►
From YouTube: Lang Team Triage Meeting 2019-10-24
Description
Main topics discussed:
* `!` stabilization
B
A
A
D
D
A
C
A
C
Well,
this
is
an
interesting
one.
Yeah
I
haven't
fully
caught
up
on
this
conversation,
but
last
week
so
are
in
and
around
the
unwind
and
FFI
stuff.
There
was
a
sort
of
confusing
to
me
turn
where
there
was
this
PR
opened,
which
allows
us
plus
exceptions
to
safely
unwind
through
rest
code
or
claims
to
and
there's
some,
which
is
great,
did
a
lot
of
important
implementation
work.
C
But
one
thing
that
came
up
around
there
was
a
debate
about
whether
it
makes
sense
to
we
had
been
assuming
that
X,
R
and
C
would
mean
no
unwinding,
but
there
was
a
case
to
be
made
that
that's
the
wrong
default,
because,
basically
it's
the
opposite
of
what
she
does.
For
example,
we
accept
at
least
and
that
it
would
make
more
sense
to
support
unwinding,
which
isn't
that
hard
to
do
and
allow
people
to
opt
out
and
perhaps
like
if
you're
using
panic,
equals
abort
I,
don't
know.
C
Maybe
that
has
some
other
semantics,
but
there's
been
a
lot
of
back
and
forth.
What
was
surprising
to
me
was
that
kind
of
most
people
like
guns
of
look
and.
C
And
on
you,
and
other
people
were
arguing
more
in
this
direction,
they've
seen
the
opposite
of
what
we
had
been
saying
for
quite
a
long
time,
so
I
haven't
really
caught
up
with
it,
but
the
one
thing
I
like
to
float
out
here
is
like
I,
think
the
main
objection
to
permitting
unwinding
throughout
the
five
boundaries
is
around
dead
code
elimination.
Perhaps
there's
a
correctness
question
whether
people
expect
unwinding,
although
I
don't
know
I'm
kind
of
curious
to
take
a
quick
temperature
or.
E
C
There's
a
correctness
argument
around
there's,
tour
of
to
safety
arguments
and
I
think
you're,
making
one
of
them,
which
is
especially
for
FF
I,
which
tends
to
have
unsafe
code
in
its
surroundings.
It's
really
important
to
be
sort
of
aware
of
online
safety,
and
a
lot
of
people
may
not
expect
on
lining
across
this
boundary.
Is
that
kind
of
the
argument.
F
Yes,
but
also,
just
in
general,
like
arbitrary,
unwinding
across
rust
code,
isn't
something
that
we've
well-defined
right.
Well,
that
would,
what
do
you
mean
I
mean
we
would
be
different
like
I
would
be
the
point
defining
that
to
be
like
you,
you
would
expect
drop
hooks
to
still
run
everywhere.
Yes,
PPR.
E
C
E
C
Think
catch
unwind
has
been.
There
is
some
debate
about.
This
is
sort
of
an
orthogonal
question,
but
if
it
affects
people's
opinion,
I
think
it'll
be
important
like
either
way.
We
have
to
say
what
kitchen
line
will
do
when
the
panic
problem,
when
an
exception
goes
I,
think
he
can
be
set
up
either
way.
I
mean.
F
A
A
C
C
Catch
on
wine
does
not
catch
it.
That's
what
the
current
PR
does.
We
could
implement
that
by
as
we
do
today
by
integrating
tightly
with
the
native
mechanism
or
you
could
integrate
it
by
catching
exception,
translating
it
to
some
other
thing
propagating
it
in
your
own
way,
but
they
sort
of
not
use
your
visible.
A
F
I
mean
because
so
and
and
still
for
well
I,
guess
that's
a
question
so
on
panic
equals
the
Bohr
code
panics
from
exceptions
from
rust
code
are
clearly
not
well-defined
right.
Does
this
mean
that
we
would
still
need
to
generate
unwind
hooks
for
exceptions
coming
from
non
rust
code,
because
that
seems
I.
C
F
F
A
C
No
I
think
the
difference
would
be
that,
for
one
thing
we
don't
behave
right
now.
Our
behavior,
with
respect
to
foreign
exceptions,
is
not
very
it's
not
very
consistent
across
platforms
and
for
another
we
say
it's
undefined
behavior
right
now,
whereas
we
would
be
saying
these
we
would.
We
would
not
be
adding.
C
F
C
C
F
F
F
B
B
C
A
C
Wrong:
okay,
listen
thanks!
All
right!
Well,
I
will
go.
I
will
propagate
these
concerns
around
into
that
document
and
try
to
understand
better
what
makes
any
change,
but
I
think
that
it
seems
like
the
consensus
in
this
room
is
still
that,
if
I,
something
like
the
most
vast
majority
of
regardless
of
sees
president
the
vast
majority
of
effort
find
roles
do
not
unwind,
and
thus
we
would
prefer
to
have
it
be
that
you
explicitly
indicate
when
you
do
expect
unwinding,
then
be
a
reverse.
Oh.
F
F
Actually,
so
that's
that's
an
interesting
clarification,
though
I
want
to
make.
The
argument
here
is
over
what
today's
default
should
be,
not
whether
we
should
still
have
those
options,
there
would
still
be
a
way
to
specify
no
like
this
function
specifically
is
no
except
you
would
just
have
to
do
it
everywhere.
Oh
yeah,
that's
right!
Okay,.
F
C
I
think
I
find
this
argument
kind
of
persuasive
that,
regardless
of
I,
don't
first
of
all
I,
don't
think
C
programmers
have
a
really
strong
expectation
around
this
or
like
I,
don't
know
at
least
in
other
words,
the
argument
for
being
corollary
with
C
is
kind
of
weak,
but
that
indeed
unwinding
does
require
extra
auditing
and
it's
unusual.
So
it
makes
sense
to
call
it
out.
A
C
C
B
C
A
C
G
C
Was
a
lint
that
triggered
on
fall
back
would
be
easy.
That
would
be
way
to
probably
wait
to
course,
and
we'd
have
to
iterate
a
little
bit.
That's
a
sort
of
development
side
thing
so
I
said
well.
Do
you
have
time
like
if
I
were
to
mentor
you
if
you're
doing
that?
Would
you
be
interested
in
driving
debt
or
said
else?
Who
is
yes.
A
C
A
B
C
G
B
A
B
That's
like
it's
because
the
main
reason
to
stabilize
bang
with,
like
we've
already
had
to
work
around
the
fact
that
we
haven't
stabilized
thing:
I
introduced
the
unpalatable
type
like
sure.
So
it's
not
a
years
now
or
like
blocking
a
change.
It
has
like
really
material
who,
like
benefits,
to
like
the
API,
for
like
a
fallback
change.
B
But
because,
like
the
link
team
blocked
that
online
commit
the
fallback
change,
he
then
went
and
found
a
different
workaround
of
making
the
infallible
type
and
stabilizing
that
in
the
center
library
and
like
we've
just
been
like,
were
working
around
not
able
to
have
this
API
for
years.
And
it's
been
blocked
on
like
no
one
really
being
willing
to
implement
this
fallback
thing
and
I.
Think
it's
not
reasonable
to
keep
like
blocking
on
this.
One
I
think
I.
Think.
B
B
Definitely
think
that,
like
you,
should
drive
implementing.
If
you
want
to
fall
back
change,
you
should
diplomat
the
like
tools
necessary,
but
holding
like
that.
We
can't
move
forward
the
things
that
are
not
actually
blocked
on
the
far
back
change
doesn't
seem
like,
like
seems
like
inappropriate
and,
like
you
said
in
the
last
meeting,
that
you
thought
that
the
introducing
bang
was
not
worthwhile
had
the
father
change,
and
so
I'd
like
to
talk
more
about
like
that
few,
because
I
don't
agree
with
and
I
would
like
me.
A
F
Think
central
I
sympathize
with
that
and
I
agree
that,
like
that's
sort
of
semantics
Lee,
that's
what
I
think
is
like
I
think
that's
wrong.
I
think
that's
not
the
behavior
that
I
bought
in
the
language
right,
but
I
also
don't
think
that
it's
gonna
ruin
too
many
people's
day
and
certainly
is
gonna.
Be
far
less
can
be
like.
Let's
have
a
problem
for
those
people
to
have
a
sort
of
not
well
that
you
would
necessarily
want
fallback
to
them.
F
A
B
And
that's
what
everyone's
doing,
and
that
means
that
they
are
not
like
interconvertible
right.
It
would
be
we'll
just
use
bang
great.
So
everybody
just
has
like
this
some
other
type.
It
would
be
like
if
you
had
to
create
a
new
unit
type.
Every
time
you
wanted
to
use
unit,
because
every
unit
type
had
a
name
that
give
it
like
some
semantic,
meaning.
C
A
A
C
C
Isn't
it
is
an
infallible
one
or
is
a
visit,
falls
back
to
bein,
basically,
and
that
takes
a
question
might
as
well.
It
does
yeah
and,
and
the
reason
I'm
bringing
this
up
only
is
that
my
point
is
that
it's
not
like
this
fall
back
only
ever
affects
dead
code.
This
is
a
case
where
it
affects
life
code
granted,
weird
and
not
very
great
life
code,
and
that's
why
I
think
it's
kind
of
okay,
but
there's.
D
C
B
C
C
B
So
yeah
I
mean
but
like
Central
I,
think
they'll
like
it's
like
great
for
you
to
drive
like
making
it
the
tooling
necessary
to
make
the
fall
back
change
if
you
like
really
believe
in
it,
but
it
just
doesn't
seem
like
I.
Never
has
likely
practical
use
cases
that
we
like
would
like
immediately
and
like
blocking
on
making
an
breaking
change
that,
like
the
breaking
change,
is
blocked
on
never
existing,
but
ever
is
not
blocked
on
the
breaking
change
happening.
So
instead.
B
F
F
F
I'm,
raising
my
hand,
is
like
the
person
who
put
the
concern
on
the
original
proposal
of
stabilizing
just
being
a
fallback,
because
I
thought
that,
like
in
in
everybody's
head
that
that
was
tied,
you
the
the
fallback
issue
and
that
the
fallback
wouldn't
sort
of
get
dropped
on
the
floor,
but
I
think
central
step
up
and
say
that
you
know
you're
interested
in
implementing
the
lint
and
that's
something
that's
I'm.
Assuming
is
you're
like
you
seem
very
passionate
about
the
the
fallback
thing.
F
Far
more
than
I
ever
was
do
I
have
faith
that
you'll
actually
go
through
and,
like
you
know,
implement
something
there
and
like
try
and
have
C
motion
on
this,
and
so
I
feel
comfortable.
Saying
that,
like
like
I,
don't
think
we're
gonna
forget
about
it
and
I
think
it
would
be.
I
would
rather
have
a
stable,
never
type
than
not
have
a
stable,
never
type,
and
so
whether
or
not
Mike
yeah.
So.
A
Isn't
so
I
mean
I
mean
I
do
have
concerns
about
like
having
sleeping
there.
It's
like
no,
but
I,
don't
see
what
concerns
about
like
I
mean
if
I,
if
I,
if
I
do
they
actually
work,
I
don't
want
to
see
like
concern
on
something
that
is
not
realistic
to
implement
one.
So
you
wanna.
Let
me
go
change
my
way.
F
A
B
I
can
drive
Central
will
like
change
the
P
R
to
do
the
thing
that
I'll
just
remove
the
canary
with
their
common.
F
E
Like
that
can
I
can
I
just
confirm
I
understand
the
conversation
that
just
happened.
We
have
basically
collectively
agreed
that
we
are
willing
to
decouple
the
stabilization
of
bang
from
a
change
to
the
fallback
semantics
and
the
central
can
independent
work
on
this
Lintz
they'll
be
attached
to
a
hypothetical
change
to
fall
back,
but
that
can
happen
at
its
own
pace,
independently
of
landing.
Never
stabilization,
never
ambassador.
A
B
C
E
C
C
C
A
F
C
C
F
A
B
F
G
B
B
I
must
use
like
it's
sort
of
like
making
less
use
into
an
auto
trait
right,
where
it
sort
of
like
leaks
into,
but
like
anything
that
contains
the
most
used
is
also
must
use,
should
definitely
be
an
RFC
where
we
have
like
a
larger
discussion
about
it,
not
a
PR
and
I'm
kind
of
like
not
convinced
that
it's
still
worthwhile
change
based
on
these
results.
Yeah.
F
A
A
B
B
B
F
C
A
C
A
There's
a
small
amount
of
follows
that
came
from
this
times
executed
last
week:
yeah
I
need
to
what
did
we
land
on
him
soon
or
not?
No,
okay!
No!
Yes
I!
It
was
actually
feet,
but
I
think
we
I
didn't
feel
very
sure
about
our
rationale
for
EXCI
being
it
was
more
like
okay.
Why
not
so
I
think
that
we'll
need
to
read
it
in
detail,
but
I
can
be
done.
Amazing,
okay,.
C
A
A
That's
the
current
behavior
as
well.
So
if
we
decide
that
we're
fine
with
this,
that
it's
basically
here
CP
learnt
or
at
closing
or
something
I
guess
we
can
just
that
basically
legit,
yes
for
the,
so
that
mean
community
can
object
or
something
if
they
on
the
other
hands.
It's
also
a
conservative
pack.
Yes,.
C
C
Currently,
we
warn,
if
the
end
point
of
one
and
the
end
point
of
another
overlap.
This
was
narrowed
down
from
broader
the
thing
where
it
weren't
over
any
overlap.
And
now
the
question
is:
do
we
want
to
learn
from
partial
overload?
We
know
we
don't
want
to
work
for
subsets
because
that's
considered
useful
I
don't
really
care
it's
my
opinion.
A
C
C
C
A
A
A
C
A
C
A
B
C
Sort
of,
like
mildly
built
into
the
compiler
but
I'd
be
happy
to
have
live
scan
period,
but
I'm,
okay.
It
seems
like.
B
C
C
C
A
C
A
C
A
C
There
you
said
exactly
what
you
wanted:
I,
don't
know
to
me:
star,
expands
to
I
realize
it's
not
exact,
that
works
in
the
mechanism
but
liking
because
of
precedents,
but
it
to
ative
Lee.
It
feels
to
me
the
star
expands
to
some
number
of
names,
which
could
be
none,
and
if
it
is,
maybe
you
know,
maybe
it's
not
that's
how
it
works.
So
you
so
basically.
C
I
A
F
A
B
A
G
A
C
Basically
saying
you
could
think
of
prefix
something
as
the
I
feel
like
there's
a
you
could
think
of
it
as
a
filter,
and
so
foo
would
be
like
I
select
the
item
from
prefix
that
matches
foo
and
star
as
I
select
the
items
that
match
I
mean
any
items
and
either
way
for
all
filters.
It's
an
error
if
it
comes
to
zero,
but
I
feel
like
what
that
saying.
C
H
A
C
Apparently
so
the
consensus
is,
we
feel
like
intuitively,
it
would
be
a
warning
or
I
would
do
what
users
want,
but
if
there's
some
other
argument
like
that,
it's
helpful
for
name
resolution
and
some
other
way
that
we'd
like
to
know
yeah
I
also
think
the
backwards
compatibility
rule
applies
here
like
there's,
not
a
strong.
You
know
to
change
it
yeah.
It's
just
white
like
this.
Okay,
so.
C
B
A
C
B
C
C
B
We
real
quick
next
meeting.
Could
we
talk
about
the
turbot
fish
RC,
specifically
how
to
unlock
it.