►
From YouTube: Lang Team Triage Meeting 2019-09-05
Description
Triage meeting
Major topics covered:
- nounwind attribute and RFC 2753
- misc nominated PRs and regressions
Paper document
https://paper.dropbox.com/doc/T-Lang-Meeting-2019.09.05--AkmYatur2MDikhHZuCgppWpxAg-qA0w3GsE7WJMX3v05orXV
A
C
Yeah,
a
quick
review
on
that
based
on
last
time,
we
have
a
an
RFC
posted
and
that
would
define
unlined
attribute
in
the
simplest
possible
way
that
would
allow
us
to
move
forward.
But,
given
that
you
said
it
was
worth
the
26
that
the
beta
would
be
branching
for
the
next
release
was
that
the
day
28
for
maybe.
D
C
Centons
20
something
given
that
the
twenty-something
is
when
the
beta
would
branch
for
the
next
release:
I'm
really
skeptical
about
us.
You
know
accepting
this
RFC
adjusting
the
current
implementation
to
match
that
RFC
have
been
stabilizing
the
resolve
all
by
the
twenty-something,
so
that
we
have
a
solution
for
people
that
doesn't
invoke
undefined
behavior
I
think
it
might
make
sense
for
us
to
go
ahead
and
merge
this
BR,
which
just
says
hey.
If
we're
gonna
abort,
we
should
not
let
tho
unwind
on
things
and
then
we
can
add
no
unwind
back
when.
D
B
D
A
E
F
E
C
A
E
I
saw
comments
from
like
gnz
LBG
I
think
it
was
saying
that
they
were
opposed
to
the
like
unwind
only
attribute
what
was
I
I
I
haven't
been
following
this
close
enough
because
it
already
got
somebody
it
comments.
In
the
last
like
two
days,
somebody
who's
been
following
the
thread
more
recently
like
it'll,
like
they
can
summarize
what
what
the
opposition
is.
C
Rflp
identity,
sorry
go
ahead,
go
ahead,
Josh
fine
I
was
just
gonna
say
as
far
as
I
can
tell.
The
main
objection
is
whether
we
are
just
providing
a
minimal
solution
for
what
people
are
doing
today,
with
the
understanding
that
people
will
still
have
to,
for
instance,
build
their
C
code
with
F
exceptions,
and
that
this
only
works
at
the
rust
line
mechanism
aligns
with
the
C++
or
C
unwind
mechanism,
but
that's
what
people
have
today.
The
objection
is
basically,
oh,
no,
we
shouldn't
do
an
interim
solution.
C
It
appears
to
be
the
objection
of
know
who
we
want
to
fully
specify
how
you
handle
cross
language,
unwinding
and
specify
the
restaurant
line
mechanism
and
avoid
any
possibility
that
people
will
have
to
actually
know
what
they're
doing.
In
order
to
use
this
feature
and
to
unwind
from
rust
to
see
rust.
There
is
a.
C
A
B
C
B
C
B
E
A
A
C
C
B
D
A
E
E
B
E
B
A
A
C
Well,
whether
we
want
such
a
thing
or
not,
I
think
I
just
made
a
comment
on
RFC
27:53.
If
people
could
take
a
look
at
that,
that
effectively
says,
let's
specify
what
we
currently
assume
and
then
it
defer
to
some
future
RFC
if
any
any
mechanism
that
would
change
that
assumption.
I,
don't
want
to
change
that
assumption
right
now.
I
just
want
to
specify
the
current
assumption.
I.
A
Yeah
I
well
I
think
the
challenge
so
well
I
think
we
should
do
this
at
the
moment.
I
in
fairness
to
central
I
think
there's
a
point
that
if
we
we
are
definitely
narrowing
our
future
options,
if
we
ever
want
to
type
I
just
happen
to
think
we
don't
and
I'm
okay
with
it
or
that
if
we
do
it's,
okay,
that
the
type
of
opt
in
so
anyway,
I
agree
with
your
interesting.
It's
not
like
that's
a
free
choice.
It
sounds
like
I
think
we
need
to
move
on
from
this,
but
it's
like.
A
C
A
C
A
D
C
A
D
B
A
Okay,
let's
move
on
it's
more
minutes.
In
so
P
height
issues,
I
mean
I.
Think
we're
reasonably
good.
You
agree
at
this
okay,
P
ayisha's.
So
there's
not
much
to
say
about
this
one.
You
can
read
my
update
if
you
like,
fYI
there's,
we
been
chosen
considered
an
unstable
feature,
but
we
used
to
accept
the
odd
compute
since
they're
doing
anything
really
productive
with
the
bench
part
using
an
unstable
library.
A
A
It
was
not
a
lot
of
stuff
affected
by
enough,
so
there's
some
nominated
PRS
six
402
one
redefines
the
see
start
type
to
be
a
new
type
of
an
extra
tight,
thus
making
it
a
shallow
or
a
thin
pointer,
rather
so
that
it
could
actually
be
used
in
a
bi
functions,
but
central
points
out
that
that
sort
of
exposes,
in
some
sense
external
types,
enjoy
stable
surface.
So
the
question
at
hand
is
like:
are
we
committing
to
some
implementation
detail?
We
might
regret
by
doing
so.
A
A
D
D
D
B
A
A
D
E
What
is
the
reference
science
winter
null?
The
constructor,
the
the
function
only
works
for
size
types
today.
Yes,
it
has
a
generic
bound
great.
So
this
you,
you
already
cannot
construct
a
a
null
pointer
to
a
sister,
and
you
continue
after
this
to
not
be
able
to
construct
that
pointer.
Why
is
he
going.
A
C
C
D
C
C
E
C
B
E
C
E
A
A
A
So,
okay,
there's
a
PR
this
one.
Maybe
we
can
defer
but
I'll
cover
it
quickly.
So
currently
you've
got
a
warning
if
you
have
a
range
which
is
a
subset
of
another
range,
but
you
don't
get
a
warning
of
any
kind
if
you
have
a
range,
that
is
a
partial
overlap.
This
PR
changes
that
go
no
unreachable
patterns,
length,
which
is
sort
of
not
strictly
correct.
The
pattern
is
only
partially
unreachable.
The
question
is:
do
we
want
this
at
all?
If
so,
should
it
be
its
own
lint?
E
A
E
D
A
E
B
C
Could
find
ways
to
avoid
the
false
positive
so
as
an
as
a
note
here,
I
think
that
it
might
make
sense
to
lint
the
partial
overlap
case.
That
is
in
the
second
line
here.
If
you
have
0
to
10
and
5
to
15
linting
on
that
makes
more
sense
than
if
you
have
say,
5
to
10
and
then
0
to
20,
if
you're
completely
subsuming
the
previous
pattern.
I,
don't
think
you
want
to
lint
on
that,
because
that
is
the
potential
base
of
and
now
I'm
handling
it
all
gates.
That's.
E
C
B
B
A
F
F
Think
I
had
kind
of
objected
previously
it's
a
kind
of
having
macros
and
any
more
positions.
Basically,
because
we
weren't
sure
about
the
hygiene
situation
and
as
far
as
I
know
was
still
not
like,
we've
been
using
it
and
it
works
and
people
have
been
using
it.
No
one's
complained,
so
you
know
there's
it's
probably
ok,
but
what
I
was
saying
to
central
the
the
other
day
was
like
realistically
like
to
get
the
kind
of
understanding
that
I
want
and
I
think
others
want.
F
You
know
we're
probably
looking
at
like
an
entire
PhD
thesis
or
something
and
there's,
and
even
then
only
with,
like
you
know,
Sam
th
or
another
couple
of
people
would
like
supervised
that
so
like
I,
think
realistically,
when
we
should
admit
that
we're
never
going
to
get
the
deep
understanding
of
macro
hygiene
that
we
actually
want,
and
maybe
that's:
okay,
like
it's
macros
that
they
all
disappear
by
the
time
we've
finished
parsing.
So,
theoretically,
we
can't
cause
any
soundness
issues
but
maybe
like
best
effort.
F
E
A
F
Correct
like
it
feels
like
it
is
like
I,
don't
think,
there's
a
problem,
but
this
is
novel.
This
isn't
like
borrowing
something
from
somebody
else
has
already
done.
The
work
like
what
we
have
mostly
tried
to
do
with
with
rust.
This
is
kind
of
completely
new
and
in
an
area
where
nobody
here
has,
you
know
really
deep
expertise,
yeah
so.
A
A
A
B
B
A
E
A
B
D
A
Yeah,
the
one
thing
I
will
say
that
gives
me
a
certain
measure
of
good
feeling.
Instead,
we
transitioned
our
name
resolution
algorithm
through
several
improvements,
including
across
2018,
a
vaguely
unified,
2015,
2018
model
sort
of
successfully
and
I
suspect
that
hygiene
will
ultimately
understand,
of
course,
where
we
will
kind
of
come
to
something
better
understanding,
but
it
will
not
be
a
hundred
percent
compatible
around
all
the
edges,
but
it
will
mostly
work
we'll.
B
F
Yeah
and
I
think
like
it,
we
should
like
fighting
the
we
have.
The
mone
is
mostly
correct,
so
I
think
like
even
if
we
discover
there's
something
fundamentally
wrong
about
it,
and
we
have
to
come
up
with
like
a
new
system,
then
like
it's
still
going
to
be
just
the
edge
cases.
That
change
like
most
macro
code
will
continue
to
work.
I
think.
F
A
A
E
E
E
Don't
feel
like
I
have
a
ton
of
context
on
it,
but
I'm
happy
to
go.
Look
around
it.
What
tests
exist
and
what
you
know
what
surface
area
is
being
expand,
the
PR
itself?
Oh,
so
you
just
want
me
to
leave
a
comment
and
there's
already
a
stabilization
report.
Although
I've
done
yeah
I
guess
I
can
do
it.
Yeah
I
mean
I.
Read
your
I
read
to
report
I
thought
you
were
asking
for
more
of
something
for
me.
So
okay
I
will
do
it.
A
Let's
jump
down
to
this,
everybody
go
check,
boxes
or
John
destinations,
but
on
this
stabilizing
storyline
is
constant
Thanks,
so
so
yeah
bring
up
here.
No
no
stuff.
We
were
discussing
the
parallel
migration
and
we
actually
said
what
did
you
want?
You
can
like,
like
we
were
discussing
it.
The
release
team
had
some
discussions.
A
Bring
it
up
here,
we
would
like
it
would
be
really
nice
to
have
only
one
bar
a
compiler
one
block.
Let's
simplify
things,
we
could
watch
sound
souls
and
we
would
unblock
further
improvements,
we're
kind
of
targeting
the
idea
we
had
a
vague
target
at
the
end
of
the
year,
which
is
one
for.
Oh
that's,
coming
up
turns
out
when
you
cycle
there's
currently
six
1622
known
crates.
A
B
E
A
Okay,
so
the
point
is
there
will
definitely
there
would
definitely
be
some
impact.
Just
did
it
today.
We're
currently
planning
something
like
this.
First
of
all,
we're
gonna
target
the
most
common
root
regressions.
Most
of
them
already
have
new
versions
for
the
major
new
minor
versions
available,
but
I
think
it
looks
like
there's
a
few
left
to
go,
maybe
and
then
essentially
try
to
draw
attention
to
what's
happening.
A
B
Big
key
I
think
of
one
action
we
could
be
right
away.
Is
you
make
it
make
like
the
migration
things
are
there's
on
2018?
For
now
it
shouldn't
be
very
unlikely
for
like
if
you
have
switched
to
Edition
2008,
and
then
you
haven't
seen
the
moorings
to
do
that.
Being
like
given
fair
warning,
don't
shoot
really
clear
things
some
stuff,
but
it's
pretty
safe.
You
do
that
just
right
or.
A
We
wanted
to
make
it
easy
for
people
to
move
to
2018,
that's
right,
so
yeah,
that's
right
and
there's
a
few
options,
basically
of
what
we
can
do
sides
these
two
things,
which
is
the
one
central
just
said
we
might.
You
could
also
do
as
we
have
done
in
some
of
fine
specific
cases
like
right
now
we
just
sort
of
dip
the
ears,
but
we
can
find
cases
that
do
not
you
kind
of
make
some
of
the
errors
Partiers.
We
can
do
both
of
those
right.
We
can
do
both
of
those
and
yeah
plausibly.
A
A
E
B
We're
in
caplets
mode
so
I'm
looking
at
the
rest
crates.
Almost
all
of
them
are
like
it.
Up.Once
probably
have
no
findings
of,
and
very
nor
actually
wait.
There's
not
that
many
there
are
and
gets
up
just
just
just
throw
the
question
out
there:
how?
How
would
you
feel
about
like
making
it
all
the
harder
on
2015
right
now,
I.
A
A
A
Just
for
2018
edition,
I
think
doing
139
2018
140
2015
feels
pretty
ok
and
it
lets
us
make
an
announcement
give
people
a
little
time.
It
does
make
a
difference,
though,
that
that
these
regressions
are
primarily
like.
It
would
be
good
to
note
the
number
of
things
there
aren't
crazy
over
society
blows.
Give
me
those
because
I
care
a
lot
more
about
I.
Think
people
will
fix
their
own
gear,
because,
if
they're
maintained
but
create
IO
can
be
opinion,
just
lingers.
E
B
A
A
Okay,
I'm
gonna
run
through
a
few
small
things
very
few
more
like
FY
eyes.
One
of
them
is.
We
got
some
regressions
recently
because
we
tweaked
how
the
pretty
printer
works
and
people
were
pretty
printing
tokens
and
then
relying
on
details
like
is
there
a
space
before
or
after
comma
and
so
forth,
and
didn't
affect
too
much
good?
We
mostly
if
it
was
a
really
small
thing
and
it's
reasonable.
A
E
B
A
A
There's
similarly,
some
regressions
around
but
they'll
kind
of
isolated,
around
tweeks
resolution
and
hygiene
to
earlier
topic,
mostly
in
the
what
appears
to
be
sort
of
bug,
fixes
like
to
look
at
this
one
here
using
debug
and
derive
debug
there's
a
kind
of
cycle.
Where
is
this
debug
like
this
refers
to
some
macro
bots
in
scope?
Well,
here's
something
that's
in
scope!