►
From YouTube: Parallel Rustc Planning Meeting 2019-10-07
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
B
B
So
yeah
there's
a
link
in
so
the
porpoise
is
there
there's
also
links
throughout
the
dock
to
like
pieces
of
code
and
such
so.
We
have
basically
there's
this
structure
that
contains
the
current
dependency
graph,
which
we
edit
as
we
go
through
the
compiler.
Like
lorries,
are
running
such
the
current
situations.
We
just
keep
the
whole
structure
behind
one
block
and
acquire
it
whenever
we
want
to
change
anything
or
commuting
thing.
A
B
B
A
B
A
B
A
B
B
A
B
A
B
A
A
B
A
A
B
A
B
A
A
B
B
A
A
A
A
A
B
A
B
Is
one
other
concern
with
total
recount
in
total
duplicate
recount
we're
moving
them
to
its
moments?
And
there
is
sort
of
this
weird
situation
where
we
are
such
add
with
eventually
consistent
ordering,
which
is
fine.
But
then,
when
we
read
them
where
for
some
reason
using
acquire
ordering,
which
is
correct.
But
it
seems
odd.
B
A
A
B
B
B
A
A
The
real
the
real
boundary
is,
the
dev
graph
I
think
is
the
answer,
whether
that's
okay,
I
guess
it's
a
question,
but
this
is
the
real
structure.
Mm-Hmm.
A
B
A
A
A
B
A
B
This
sure
yeah,
so
we
have
the
limb
store,
is
yeah.
We
have
three
store
what
we
call
white
pass
objects
which
gets
into
sort
of
a
fourth
of
our
registration
system
where,
instead
of
registering
links,
you
register
essentially
passes
which
are
like
visitors,
and
we
will,
when
we
get
around
to
running.
Let's
take
the
list
of
passes
out,
feeling
it
and
then
muted.
B
Li
of
now,
a
unique
pointer
to
the
each
pass
objects
run
the
pass,
which
basically
involves
traversing
like
a
tree
of
share
nodes
or
whatever,
and
then
that
emits
into
like
session
Diagnostics
and
such
and
that's
like
that,
we
have
the
fog.
The
other
reason
is
that,
when
we're
sort
of
building
up
session,
we
don't
always
have
unique
access
to
it
and
like
a
plug-in
or
other
sort
of
places
in
the
comparable.
If
we
want
to
register
things.
A
B
B
A
B
A
Apart,
so
if
we
were
to
refactor
this,
the
hardest
part
probably
would
be
coordinating
Clippy
like
it,
doesn't
sound
that
hard
to
register
a
builder
does.
B
It,
the
complication
in
some
sense,
is
that
currently
for
performance
reasons,
we
don't
actually
like
the
compiler
itself
registers
one
element
into
each
of
these
vectors
by
combining
all
of
the
visitors
into
one
single
like
visitor
and
calling
itself
recursively.
When
you
register
a
builder,
you
sort
of
I
mean
you
could
do
something
similar,
I
guess,
yeah.
B
A
B
B
A
B
B
A
Could
enforce
its
on
Hawaii
but
I
think
it's
ok
to
document
for
now.
Why
is
this
all
public?
By
the
way?
Maybe
we
could
enforce
it
by
like
a
little
bit
better
by
am
making
this
field
not
public,
so
that
when
you
yeah.
B
A
B
B
A
Be
how
would
that
even
work
like,
ideally,
you
would
make
the
lid
store
and
set
it
up
and
then
create
the
session
and
give
it
a
setup
and
story.
Would
that
be
super
hard
to
do
probably.
B
That
is
with
due
to
plugins
and
like
impossible,
modular
plugins.
The
main
concern
is
that
they
I
think
we
could
probably
throw
it
inside
like
to
a
context
like
steal
it
from
the
session
and
then
say
everyone
should
be
using
it
from
there
as
like,
a
just
a
read-only
access
via
like
a
method
or
whatever
it
would
be,
not
super
party
I.
Imagine
that's.
B
A
When
you
said
frozen,
were
you
imagining
like
a
rewrite
lock
that
you
could
sort
of
hidden
inside
of
another
thing
and
you
can
call
a
freeze,
method
and
sort
of
at
that
point?
It
no
longer
permits
right
locks
right,
and
maybe
it
has
some
unsafe
code
to
let
you
get
an
interior
reference
out
or
something
right.
B
A
A
So
it
seems
like
we
got
two
things
then
first
we
can
clean
up,
so
it
doesn't
happen,
Tyrion
inability
and
then
we
can
look
at
how
to
take
this
session
pattern
and
either
add
a
frozen
to
make
it
explicit
or
move
to
TCX
query
I
mean
what
you
could
even
do
is.
Have
there
be
an
actual
query
that
steals
from
the
list
from
the
session?
It's
not
like.
I
love,
it's
actually
how
we
want
to
do
it
when
you
first
use
the
query
at
that
point
it
gets
stolen.
B
A
A
A
Lock
of
option,
or
whatever
inside
of
mints,
store
and
read,
register
a
builder,
they
don't
pass
itself
I
wonder
if
it
would
be
worth
pinging
money
or
Ollie
to
ask
them
about
that.
If
there's
like
some
reason,
I
feel
confident
that
some
lids
do
something
horrible,
though
I
don't
know
what
it
is,
not
sure
exactly
what
that
would
be.
A
A
B
A
B
B
A
B
And
then
we
also
have
some
links
at
the
bottom
of
the
planning
dock
down
here,
yeah.
So
there's
two
links
to
perf
suite
with
sort
of
virtual
cores
and
just
physical
cores,
which
corresponds
to
J
4
and
J
on
the
first
server
and
then
there's
also
a
link
to
j1
or
not
jin-hwan,
but
like
threads
equals
1.
It's
just
like
worst
case
and.
A
B
B
A
B
A
A
B
A
A
B
B
A
A
B
B
A
B
A
B
A
Don't
know
my
immediate
concern
for
rayon
is
just
laning
that
rebase
and
I
have
pinged
azox
heel
on
it
and
not
got
any
reply.
So
I,
don't
know
if
you
or
someone
else
can
take
the
reins
on
that
I
think
we
can,
because
we're
literally
under
rebase
the
only
thing
that
could
be
hard,
which
I
should
be
able
to
get
I
get
things
asked
you
that
I
think
at
the
moment,
if
I
recall,
roxy
is
the
only
owner
of
this
crate.
Oh
no!
It's
not
true
anymore.
Okay,
very
good
check
the
core
too,
but
yeah.
A
A
A
I
think
I'd
rather
talk
about
the
jobseeker
stuff,
because
that
feels
like
more
important
and
also
relevant
right
and
it's
related
to
ran
because
it's
one
of
the
patches
was
to
put
hooks
every
time,
Rihanna
sleeps,
basically,
and
it's
releasing
an
acquiring
job
server
handles
when
it
does
that
right.
One
of
the
things
I
was
just
thinking.
A
One
of
the
things
we
could
do
as
an
outcome
from
this
like
meeting
is
start
to
develop
some
parallel
guidelines
and
it
feels
like
we
came
up
already
with
some
heuristics,
but
we
didn't
write
them
down
like
if
you
have
interior
mutability,
you
should
do
XXX
documentary
lock
order
and
try
to
have
a
module
boundary.
That's
strong
and
also
the
patterns.
We
are
seeing
like
the
initialization
pattern,
you
talking
about,
isn't
a
general
guide
or
a
rest.
A
C
specific
guide,
rusty,
specific
I'm
talking
about
I'm
thinking
about
a
rusty
guide
chapter
with
here
are
the
guidelines
and
we
have
audited
all
the
code.
But
new
code
should
be
audited
against
this
guidelines
right
and
we
have
to
sort
of
develop
them
as
we
go.
But
our
guidelines
set
the
right
term.
I,
don't
know
guidelines
and
conventions.
A
A
Atomic
orderings
you
sequentially
consistent
everywhere,
unless
there
is
a
actually
I
wonder
if
they
could
do
strong
reason
to
to
otherwise.
If
so,
it
should
be
documented,
I
wonder
if
we
could
make
a
constant.
Actually,
oh
I've
noticed
one
thing
I
find
annoying
about
orderings.
Is
that
if
you
just
use
sequentially
consistent
everywhere,
there's
no
document
like
it's
not
obvious.
If
you
require
that
for
some
reason
or
you
just
didn't
want
to
think
about
it,
I
would
never
forget,
make
a
constant.
That's
like
some
ordering
I
didn't
want
to
think
about.
It.
B
A
B
A
B
A
With
that
I
think
it'll
still
be
mildly,
ambiguous
I
would
as
long
as
we
doctor
and
I
think
it
would
be.
Fine,
like
I,
could
see
then
see
you're
saying
if
it's
crucial
that
you're
used
sequentially
consistent.
You
should
use
the
explicit
version,
that's
pretty
unusual.
It
happens
in
rayon
here
and
there
I.
B
A
B
A
A
A
B
A
B
A
Should
yeah
should
definitely
pink
Alex,
because
I
think
the
other
thing
to
discuss
would
be
like
how
the
olivine
that's.
A
A
This
all
sounds
great.
Do
we
want
to
the
only
other
thing?
Was
this
guidelines
thing
I
think
shouldn't
take
very
long
just
once
like
a
preliminary
Russy
guide,
PR
or
something.