those
mutations
and
there's
no
way
to
take
those
to
take
that
power,
and
in
the
other
case,
you
are
intentionally
giving
them
access
to
produce
mutations
or
to
to
make
a
mutation
on.
On
the
other
realm,
so
that's
a
okay,
I
would
say
it's
okay
compromise
in
the
sense
that
yes,
we're
not
protecting
a
mutation.
Yes,
you
can
give
something
to
the
realm
that
produce
a
mutation
or
or
allow
you
to
make
imitation.
B
B
B
Yeah
and
by
that
he
also
understood
as
a
mutation,
so
there
is
nowhere
else
in
the
spec
that
we
plan
to
have
that
we're
going
to
have
in
this
pack
saying
like
we're
just
a
long
invitation.
So
dominic's
complaint
was
because
we
once
said
we're
just
allowing
mutation
across
realms
and
now
and
then
he
said.
Oh,
the.
The
actual
report
is
a
law
invitation.
So
that's
like
you're
breaking
your
own
contract,
but
we
now.
D
We
don't
have
that
anymore,
so
you're
saying
that
if
we,
if
we
remove
that
warning-
and
we
say
simply
that
anything
that
added
by
the
host
must
be
deletable
is
sufficient
to
say
we're
not
in
the
basis
of
protecting
against
mutations
mutations
will
occur
one
way
or
another.
Yes,
they
might
be
observable
or
not,
depending
on
what
the
user
is
trying
to
do
with
the
realm,
that
we
don't
care.
Yeah.
C
So
the
I
think
the
cases
to
keep
in
mind
is
you
know,
you're
on
a
browser
and
someone
wants
to
to
emulate
the
node
environment
or
your
own
node,
and
somebody
wants
to
emulate
the
browser
environment
or,
of
course,
you
know
any
of
those
things
cross
and
embedded.
So
the
the
host
virtualization
was
part
of.
So
it's
not
just
the
security
of
of
of
lockdown
that
I'm
trying
to
ensure
it's
the
faithful
virtualizability,
and
I
think
all
this
is
consistent
with
that.
A
The
the
the
the
one
thing
that
that
we
need
to
be
careful
of
with
that
relaxation
is
to
ensure
that
lockdown
is
shimmable
and
and
to
make
and
to
make
it
possible
to
shim
realm
locked
at
lockdown
realms
as
well.
I
think
that
we're
covered,
but
just
to
talk
it
through,
supposing
that
you
are
is
a
fly
in
the
ointment,
is
that
the
dynamic
import
of
a
child
realm
is
not
deniable,
so
that
the.
A
But
I
think
that
that
I
think
that
the
story
remains
the
same
in,
in
the
absence
of
an
implementation
of
compartment
being
added
to
this
added
to
the
a
specification
being
created
and
the
implementation
being
added
to
a
vendor
implementation.
We
still
have
to
shim
both
compartment
and
locked
down
realm,
and
I
think
that
that
puts
us
in
a
position
to
deny
dynamic
import,
as
we
we
already
have
are
in
a
position
to
deny
dynamic
import.
It's
not
ideal,
but
with
our
emulated
module
system
and
the
censorship
that
the
session
does.
A
We
can
do
that.
The
the
then
the
question
so
locking
down
a
realm
is
a
matter
of
replacing
the
realm
constructor
with
a
new
realm
constructor
that
implicitly
runs
lockdown
before
running
any
client
code
in
a
child
realm
such
that
it
denies
it
the
ability
to
do
dynamic,
import
and
grants
it
only
ability
to
communicate
through
the
through
the
the
incubator
realm.
C
I
think
so
another
another
restriction
to
to
just
notice
is
kind
of
in
the
same
left.
Side
of
your
diagram
versus
right
side
of
your
diagram
transition
is
that
post
lockdown,
sloppy
evaluation
is
not
possible
and
that's
through
the
entire
hierarchy
of
evaluators.
It's
just
you're.
Now
in
a
world
in
which
in
which
the
concept
of
sloppy
evaluation
has
been
removed.
From
from
you
know,
it's
no
longer
a
concept.
C
It's
no
longer
something
that
that
one
can
a
way
to
evaluate
anything,
even
if
you
cr,
even
if
from
the
realm,
you
create
a
new
thread
or
a
new
process
in
the
lockdown
world,
you're
creating
a
strict
realm
and
a
strict
process.
So
I
think
it's
it's.
I
think
all
of
these
restrictions
go
together
in
one
coherent
story,
since
we
weren't
willing
to
make
the
left
side
the
normal
realm
api
require
strict.
Only
then
all
the
rest
of
this
sound
sounds.
You
know
in
a
similar
category.
C
A
Environment.
Another
thing
that
I
like
about
this
particular
route,
which
is
somewhat
dangerous,
is
that
that
it
makes
it
possible
to
strike
a
better,
a
better
compromise
between
security
and
performance.
In
some
cases
like,
for
example,
we
like
what
the
the
true
the
safest
option
to
do
for
fetch
is
to
virtualize
fetch
such
that
anything
that
is,
that
calls
fetch
is
virtu,
is
virtualized
through
the
parent,
the
parent
realm
right,
so
it
would
have
to
so
so.
A
A
request
would
have
to
transit
the
realm
to
realm
boundary
and
the
fetch
would
have
to
occur
in
the
parent
and
then
and
and
subject
to
any
attenuation.
But
that
comes
with
a
performance
penalty
that
might
not
be
acceptable
in
all
cases,
and
it
might
be
necessary
in
some
cases
to
compromise
and
say
that
the
lockdown
realm
api
may
have
performance
motivated
options
of
its
constructor
to
say,
if
I
did
have
direct
access
to
fetch,
I
can
bequeath
that
to
my
child
realm
for
performance
reasons,.
B
Yeah,
so
when
we
reached
10
30,
I
text
jordan
and
saying,
like
yeah,
I
hope
to
see
in
the
see
you
in
the
meeting
he
first
told
me
he
didn't
have
an
invite,
but
I
I
confirmed
I
had
this
like.
I
talked
to
jordan
on
last
wednesday
after
the
meeting
in
the
afternoon,
and
we
I
thought
about
like
having
him
joining
us
for
to
discuss
this,
which
I
confirmed
like
I
sent
the
calendar,
invite
I
actually
sent
the
the
the
sas
meeting,
invite
I
invited
his
email.
B
He
told
them
that
he
had
conflicted
meetings,
so
he
cannot
join
us
today
anymore,
and
then
I
checked
my
calendar
again.
He
his
email
doesn't
show
up
there
anymore.
My
work
email
was
not
showing
up
in
this
calendar,
invite
as
well.
On
monday,
I
I
I
have
my
personal
and
my
work
email.
I
always
have
like
double
triple
checks
for
sas
media
and
I
have
like
a
tc39
calendar.
I
have
for
my
personal
email
and
for
my
work,
email.
It
was
not
showing
there.
B
He
cannot
come
next
week.
I
already
talked
to
him
like
I.
I
just
had
the
answers
when
I
was
talking
to
you,
that's
why
I
was
having
like
interruptions,
because
I
was
just
receiving
the
news,
like
he's
not
being
been
able
to
join
today
he's
off
next
week.
He
can
come
up
like
the
second
wednesday
after
this
that
worked
yeah
yeah,
so
other
people
have
work,
email
that
disappeared
from
the
meeting
invites
like.