►
From YouTube: Lang Team Triage Meeting 2019-11-21
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).
B
A
A
A
Some
notes
on
what
we
were
saying:
what
central
and
I
were
chatting
a
bit
about
it
in
the
pre
RC
pre
triage,
and
we
were
saying
that
maybe
we
want
to
kind
of
propose
that
encourage
experimentation
by
right.
C
C
This
fall
under
the
purview
of
the
the
working
group
that
josh
has
repaired.
He
with
Sears
that
not
really
this.
A
That's
a
good
point,
though:
I
was
going
to
write
a
comment
on
this
I.
Don't
exactly
remember
what
I
was
gonna
say,
probably
that
we're
not
gonna,
but
we're
exploring
some
options
here
and
we'll
try
to
address
the
we'll
make
a
decision
in
the
FFI
online
working
group
when
we
get
to
it.
But
I
didn't
I'll.
Try
to
do
it
today,
Felix
to
urge
this
6
4,
2,
7,
3,
I,
think
Felix.
You
got
some
concerns
around
I.
C
Rolled
a
comment,
and
then
I
realized
I,
didn't
understand
some
the
notes
in
line
team
meeting
and
so
I
worked
to
try
to
make
sure
I,
understood
them
and
once
I
didn't
understand
them.
I
said
we
probably
do
want
the
documentation,
that's
in
request.
Basically,
there
was
someone
mentioned.
We
should
have
Atlanta
doc,
saying
how
out
of
line
modules
are
treated
by
macros
and
there's
what
macros
see
they?
Don't
they
don't
see?
The
micro
isn't
gonna
automatically
jump
to
the
file
and
give
you
the
tokenized
stream
from
the
file.
C
It's
just
gonna
give
you
the
fact
that
there's
that
identifier
and
a
semicolon
or
whatever
and
at
first
I,
was
like.
Why
would
we
bother
documenting
this
and
snare
anyway?
You're
gonna
get
a
stability
error
and
then
I
realized
the
macro
it
does
run
and
I
could
does
see
this
thing,
and
then
you
see
the
error.
C
C
C
C
C
A
Okay,
one
thing
about
so
absolutely
cancel
default,
behavior
of
unwinding
I'll.
Do
that
so
anything
else
to
say
on
it.
I
do
anything
about
uplift.
Oh
I
brought
this
up
in
the
I
brought
this
up
in
the
sink
meeting
and
Manish
was
like.
If
I
can
summarize,
roughly
speaking
said,
sounds
fine.
We
think
the
compiler
team
should
do
that
work.
A
It
seems
like
there's
a
sort
of
impassive
I
think
my
impression
was
that
the
Clippy
team
wasn't
too
enthusiastic
about
going
through
and
making
a
bunch
of
rain
appears.
So
maybe
there's
some
more
discussion.
I
need
it
here,
but
I
I
think
the
plan
to
do
it.
Piecemeal,
like
piecemeal,
is
fine.
The
question
is:
will
do
it?
Probably
nobody
mess
Eames?
Okay.
If
nobody
cares
enough
to
write
the
PRM.
B
A
A
D
A
A
A
We
ran
the
results.
Sixty
create
versions
for
dressing
I
haven't
dug
further
into
this,
but
it
seems
to
back
up
the
you
can
see.
A
lot
of
them
are
also
the
same,
create
just
different
versions
of
the
same
grade.
It
seems
to
back
up
the
idea
that
the
impact
of
this
is
non-trivial
but
focused
on
a
relatively
narrow
set
of
crates.
I'm,
not
sure
what
I
should
do
is
the
next
step.
Actually
so,
I
wouldn't
mind
a
little
thoughts
on
that.
A
A
A
B
A
B
A
D
Aborting
on
unsafe
reporting
on
unwind,
we
didn't
say
a
lie,
except
that
I
was
supposed
to
write.
Something
and
didn't:
do
you
have
opinions?
I
would
I
mean
I,
I,
think
I
stated
them
all
in
missoula,
but
I
would
say
that
those
comments
are
possibly
worth
reading.
If
Lange
team
members
haven't
done
so
yet
I
can
provide
a
link
in
the
discord.
D
D
So
Josh
in
that
thread
had
said
it
seems
wrong
to
change
code
Jen
based
on
the
presence
of
unsafe
and
I,
am
I
agree
with
that
pretty
strongly.
So
my
vote
would
be,
you
know,
choose
one
of
the
other
two
approaches
either
abort,
always
on
unwind
across
death.
If
I,
you
know
just
stabilize
that
behavior
that
was
previously
stabilized
and
then
backed
out
or
just
stop
emitting
no
unwind,
which
was
another
PR
right,
I.
Think,
okay,.
A
I
think
what
I
planned
to
write
was
that
we
will
not
do
this.
You'll
basically
do
keep
status
quo.
While
we
try
to
investigate
overall
strategy
want
to
take
and
then
based
on
that.
That
would
inform
our
actions
here,
which
I
guess
is
not
neither
pro
or
con,
but
definitely
not
that
we
would
land
that
PR,
because
I
think
we're
all
there's,
definitely
no
census
around
unsafe.
Having
a
semantic
effect.
D
D
Don't
think
the
decision
can
rely
on
the
assumption
that
the
opt-out
mechanism
mentioned
above
will
be
ready
soon.
The
private
group
does
not
currently
have
a
design
recommendation
ready
nor
any
estimated
timeline
for
delivering
such
a
recommendation
and
therefore
I
think
it's
in
the
rest
of
projects
best
interest
that
one
of
the
three
immediately
available
approaches
be
stabilized
as
soon
as
possible.
D
A
B
B
B
C
No
I've
been
I've,
been
following
this
discussion
a
little
bit
but
I
think
right
now
it
looks
like
there's
a
disconnect
between
what
Ralph's
interpretation
of
what
the
semantics
is
possible.
The
lbm
developers
disagree
about
the
intended.
Semantics
means
writing.
Sorry,
though,
I'm
jumping
in
without
giving
background
information,
the
there's
a
bunch
of
different
issues
here
that
are
all
sort
of
collected
in
my
head,
but
the
biggest
one
that
we're
also
pointing
out
is
that
we
currently
with
this
dereference
of
the
law,
attribute
unbox
and
the
semantics
of
dereference
a
little
as
it
stands.
C
Today,
we
think
is
meant
to
mean
this
things
dear
principal,
for
the
entirety
of
the
function
body
which
that
works
for
like
a
borrowed
pointer,
like
you,
know
something
tight
for
Santi,
but
it
doesn't
work
for
box
because
you
can
drop
the
Box
halfway
through
your
function.
So
what
Ralph
is
suggesting
for
this
particular
bug
is
I
believe
this
broth
is
just
we
should
get
rid
of
it
or
you
should
not
omit
that
you're
a
principal
attribute
for
the
box
type,
because
in
general
it's
not.
C
Is
this
is
where
you
then
get
into
the
things
that
LVM
is
currently
exploring
and
I?
Don't
know
if
we
have
time
to
try
to
discuss
that,
but
they
are,
they
are
exploring
the
design
space
here,
someone
suggested
to
your
principal
in
scope
originally,
and
there
was
some
pushback
against
that
concept.
That's
led
them
to
propose
other
things
that
Ralph
Ralph
is
actually
jumped
in
and
said
he
doesn't
think
their
design
as
proposed
works
for
us.
So
there's
active
discussion
here
also.
C
A
A
C
Hey
my
my
takeaway
here
is
that
it
seems
like
LVM
themselves,
are
sort
of
still
trying
to
go
out
even
what
the
semantics
they
want
to
adopt
is
in
the
first
place,
and
we
probably
do
wanna
be
part
of
that
conversation.
To
some
extent,
when
I
say
we
I
mean
the
breasts
project,
I,
don't
say
mean
the
lying
team.
The.
C
B
E
A
A
I'm
much
more
worried
about
the
ampersand
and
implications
there.
I
think
we
should
try
to
better
if
they're,
worried
and
I
try
to
focus
on
something.
I
would
be
interested
in
getting
some
time
of
Ralphs
and
focusing
on
D
referenceable.
It
seems
to
be
especially
for
the
ampersand
unsafe
cell
interactions.
If
there's
like
immediate
steps,
we
can
take.
A
E
And
in
doing
this,
I
have
a
playpen
which
shows
the
unsoundness
in
like
less
than
30
lines
of
code.
So
that's
you
just
link
to
that.
E
E
So
there
are
basically
two
possible
ways
to
close
the
soundness
hole.
One
is
to
make
to
add
unsafe
marker
traits,
like
it
imagine,
pin
safety
refuge
and
pins
everywhere,
with
pin
currently
use
d,
rekt,
mute
and
now
uses
this
unsafe
market
rate.
We
would
also
need
unsay,
hence
a
clone,
and
that
would
close
to
sound
this
whole
while
still
allowing
you
to
add
these
impuls.
The
other
approach
would
be
to
just
to
closing
this
soundness.
Hole
is
to
you
know,
make
it
so
that
these
symbols
cannot
be
written
by
somehow
reserving
you
know.
E
E
Ralph
I
don't
know
a
Ralph's
opinion
on
reserving
these
simple
suits,
but
he
does
think
that
we
should
have
market
rates
I
think
maybe
we
could
focus
on
the
question
of
like
ignoring
pin.
Should
we
preserve
these
market
rates
might
be
something
that'd
be
good
to
focus
on
because,
like
a
part
of
my
opinion
is
that
pin
was
designed
based
on
the
assumption
that
you
know
the
pointer
types
in
a
standard
library
can
be
assumed
to
behave
reasonably.
E
You
know
you
can
assume
that
shared
references
behave
like
the
way
the
share
references
are
supposed
to
behave
and
I
think
that
all
users
should
be
able
to
make
that
assumption
and
so
like.
We
should
just
make
it
so
that,
indeed,
should
references
never
implement
DRF
mute,
because
that's
like
an
assumption
that
I
think
is
reasonable
to
make
and.
E
So,
possibly
through
the
reservation
mechanism,
Nico
counted
that
it
might
not
be
possible
with
exactly
how
that
is
currently
implemented,
but
basically,
as
long
as,
if
the
reservation
it's
like
depends
on
how
the
reservation
works
with
specialization.
Basically,
the
reservation
treats
it
as
a
specialized
about
imple,
regardless
of
anything,
then
that
is
that
doesn't
work
because
you
kept.
We
can't
allow
you
to
specialize.
Do
you
have
food
for
your
local
reference?
Yes,.
A
I,
don't
believe
reservation
ever
prevents
you
from
writing
in
imple.
It
only
that's
not
obviously
that's
too
strong,
but
I,
don't
believe
it
prevents
you
from
writing.
These
Impuls
I
believe
what
it
prevents
you
from
doing
is
writing
impulse
that
assume
in
the
thing
we're
trying
to
guarantee
actually
like,
like
you,
can't
assume
that
amber
Santa
does
not
implement.
Do
you
because.
E
A
You
add
the
info
later,
because
people
might
have
this
probably
should
be
reworked.
For
this
reason,
with
the
documentation
was
getting
at
was
precisely
because
people
could
add
these
impulse
leader.
You
can't
just
add
the
reserve
employ
yourself
unless
you
can
be
sure
that
it
would
be
allowed
through
some
other
mechanism,
I
see
so.
E
Like
some
change,
the
fundamental
it's
Alston
asked,
if
add,
making
direct
mute
and
clone
fundamental
would
solve
the
problem,
which
may
be
a
woods
because
for
the
winter
actually
been
fundamental
on
traits
and
feminine
types.
That
gives
me
a
headache.
So
it's
possible
that
makes
sense,
but
I
don't
know
and
I
also
don't
know.
If
that
would
be
cooperate.
We
hoped
that
would
yeah.
B
B
A
We
should
worry
less
about
the
mechanism.
Well,
we
can
worry
about
it
and
we
should
but
I'm
interested
in
the
promise
that
was
put
forward.
That,
like
this
is
something
worth
guaranteeing
I
definitely
feel
like.
Overall
I'm,
you
know
unhappy
that
DRF
mutant,
deer,
F
can
diverge
and
I
know
why
it's
the
case
and
I
don't
know
what
I
think
it's
the
best
fix,
but
it
seems
to
be
a
lot
of
I
mean
a
common.
B
B
A
If
we
added
the
market
rates,
I
mean
I,
do
think.
We're
gonna
need
at
some
point
some
form
of
marquetry
around
the
you
that
says
this
is
well
behaved.
Well
is
a
strong
statement,
but
I
expect
us
at
some
point
to
want
to
talk
about
well
behaved
to
us
into
your
efforts
for
the
purposes
of
optimization
and
other
things.
Unfortunately,
I
don't
exactly
know
what.
E
Yeah
I
think
here
is
but
they're
pure
is
not
like
it's
a
stronger
constraint.
Yeah!
It's
a
stronger
can
strengthen
this
because,
like
you
could
mutate
something
else
in
your
D
or
F
mute
as
long
as
you
continue
return,
the
same
address
as
your
Dieruff
right
and
that
address
is
somewhere
that
it
doesn't
move
when
the
pointer
moves
and
stuff
like
that.
C
E
C
E
A
A
A
A
I
think
we
I
don't
know
how
much
more
time
under
spin,
but
I
I
would
say
that,
in
terms
of
the
mechanism,
I
think
I
have
long
hoped
we
would
add
a
form
of
explicit
negative
input
which
would
serve
as
a
way
to
promise
that
something,
basically
to
rule
out
something
can
be
implemented
or
a
promise
that
it
never
will
be.
Yeah.
C
A
Rules,
but
if,
but
you
could
imagine
saying,
like
I-
implement
not
copy
for
box
with
T
for
all
T,
it's
not,
it
would
actually
be
quite
easy
in
the
sense
of
like
we
could
make
it
do
what
we
want
to
have
this
mechanism,
but
I'm,
like
not
super
I,
think
there's
an
obvious
like
like
reservations
but
different,
because
this
is
unbounded
in
time.
The
only
fix
to
solving
that
would
be
to
you
know
somehow
address
a
negative
reasoning
in
a
more
cohesive,
coherent
way.
B
A
There's
not
I
mean
what
this
is.
This
is
limited
to.
We
can
talk
about
line,
but
it
will
be
limited
to
coherence
or
basically
be
saying.
I
can
imagine
a
pretty
limited
mechanism.
It
basically
says
I
can
declare,
but
this
is
not
implemented
and
coherence,
checks
that
all
positive
and
negative
impulse
don't
overlap.
A
I
would
or
it's
already
what
coherence
does
when
you
have
positive
and
negative
animals,
and
so
all
it
does
is
effectively
prevent
you
from
writing
impulse
which
doesn't
have
any
real
interaction
with
the
rest
of
the
language,
since
those
are
all
defined
based
on
what
impulse
exists.
So
in
that
sense
it's
fairly
targeted
but
I
think
what's
worrisome
about.
It
is
not
to
me
as
much
the
semantics
as
it
is
the
like.
What
is
the
full
rounded
feature
look
like,
and
is
this
a
good
sound
part
of
it
yeah.
E
E
A
You
know
working
out,
people
expect
it
to
write
like
for
one
thing
you
wouldn't
be
able
to
rely
on
it
in
that
would
be
you've
really
allowed
you
to
rely
on
it.
Well,
it's
still
relatively
limited
because
it's
in
coherence,
but
it
would
be
then
getting
more
in
the
direction
of
like
I.
That
starts
to
permit
code
we
didn't
used
to
permit,
whereas
the
way
I'm
talking
about
it,
I
think
we
could
pretty
clearly
say
would
only
forbid
code.
We
used
to
do
and
not
affect
the
semantics
of
good.
E
B
A
B
Well,
I
will
leave
it
nominated
now.
It's
probably
something
we
visit
anyways
unused
parens
triggers
on
macro
bite
underfoot,
so
we
recently
added
made
the
onions
parents
limbs
also
worked
for
four
types.
Tell
descriptions:
we
didn't
really
change
the
mins
than
that,
but
it
seems
like
some
don't
matter
based
most
like
relying
in
not
having
warnings.
A
There's
the
basic
raw
bean
idea
being
that,
in
this
instance,
the
heuristic
of
like
how
useful
the
likelihood
of
it
being
a
false
positive,
goes
significantly
up.
If
it's
coming
from
a
macro,
I
will
definitely
agree
that
I
have
done
exactly
this
trick
numerous
times
in
metros.
It's
a
pretty
useful
trick.
B
E
A
Think
macro
generated
code.
Anything
on
it
is
usually
pretty
good
fee.
I
guess
the
only
thing
would
be
if
you
had
a
big
like
if,
in
the
expression,
try
remember
how
the
exactly
how
we
enforce
this
I
don't
know,
but
in
other
words,
imagine
the
the
redundant
parenthesis
was
around
the
3-year
and
not
generated
by
the
macro.
Ideally,
that
would
still
trigger
yeah
I.
Think.
C
A
B
B
A
B
So
so
this,
instead
of
enforcing
so
we
move
so
what
this
PR
does.
Is
it
moves
the
restriction
on
not
having
disabilities
and
trade
items
and
even
violence
to
asthmatic
on
instead
of
a
syntactic
one
so
for
traits?
This
is
makes
it
easier
to
unify
the
item
grammars
and
for
enums
well,
gives
you
better
recovery
and
stuff,
and
it
would
facilitate
allowing
it
semantics
in
one
day,
and
it
might
also
enable
some
some,
like
macro
based
yourselves
and
such
I
can
see
that.
A
A
A
B
Basically,
we
have
a
somersault
from
2013
and
it
can
can
use
combinations
and
we
have
an
unstable
flag
to
change
which
makes
the
ping
sound,
but
has
some
progressions.
The
my
proposal
is
that
we
introduced
from
intrinsics.
We
expose
them
in
some
foreign
for
as
in
stable
and
then
we
switched
defaults
so
so
that
it's
sound
by
default
and
we
can
introduce
and
a
flag
for
maybe
eight.
A
Have
two
thoughts?
First
of
all,
does
anyone
want
more
detailed
background?
I
can
give
my
already
touch
without
going
into
what's
a
terrain,
yes,
which
I
don't
think
is
all
that
important,
but
I
will
not
die.
I
might
do
I,
say
I
have
two
questions.
Maybe
your
thoughts
one
is
that
I,
don't
I,
don't
know
if
we
need
the
unsaturated
afloat
casts,
but
the
that's
a
minor
point.
The
more
interesting
point
to
me
is:
how
do
we
make
this
decision?
I
feel
like
there's
an
RFC
here.
A
There's
at
least
some
surface
impacts
area
I
mean
the
basic
gist
of
the
proposal.
Is
we're
gonna,
add
explicit
methods
for
all
the
different
ways
you
can
all
the
different
modes
of
casting
floats
to
ins
and
we're
gonna
pick
one
for
what,
as
does
right
and
that
one
that
we
picked
will
not
always
be
the
best
choice
from
a
performance
point
of
view,
but
at
least
it's
a
consistent
choice
as
opposed
to
cease
crazy
done.
A
B
E
C
That's
well
wait.
There's
the
detail.
I
think
that
the
default
for
ad
should
be
saturating.
I.
Think
there's
about
that,
but
I
think
there's
also
an
assumption
that
we
have
to
provide
the
alternative
entry
points
and
I.
Don't
know
if
there's
consensus
about
what
all
the
alternative
entry
points
need
to
be.
D
B
B
Think
I
think
there
is
consensus
that
we,
because
people
want
the
alternatives,
the
unsaturated
and
unsay
versions
that
we
should
provide
them.
But
I
think
there
is
still
some
discussion
about
exactly
how
we
expose
the
methods
like.
Should
they
be
on
the
primitive
site
should
they
did?
Should
we
use
traits
or
something
I.
B
A
Feel
like
we
would
I
feel
like
I'm,
not
sure
I'm,
convinced
by
there's
an
old
issue
with
a
lot
of
comments
and
therefore
we
don't
need
in
order
see,
but
I
do
think
that's
going
into
why
the
RFC
could
move
relatively
quickly.
We
could
be
aggressive
about
saying.
Well,
we're
not
gonna.
Consider
too
many
variations
here.
On
the
other
hand,
I
don't
think
that
artsy's
the
place
to
work
out
this
you
guys
surface
either
I
think
that
works
better
in
a
PR
to
a
certain
extent.
E
A
C
B
A
C
A
C
There's
more
personally,
I
think
that,
for
example,
you
could
have
there's
been
some
recent
discussion
about
this
holy
I
reason.
This
isn't
blah
for
so
long
is
because
a
lot
of
people
were
proposing
using
up
some
sort
of
freeze
intrinsic
from
LLVM
to
avoid
the
undefined
behavior
you'd
have
done.
Deterministic
behavior
in
terms
of
value,
get
it'll
be
fixed
as
one
it
wouldn't
be
undef
or
whatever
it's
called,
and
so
you'd
actually
have
a
bit
pattern
that
would
be
fixed,
and
thus
you
wouldn't
get
on
to
my
behavior
from
it.
You
just
get
unknown.
C
You
know
value,
but
you
can
rewind
what
ralph
has
recently
pointed
out
when
I
was
asked.
So
why
I'm
part
of
the
reason
this
has
come
up
again,
because
I
was
the
one
who
posted
12
hours
ago.
Saying
hey
I
noticed
that
somebody
tweeted
that
LEM
does
have
fries
now.
Can
we
move
on
this
because
of
that
and
Ralph
pointed
out
there
and
elsewhere?
That
fries
is
not
ideal,
because,
even
though
it
voids
did
upon
behavior
has
information
leakage
problems,
you
know
security.
You
have
social
security
risks
I.