►
From YouTube: SES Meeting: Near Membrane Performance
Description
Mathieu shares progress on symbols as weakmap keys and Leo and Rick share performance figures for near membranes.
A
Welcome
to
the
assess
meeting
today
is
april.
6Th
and
the
topic
for
today
is
symbols
as
weak
map
keys,
which
we,
which
I
hear
were
discussed
at
a
records
and
tuples
meeting
yesterday
right
and
and
we're
going
to
call
it
a
short
meeting
unless
that
topic
stretches
out,
and
we
have
decided
that
we
will
defer
the
topic
that
leo
brought
for
today
in
order
to
give
rick
and
rick
and
and
team
an
opportunity
to
prepare
material.
If
that's,
okay,.
B
D
Yeah,
so
a
real,
quick
update.
Yesterday
there
was
the
records
and
tuples
monthly
meeting.
So,
as
we
all
know,
there
is
no
object,
placeholder
or
box
anymore
anywhere
near
the
recording
double
proposal.
The
approach
is
going
back
to
using
symbol.
As
with
map
keys,
there's
been
contention
about
what
kind
of
symbol
to
allow
in
which
map
keys
and
what
kind
of
symbols
to
disallow.
D
The
next
type
of
symbol
is
a
unique
symbol.
Everybody
agrees
that
those
should
obviously
be
usable.
The
last
kind
of
symbol
is
well-known
symbols
which
are
expect
as
unique
symbols.
However,
they
are
somewhat
permanently.
They
will
never
really
be
collected
because
their
intrinsics
that
need
to
always
be
there.
Some
of
them
are
undeniable
in
the
current
realm,
maybe
all
of
them,
some
of
them
can
always
be
reached
again.
If
you
create
a
new
realm,
and
so
at
the
end
of
the
day,
it
doesn't
make
much
sense
to
use
those.
D
E
E
E
D
I
think
that
sounds
like
I
mean
it.
It
sounds
like
a
bit
of
a
niche
case.
I
would,
I
would
put
it
under
the
same
kind
of
edge
case
like
if
someone
deletes
the
symbol
out
of
the
global
and
delete
shadow
realm
or
any
way
to
create
a
new
realm.
They
would
not
be
able
to
get
at
some
of
the
intrinsic
symbols
anymore,
either.
B
B
B
E
D
Unfortunately,
no
so
that's
what
we
wanted
originally
a
predicate
that
says
like
is
something
usable
as
weak
map
key.
However,
the
position
seems
to
be
that
a
predicate
that
just
returns
a
value
if
there
is,
should
not
change
its
its
answers
over
changes
in
the
spec.
So
if
today
we
don't
allow,
if,
tomorrow,
we
don't
allow
records
and
tuples
as
weak
map
keys
and
recorders
and
phones
introduce,
and
that
predicate
first
says
like
no
in
the
future.
That
predicate
will
always
have
to
say
no,
even
when.
D
Records
and
tuples
some
of
them
may
become
usable
as
witness
keys,
so
the
the
position
of
delegates
on
on
that
which
I'm
not
surprised,
but
I
I
do
understand.
The
argument
is
that
if
something
throws
it's
fine,
if
you
build
a
predicate
that
says
true
or
false
based
on
that
same,
would
it
throw
or
not
it's
not
okay,
to
have
that
as
an
api.
E
B
E
D
D
The
complexity
of
course
comes
from
the
fact
that
records
and
tuples
don't
have
a
live,
really
livened
definition
of
their
own,
but
that
their
liveness
is
a
compound
concept
of
all
the
other
weak
map
keys.
They
contain
so
basically
all
the
deeply
nested
symbols
that
they
may
contain
unique
symbols
that
they
may
contain.
D
So
it's
surprising
to
hear
that
implementations
have
that
concern
and
they
may
be
pushing
back
on
that,
especially
because
ashley
actually
created
a
a
shim
in
in
entirely
in
javascript.
That
allows
to
do
this.
I
need
to
look
at
it
again
but
to
see
if
it
doesn't
have
other
kind
of
memory
leaks
issues,
but
it's
yeah
it's
it
would
be.
B
B
B
B
B
I
don't
know
how
to
remove
this
prevail,
screen,
yeah,
so
yeah.
I
am
using
this
near
membrane
to
get
like
some
quick
things
here
as
like
I'm
creating
an
instance,
and
I
have
some
dome
manipulation
inside
this
near
membrane,
just
to
make
sure
like
kind
of
like
I
want
to
reflect
a
little
bit
of
final
usage,
but
this
is
just
like
touching
the
surface
of
what
we
can
use.
I
do
some
evaluation
here.
B
B
I
think
there
is
much
more
one
of
the
out
of
curiosity
curiosity.
I
was
trying
to
load
this
and
I
like.
I
saw
that
if
I
wanted
to
load
like
a
thousand
namespaces,
which
is
an
exaggeration,
I
couldn't
load
a
thousand
namespaces
with
iframes,
but
I
could
easily
load
a
thousand
namespaces
using
shadow
realms.
That
speaks
a
lot
to
the.
C
C
B
But
like
with
the
console,
I
get
like
speeds
up
like
very
much
faster,
but
I
need
to
understand
because
this
is
the
memory
footprint
that
is
talking
about
when
I
have
the
console
open
but
like
for
execution
time.
I'm
not
seeing
like
a
super
firm
result
right
now
to
share
it's
like
I,
I
can
tell
shadow
realms
so
far
with
the
neo-ring
brains
and
keep
in
mind
the
new
main
brains.
B
As
I
understand
they
are
not
highly
optimized
for
shadow
realms,
yet
they
are
optimized
today
for
the
iframes
work,
they're
like
using
yeah
rick.
I
know
I
see
your
face.
I
just
seen
like
I
I
expect
like
the
more
we
have
shadow
rounds,
the
more
we
can
focus
work
on
optimizing
users
on
shadow
realms
or.
C
C
C
I
shouldn't
have
said
that.
Let
me
just
say
what
was
I
saying
near
membrane
is
currently
designed
with
the
shadow
realm
constraints
with
that
being
primitives
and
callable
passing
only
it
has
been
since
last
summer
into
fall
time
when
charity
and
I
oh
I'll,
say
carrie,
let
it
and
I
just
made
sure
that
it
was
on
the
rails
with
testing
and
stuff,
but
yeah
it
is
it's.
C
It
is
definitely
can
always
be
further.
Optimized
and
john
david
dalton
has
been
like
really
like
pressing
the
edges,
trying
to
figure
out
ways
to
reduce
the
number
of
proxy
trap.
It's
like
think
of
it
as
like
pile
up.
You
know
like
just
the
meta
object.
Programming
has
like
trapped
pile
up,
which
is
you.
A
C
Just
it's
acceptable
because
that's
just
how
net
object
programming
is
in
javascript,
it's
not
something
that
needs
to
be
changed,
but
it
is
indeed
something
that
we
are
able
to
figure
out
ways
to
optimize
around.
I
would
say
when
you're,
when
you
run
through
the
shadow
realm
path,
you
are
indeed
experiencing
an
even
playing
field
with
the
iframe,
I
mean
it's
actually
an
improved
playing
field.
C
Frankly,
because
I
don't
so
to
actually
matthew
answer
your
question,
I
don't
actually
anticipate
there
being
any
real
difference
in
terms
of
improvement
with
shadow
realm,
evaluate
versus
you
know,
eval
and
iframe.
They
don't.
Actually.
I
never
really
anticipated
that,
like
that's,
not
even
something
that
I
had
considered.
C
What
I
did
consider
was
the
initialization
cost
being
drastically
improved
and
that
you
know
that
was
like
my
you
know.
I
theorized
that
this
we'll
see
drastic
improvement
in
initialization
costs
of
creating
our
near
membrane
and
basically
building
the
near
membrane
world
inside
of
the
shadow
realm
that
has
certainly
improved
I'm
frankly,
it's
because
we're
not
creating
an
iframe
right,
we're
not
doing
we're,
not
we're
not
relying
on
dom
stuff
anymore,
to
build
our
virtual
environment.
We're
going
straight
to.
E
C
But
I
hadn't
given
much
thought
to
what
should
we
expect
in
terms
of
performance
of
actual
operation
execution
between
the
two
and
so
far?
I
don't
think
we're
seeing
anything.
That's
we're
not
seeing
any
any
negative
impact.
That's
for
sure
for
sure.
I
would
have
to
think
a
little
bit
more
deeply
about
how
we
could
how
we
could
measure
that
in
a
useful
way,.
D
I
wasn't
certain
how
your
near
membrane
was
structured,
but
I
figured
like
you
might
have
one
case
where
the
near
membrane
runs
on
top
of
the
callable
boundary
and
then
in
the
iframe
case.
You
basically
have
near
membrane
running
on
bare
metal
where
you
don't
have
the
wrappers
in
between,
but
you
still
have
the
you
call.
C
Gotcha,
so
in
order
for
us
to
do
that,
we
would
have
to
grab
a
historic
release
that
predates
our.
So
we
did
an
interesting
thing.
We
we
had
like
another
another
goal
internally
for
our
the
project
here
at
salesforce
that
uses
the
near
membrane
library,
which
I
won't
go
into,
because
it's
right,
because
I
can't
and
to
unlock
that
capability.
C
We
had
a
few
obstacles
in
your
membrane
and
carrie
actually
had
the
sort
of
the
foresight
and
realize
like.
If
we
just
do
the
work,
we
need
to
do
for
shadow
realms
now
now
being
like
last
summer,
then
that
will
help
us
unlock
this
other
capability.
So
we've
actually
had
a
callable
boundaries.
Implementation
that
we've
been
imposing
on
our
iframe
based
near
memory.
C
C
C
There
were
notable
regressions
but
the
for
our
case
security
trump's
performance
at
salesforce.
So
we
were
able
to
basically
say
like
this
is
a
cost
that
we
knew
would
happened
so
we're
going
to
float
it
through
on.
But
the
answer
to
your
question
is
that
yes,
that
certainly
does
have
have
an
unfortunate
additional
cost.
C
We
only
recently
solved
the
how
to
do
this
lazily
and
then,
and
it
has
that
really
only
affects
initialization
time
so,
but
when
you're
in
band
like
everything
that
you
want
to
access
across
the
boundary,
which
is
just
like
everything
that
is,
you
know
dom
related
right.
The
cost
has
grown,
notably.
D
F
I
think
I
think
they
they
the
conversations
that
we
have
with
igalia,
at
least
from
what
I
remember
that's
doable,
but
it
will
not
help
that
much
wouldn't
right.
They,
in
fact
when
they
did
the
profiling
of
the
neo-membrane
implementation
that
we
have
and
so
on.
There
were
bigger
fish
to
fry,
especially
the
get
on
property
descriptors,
for
example,
that
one
was
a
big
one.
So
there
are
other
things
that
they
believe
will
give
us
a
a
little
bit
of
an
edge.
F
D
F
D
F
F
We
just
need
to
do
a
lot
more
tests
and
the
optimizations
are
going
to
be
our
primary
case
for
the
next
few
months.
I
would
say,
like
continue
optimizing
that
I
think
jdd
and
rika
has
been
wonderful,
trying
to
figure
out
what
what
the
heck
is
slowing
down
the
thing
there
is
a
lot
of
back
and
forward
between
the
two
sides
of
the
membrane.
That's
that's
the
biggest
problem
like
you
have
to
go
back
and
forward
thousands
of
times
on
any
operation,
and
that's
like
so
we'll
see.
C
That
is
like
virtualizing,
like
it's
a
virtualized
real
window
and
it's
I.
I
think
that
there's
something
going
on
and
there
was
a
leo
linked
me
to
a
maybe
what
wig
or
some
damn
thing.
I
have
no
idea
conversation
about
event,
error
or
target
or
something
I
don't
know
if
that
is
directly
related,
but
the
global
object
not
being
not
having
event
target
in
its
prototype
chain
inside
the
shadow
realm
when
you're
trying
to
map
that
global
to
vehicle.
This,
too.
C
D
C
Yeah,
okay,
so
I
don't
think
it's
a
problem
for
this
group,
but
I
would
like
to
get
this
group
thinking
about
it
so
that
we
can
like
pick
pile
on
whoever's
problem.
It
is
to
like
get
it
solved
when
the
time
comes,
because
that
is
actually
our
only
here's
a
weird
funky
funky
thing,
and
that's
basically
how
I
how
I
was
like
it's
like
when
you
get
to
this
thing.
It's
like
why
is
like.
Why
is
why
is
our
like
dunder
proto
chain,
not
working
correctly
here
either?
C
D
C
C
F
If
I
remember
so,
they
for
the
prototype
of
the
global
object
is
pointing
to
a
proxy
of
the
prototype
of
the
incubator
wrong.
Yes,
because
that's
possible
in
the
shadow
run,
but
not
possible
in
iframe.
So
in
an
iframe,
we
have
to
do
a
lot
more
work
because
we
cannot
replace
this
unforgivable,
so
we
have
to
go
and
wipe
out
whatever
that
object
has
and
then
put
a
bunch
of
linkage
to
the
other
one.
F
But
so
one
thing
that
I
do
want
to
spend
time
on
and-
and
I
promise
a
couple
of
times-
but
we
never
really
materialized.
It's
like
really
come
coming
to
this
meeting
and
explaining
what
we
have
more
of
the
perspective
of
how
the
people
might
be
able
to
use
that
specifically
aquatic.
I
think,
there's
a
lot
of
stuff
in
there.
C
D
F
F
Thing
that
that
we're
doing
is
virtualizing
the
incubator
wrong,
so
we
do
a
full
virtualization,
not
a
full
video,
but
every
translation
of
the
main
window.
So
when
you
run
code
inside
the
shadow
room,
it
looks
like
it's
running
on
the
main
one
he's
just
running
inside
the
shadow
room
and
the
reason.
F
Because
we
want
to
prevent
side
effects
or
some
side
effects
on
the
main
window
on
the
integrator
round,
like
if
you're
trying
to
modify
prototypes
or
trying
to
modify
objects
that
are
dom
api
stuff,
like
that,
it
will
not
leak
into
the
main
window.
So
that's
the
the
first
step.
The
second
step
is
that
to
be
able
to
control
any
evaluation,
carry
on
by
the
sandbox
code,
the
code
that
we
run
inside
this
virtualized
environment.
F
So
if
they
do
any
kind
of
revolution,
eval
import
whatever
they
do,
that
thing
will
will
continue
to
run
inside
the
virtualized
environment.
It
will
never
leave
the
virtual
environment.
So
as
it
does
that's
a
use
case
there,
there
are
things
that
we
want
to
prevent
that.
We
don't
know,
we
don't
want
them
to
out
to
do
and
those
we
go
through
distortions.
F
They
are
other
cases
that
are
more
complicated
distortions.
Like
a
good
example
of
it
is,
like
you
insert
an
iframe
in
the
page
in
the
main
page,
and
normally
you
will
be
able
to
access
that
iframe
by
doing
a
window.frame,
the
frame,
zero
frame,
one
frame,
two,
whatever
the
frame
is
or
simply
just
doing
window
square
bracket,
zero,
one,
two.
Whatever
so,
don't
think
those
things
you
don't
see,
you
don't
see
them
inside
or
you
might
want
to
see
them.
We
don't
know
yet,
but
there's
some
debate
there,
but
those
are
complicated
distortion.
F
When
it
comes
to
the
assertions
of
a
simple
function,
it's
easy
because
you
just
want
to
swap
out
the
the
function
with
a
different
functionality
and
that
that
works
well.
But
the
main
goal
is
that,
like
the
code
runs
in
there,
it's
controlled
anything
that
is
the
the
code
is
attempting
to
use
or
call
or
whatever
trying
to
do.
We
we
might
have
control
over
that
by
having
some
distortions-
and
it
looks
like
it
is
a
main
window.
So.
F
F
F
Just
to
highlight
one
thing:
it's
like
this
virtualization
engine.
We
also
use
it
for
node
and
we
also
want
to
use
it
for
escort
or
any
javascript
environment.
So
the
core
of
it
is
the
same.
You
have
an
environment
that
has
a
global
object.
We
want
to
virtualize
that
to
have
a
high
fidelity
virtualization
of
that
thing
in
a
way
that
code
that
runs
inside
the
sandbox
is
in
some
ways,
control
go
ahead.
D
D
Your
shim
can
grab
that
listen
to
those
and
have
a
distortion
that
when
the
code
goes
through
the
window
or
whatever,
like
the
prototype
from
the
target,
to
add
an
event
listener
on
the
on
the
real
on
your
version
of
a
global
disk
and
checks
for
some
evidence
like
it
would
either
get
the
evidence
from
the
incubator
realm
or
for
an
angled
errors,
or
things
like
that,
grab
them
from
from
the
local
source.
Well,.
F
F
I
believe
correct
me
if
I'm
wrong,
but
still
you
have
event
target,
you
can
create
things
and
you
can
do
things
without,
even
targeting
in
the
same
way
that
you
can
do
it
in
in
the
in
the
dom
sorry
in
the
in
a
page
in
a
window
without
having
to
do
anything
with
the
dom
itself.
So
it's
not
down
bound
at
that
point.
So
for
us,
none
of
that
matters
really
because
we're
going
to
replace
everything
anyways,
because
we
want
a
virtualization.
F
F
Yes,
so
if
I
can
replace
it,
then
I
will
replace
it
unless
there
is
a
strong
opinion
on
not
replacing
certain
things,
and
we
do
that
for
certain
things
that
we
just
keep
it
from
the
arena
shadow
round
intense,
but
most
of
the
things
we
just
replace
it
entirely,
but
I
I
think
the
the
event
target
thing
from
that
issue
that
was
happening
a
few
weeks
ago.
I
think
I
agree
that
should
be
I'm
there.
F
F
F
Yeah-
and
I
do
agree
with
your
proposal
having
something
in
the
global
option
that
can
mimic
that
thing
and
can
be
standardized
and
so
on.
So
I
I'm
I'm,
I'm
I'm
supportive
of
that.
I
think
it's
going
to
be
more
clear
that
you're
dealing
with
a
place
where
errors
will
occur
and
you
can
listen
for
them
any
platform
and
they
make
sense
for
what
you
want
to
do
and
so
on
so
yeah,
I'm
okay
with
that.