►
From YouTube: 2020-07-27 C/C++ SIG
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
A
Hello,
are
you
on
call
again
this
week
or
is
it
more
relaxed.
B
B
B
B
B
How
many
folks
do
we
have
hi
everyone
hi?
I
think
we
have
enough
karum
here
hi
good
afternoon.
I
think
we
can.
We
can
start
so
same
thing.
We
can
have
each
team
updating
their
progress
and
see
any
like
blocking
issue
any
dependency
or
anything
that
you
need
help
from
the
maintainers
or
approvers
here.
D
Yeah,
so
nothing
blocking
me
there's
a
mock
explorer
that
was
used
in
a
couple
different
test
files
that
I'm
moving
to
a
separate
class
and
we
renamed
that
actually
to
in
memory
span
exporter
and
I'm
also
starting
optimization
on
the
samplers.
After
that's
done,
that's
pretty
much
all
the
work
we
have
for
samplers.
So
I
have
two
more
weeks
of
my
internship
that
I'll
be
able
to
help
on
a
different
project
if
needed.
D
So
if
anyone
will
be
needing
help
at
that
point,
I'll
be
free
for
starting,
probably
next.
D
D
It's
actually
three
weeks
this
week
and
then
two
more
so
I'll
have
two
extra
weeks
standing
on
the
14th
just
to
clarify.
E
Okay,
so
listed
here,
I
have
a
blocking
issue
on
the
meter
class
and
I
mentioned
it
and
the
pr
I
just
raised
and
updated
the
meter
api
pr
to
mention
this
as
well.
I
posted
an
issue
about
it,
but
essentially
to
summarize
here,
no
standard
shared
pointer
is
missing.
E
Some
functionality
that
we
actually
need,
which
is
to
be
able
to
use
dynamic,
pointer,
cast
and
static
pointer,
cast
the
reason
that
we
need
something
like
this
is
because
we
can't
just
because
we
need
to
be
able
to
like
cast
from
an
api
component
to
an
sdk
component
to
access
functions
that
are
only
defined
in
the
sdk
functions
related
to
collecting
and
exporting
metric
data,
and
we
can't
just
do
like
a
regular
static,
cast
or
dynamic
cast
by
getting
the
raw
pointer
under
all
situations,
because,
for
example,
in
the
meter
class,
when
a
user
creates
a
metric
instrument,
they
are
returned.
E
A
shared
pointer
to
that
instrument,
as
well
as
a
shared
pointer,
also
gets
put
in
a
map,
that's
maintained
by
the
meter
class.
So
if
we
were
to
get
the
raw
pointer
and
do
the
cast
there,
then
when
the
user
shared
pointer
goes
out
of
scope,
then
that
memory
gets
deleted
in
the
next
collect
call
we'll
call
as
a
seg
fault
yeah.
E
So
I
I
expounded
more
on
some
of
the
solutions
in
the
issue
that
I
posted
there,
but
for
those
of
you
who
haven't
had
a
chance
to
look
at
it,
yet
as
it
was
just
posted,
I
think
that
the
best
solution
would
probably
be
to
implement
a
dynamic,
pointer,
cast
and
static
pointer
cast
for
no
standard
shared
pointer,
and
this
is
the
solution
that
tom
tan
on
your
team
gave
us
riley
and
we
looked
into
implementing
this
ourselves,
but
it
seems
that
the
no
standard
shared
pointer
is
missing.
E
I
believe
it's
called
the
aliasing
constructor.
It's
missing
some
constructor
that's
necessary
to
to
implement
this
functionality,
so
we're
a
bit
out
of
our
depth
here
and
I
was
hoping
that
it
was
possible
if
somebody
else
could
take
a
look
at
the
issue
that
I
posted
and
possibly
see
about
implementing
that
or
another
solution.
B
Okay,
I
I
I
have
received
your
email
and
sorry
for
my
latest
response.
So
so
my
understanding
is
like
you
believe
what
tom
proposed
should
be
able
to
solve
your
problem,
but
you
got
stuck
because
you
have
some
other
dependencies
and-
and
you
don't
have
enough
knowledge-
how
to
unblock
this
work
and
what
you
expect
is
like
come
again
here
to
offer
more
help
and
unblock
you
right.
E
Yeah,
essentially
that
the
solution
that
he
gave
us
it
seems
to
work,
especially
because,
if
I
replace
in
the
meter
class
no
standard
sharepointer
with
standard
sharepointer,
all
tests
pass
and
it
builds
and
there's
no
seg
faulting
or
any
other
problems.
So
it
seems
that
we
just
need
that
functionality
that
tom
mentioned
there
and
I'm
not
familiar
enough
with
like
stl-s
code.
So
I'm
not
sure
how
to
implement
it
myself
for
who
could
possibly
help
with
that
on
the
project.
B
Right,
let's
see
so,
my
first
recommendation
would
be
take
a
look
at
how
stl
has
implemented
this,
and
my
take
is
like
tempted
to
call
this
is
complex.
E
B
A
Yeah
so
my
progress,
there
are
a
couple
prs
that
are
ready
for
reviews,
so
these
are
related
to
metric
instruments
and
then
a
couple
also
for
the
aggregators.
I've
listed
them
all
out
here
and
right
now,
I'm
working
on
the
yeah.
That's
that's
one
of
them.
A
A
B
Okay,
so
for
this
one
I
think
the
general
feedback
is
the
isdk
should
throw
exception
if
the
user
specified
that
they
want
that.
So,
in
the
compilation,
I
believe
there
are
two
different
flavors
passed
to
the
sdk.
One
is
like
don't
throw
any
exception
at
all
another
one
is
you?
Can
you
can
give
me
exception
and
in
the
first
case
where,
instead
of
sending
exception,
we
should
give
error
code
or
we
should
just
kill
the
application.
B
I
believe,
if
you
search
the
code,
you
should
see
some
some
places
where
we
figure
their
exception
happen,
but
we
don't
allow
people
to
throw
exception.
This
way
we
just
call
terminate
if
you
search
through
the
code
base.
I
believe
you
can
find
that.
A
B
Yeah
cool
and
in
this
particular
case
I
I
think
it
makes
sense
to
have
the
like
if
deaths.
So
during
compile
time
you
can
determine
if
the
user
is
compiling
your
code,
where
they
won't
accept
it
or
not,
and
based
on
that,
you
throw
exception
or
you
just
kill
the
application.
C
Okay
for
the
ocean
metric
exporter,
I
just
made
the
pr
from
draft
to
ready
for
review
that
pr
has
a
lot
of
aggregator
code
from
the
other
pr,
so
that
should
not
be
reviewed
there,
but
it
relies
on
that
to
pass
tests
and
actually
work,
and
then
the
processor
for
metrics
is
in
development,
and
it's
almost
done
I'm
aiming
to
pr
by
tomorrow-
and
I
just
have
a
quick
question
like
in
in
the
metrics
pipeline-
we're
only
really
passing
around
aggregators
as
no
standard
share
pointers,
but
the
aggregators
can
only
merge
as
raw
aggregator
pointers
and
in
order
to
keep
state
in
the
processor,
I
need
to
merge
aggregators
whenever
it
sees
a
an
instrument.
C
A
label
set
pair
that
it
has
already
seen.
And
how
should
I
handle
this?
Should
the
aggregators
just
merge
with
no
no
centered
share
pointers,
or
should
I
cast
down
to
merge.
B
C
Yeah,
so
from
other
I
was
looking
at
python's
implementation
and
the
aggregators
stored
in
the
processor
should
not
be
the
ones
handled
by
the
pipeline.
They
sh,
which,
whenever
you
see
a
unique
instrument
label,
set
pair
it'll,
create
a
new
aggregator
and
store
that
one
for
for
keeping
state.
C
So
I'm
not
sure
if
those
need
to
be
reference
counted.
B
Yeah,
I
think
the
brush
where
python
is
taking
should
work
okay,
and
for
for
this,
like
super
fast
project,
I
I
think
we're
a
little
bit
different
where
a
lot
of
other
sdks
they
don't
allow
sharing
of
some
of
the
like
exporters
or
processors,
for
example.
B
If
you
have
multiple
providers
they're
not
supposed
to
share
the
actual
trace
processor
and
they
have
multiple
processors
they're
not
supposed
to
share
the
same
mix
polar,
but
in
this
project
I've
seen
like
like
we
use
a
lot
of
like
like
reference
company
for
sharing
components,
and
I
want
to
know
like.
Is
there
any
feedback
from
johannes?
B
Yeah,
I
think,
like
literally
for
this
class
pass
product,
we're
we're
slightly
different
from
the
spec,
where
the
spec
is
saying
like
exposure
is
referenced
by
the
basically
owned
by
the
processor
and
processor
is
owned
by
the
trace
provider,
but
here
I
think
we're
not
taking
that
ownership.
Instead,
we
share
the
reference
counting.
G
G
Is
like
okay,
one
exporter
can
only
be
used
by
one
processor.
I
see
and
one
processor
can
be
used
by
different
tracers,
so
the
tracer
I
see
the
processor
so
that
that
relationship
is
one
too
many,
so
the
same
processor
can
be
shared
by
several
tracers.
G
B
C
Yeah
and
I
think,
if
the
dynamic
casting
for
sure
no
centered
share,
pointers
added
that
might
problem
be
solved
as
well.
So
we
don't
have
memory
management
issues.
B
H
H
I
think
so.
I
just
need
one
final,
like
a
final
pair
of
certifies
on
this
so
yeah
once
this
is
merged
and
I
can
like
submit
another
pr
just
for
the
example
that's
ready
too
and
other
than
that
I've
been
looking
for
way
more
ways
to
contribute,
because
after
this
it
seems
like
the
backend
libraries
are
what
are
supposed
to
be
done
for
now,
like
like
spam
context,
maybe
or
the
metrics
ap
guy,
and
I
was
just
wanting
to
get
your
opinion
on
what
further
ways
I
could.
B
Okay,
probably
you
can
you
can
post
a
message
in
the
the
overall
open,
telemetry
c,
plus
plus
guitar
channel,
and
see
if
the
other
work
students
will
need
your
help?
Okay,
I
think
initially,
I
was
thinking
that
the
context
part
might
need
some
help,
but
I
think,
as
long
as
like
sam
is
getting
the
first
like
the
context
through
local
storage
integration
merged,
a
part
probably
should
be
okay,
so
if
the
other
part
will
need
help,
we
can
figure
out.
G
Thank
you
just
one
note
here.
I
think
if
there
is,
if
it's
hard
at
this
stage,
to
get
going
with
each
of
the
other
teams
or
if
there's
there
are
not
much
work
left,
I
mean
we
have
lots
of
issues
there.
So
there's
there's
also
the
possibility
just
to
pick
up
any
of
the
issues
that
seem
fitting
for
you
got
it
okay.
It
will
also
be
a
great
help
to
just
clean
up
our
issue.
Backlog,
yeah,
all
right.
B
Yeah
and
I
I
need
some
help
so
who
can
who
can
help
to
reveal
this
pr?
I
I
got
many
pr's
from
this
repo
and
other
reports.
Probably
I
don't
have
enough
energy
to
look
at
all
of
them.
I
think
for
matrix.
Currently,
I'm
I'm
involved
a
lot
so
I'll
I'll
be
able
to
handle
all
the
matrix
pr
and
the
context
pr,
but
for
all
the
prs.
I
think
that's
probably
like
too
many
for
me
and
I
need
help.
G
I
Okay,
so
for
me
last
week
I
was
working
on
updating
the
recordable
interface
and
implementing
links
and
event
attributes
in
the
otlp
recordable,
and
now
I'm
continuing
looking
for
more
work.
So
I
have
five
weeks
left
in
my
internship,
and
so
I
talked
to
sam
and
tian
lin
who
are
working
on
the
context
and
they
suggested
working
on
merging
the
trace
state
pr
that
emil
had
a
few
months
ago-
and
I
have
a
couple
questions
about
that.
I
Yes,
okay
and
then
the
so
I
talked
to
emil
about
this
and
he
said
that
the
reason
that
this
pr
got
put
on
hold
is
due
to
a
design
issue
so
currently
he's
using
no
std
string
view,
which
is
a
non-owning,
and
he
was
saying
that
he
couldn't
really
figure
out
a
way
to
use
an
owning
data
structure.
I
So
he
suggested
like
we
could
have
no
no
std
string,
but
then
that
doesn't
currently
exist
and
that
might
lead
us
down
the
path
of
re-implementing
a
lot
of
stl
stuff
and
then
there's
also
no
std
unique
pointer
to
a
char
array.
I
But
he
basically
said
that
all
of
the
solutions
he
came
up
with
were
not
that
good.
So
I'm
not
sure
if
anyone
has
any
thoughts
on
how
to
work
around
that.
G
G
I
B
I
Okay,
awesome,
okay,
so
I'll
be
looking
at
that
this
week
and
also
like
I
could
I'm
waiting
on
this
like
specs
pr
for
the
otlp
exporters
and
once
that
merges,
I
can
start
working
on
updating
the
rotlp
exporter
to
conform
to
those
specs.
J
And
okay,
so
I
just
have
a
couple
of
bits
left
to
fix
for
the
runtime
at
the
local
context
and
I'll
get
that
done
tonight
and
then
hopefully
we
can
merge
that
and
then,
after
those
fixes,
I
have
a
pr
ready
to
go
with
some
static
methods.
To
add
to
the
runtime
context.
B
Okay
for
for
the
helper
functions,
may
I
ask
you
to
add
some
like
comments
to
give
example
like
you,
you've
got
the
open
census
and
open
telemetry
context
api.
You
can
see
some
like
very
simple
examples
there.
So
in
a
class
project
project,
it
would
be
helpful
to
give
those
examples,
because
you
want
to
open
it
with
a
question
about
how
do
you
propagate
context
across
different
explicit
threads,
so
mention
mentioning
something
there
be
helpful
for
people
in
the
future.
J
So
but
it
should
having
the
ability
to
propagate
the
context
across
the
threads
be
included
with
that
pr.
B
B
Yeah,
because
those
additional
helper
measures
are
not
required
by
the
spec,
so
people
might
have
no
idea-
and
I
think
the
most
simple
case
would
be
important
to
take
a
look
at
like
open
census,
python
or
open
telemetry
python.
So
this
is
basically
mimicking
the
behavior
and
that
should,
but
if
we
have
some
better
like
way
of
doing,
I
would.
B
J
Sure
going
forwards,
should
we
add
a
like
johannes,
was
asking
a
way
to
propagate
these
the
context
across
threads.
Would
that
be
a
good
thing
to
work
on
next.
B
I
think
that's
a
nice
to
have
thing,
but
there's
no
dependency.
So
it's
up
to
you.
If
you
figure
you
want,
you
want
to
continue
on
that
or
you
want
to
work
on
something
related,
but
not
specific
with
this
problem.
Sure,
okay,
another
other,
like
context
management.
Besides
tls,
there
are
like
fiber
local
storage
and
cooperatives
like
coroutine
and
where
people
can
have
those
callbacks.
So
is
this
something
I
want
to
explore.
That
part
is
more
challenging,
but
it's
out
to
you.
J
Well,
we
can
discuss
it
once
we're
there.
B
K
Okay,
so
last
week's
for
last
week
I
finally
got
the
http
trace
context
to
work
and
they
passed
some
basic
testing
local
testing,
but
not
with
the
conformance
tests.
Yet,
and
so
that's
the
good
news-
and
there
are
some
downside
of
this-
and
because
some
of
the
dependency
issues
I
have
made
a
couple
of
compromises
and
workarounds
which
are
detailed
below
so
so.
K
First
of
all,
so,
first
of
all,
I
used
the
standard
string
in
within
the
http
trace
context
for
some
string
concatenations,
I'm
not
sure
if
that's
okay,
to
do
because
I'm
not
really
returning
it
outside
of
the
file,
but
just
inside
of
one
specific
function
and
never
going
outside.
I
don't
know
if
that's
okay
to
do,
and
secondly,
the
tree
state
is
not
it's
not
fully
implemented.
Yet
from
what
I
see
on
the
master
branch,
so
I
just
used
a
standard
map.
K
K
So
that's
the
dependency
issues
and
also
another
thing
is
that
it
seems
that
the
span
is
no
longer
being
accepted
by
context
as
a
input
as
a
value
input
and
I'm
not
sure
if
that's
intended.
But
now.
K
B
G
In
a
local
variable
should
be
fine
as
long
as
it's
not
there.
It's
a
member
of
any
of
the
classes
that
we
pass
across
api
like
across
binary
boundaries.
That
should
be
fine.
B
Yeah
so
quick
question:
what
about
allocation
like
if
there's
a
location,
failure?
What
kind
of
exception
do
you
expect
to
get,
or
you
always
use
no
throw.
K
There
were
some
exceptions
being
thrown
and
but
I
used
a
like,
I
handled
those
exceptions.
I.
G
Working
yeah
also
as
a
note
here,
we
say
we
throw
in
exceptions,
but
still
that
is
not
true
for
bad
and
lock
exceptions,
so
we
don't
add
like
a
try
except
head
around
every
new
call
that
we
have
so
our
code
might
still
throw
by
dialog
exception.
So
that's
that's
fine!
Actually,
if
that
is
if
that
is
the
question
here,.
B
B
The
third
one
is
not
clear
what
actually
that
is
so
probably
like
tag
us
like
tag
me
and
walnuts
on
the
actual,
like
pr
code,
explain
what's
the
problem
or
if,
like
only
going
to
take
a
few
minutes
and
expand
here.
K
So
it's
actually
just
one
one
method:
it's
called
create
from
remote
parent
and
if
you
look
at
the
code
in
in
master
branch
on
the
span
context,
you
will
find
there
is
a
to
do.
K
B
Okay,
so
you
have
a
simplified
version
to
create
that,
and
I
I
think,
as
part
of
this,
like
project
you're
supposed
to
create
the
actual
implementation.
This
is
part
of
the
context
propagation.
So,
basically,
you
receive
something
from
the
http
incoming
headers
and
you
need
to
create
a
spam.
Connects.
K
B
I
think
that
update
probably
the
overloaded
functions,
I'm
not
sure,
but
if
you
have
the
pr
then
like
you
can
tag
my
name
there
I'll
take
a
look.
B
B
Yeah,
I
I
think
it
should
be
just
like
one
one
line
of
code:
changing
the
variant
is
that
correct,
yeah?
Okay,
so
in
this
way
like
telling,
I
I
think
it's
like
a
variant
that
we
decided
people
can
add
stuff
when
they
need
and
that.
B
M
M
So
in
the
past
week
we
got
tracy
working
end
to
end
pretty
much.
We
just
are
waiting
on
a
couple
pr's
to
get
merged
and
maybe
there's
like
some
changes
we
need
to
add.
M
We
got
like
other
parts
working
and
then
we'll
start
on
cleaning
those
up
and
adding
more
documentation
before
we
start
the
prs
for
that
yeah.
So
a
couple
blockers,
just
mostly
the
http
server.
We
need
to
clean
that
up
and
then
get
that
that
testing
server
merged
into
the
open,
telemetry
repo.
M
And
then
we
just
had
like
a
couple
questions
about
some
of
the
different
things
we're
using
that
make
it
kind
of
awkward
to
add
to
the
single
sales
people
like
the
since
we're
using
javascript.
If
there's
any
standards
and
style
or
testing
we
need
or
if
we
needed
to
show
the
events
for
span
data.
B
B
B
Yeah,
I
I
think
like
like
either
they
belong
to
a
separate
repo
and
we
can
take
that
as
a
sub-module
during
build,
but
that
still
seems
too
much
or
we
can
make
that
a
separate
package
and
publish
that
to
something
like
npm,
which
is
a
standard
way
of
doing
like
those
things.
B
And
then,
when
we
build
this
z
pages,
we
have
a
custom
like
build
stack
where
we
take
the
package
from
npm,
expand
that
and
incorporate
that
into
the
project,
link
that
into
the
actual
hdb
response
and
on
the
other
side,
if,
if
you're,
like
eventually
vision,
this
is
not
a
local
tool,
you
don't
need
everything
to
be
packed
into
the
binary.
You
can
also
think
about
like
cdn,
but
that's
probably
too
much
not
a
conversation
today.
L
Yeah,
it's
just
like.
I
think
that
the
whole
thing
of
how
to
properly
package
it
is
probably
a
little
bit
beyond
the
scope
of
the
internal
project,
like
I
agree
with
what
you're
saying,
but
perhaps
for
the
guys
working
with
it
right
now.
Maybe
if
we
can
add
an
exemption
that
this
code,
like
javascript
code,
requires
this
license,
but
it
may
not
like,
like
it's
an
optional
component,
that
somebody
who
finds
that
this
is
not
a
friendly
license
for
their
development
may
just
not
even
build
with
that
in
a
way
like.
B
To
look
at
the
license
if
it
is
compatible
with
a
patch
2.0,
I
think
it's
fine
to
leave
it
now.
We
need
to
have
a
an
issue
to
track
like
when
are
we
going
to
clean
this
up,
and
I
think,
if
we're
okay
with
that,
I
think
it's
fine
to
just
leave
the
file
instead
of
having
a
sophisticated
coverage.
I
think
like
just
to
have
this
temporary
solution
to
leave
all
the
files
for
this
internship
project
and
build
all
the
ci
on
top
of
that
seems
overkill.
For
me,.
L
Do
you
know
if
sergey
is
writing
overall
initiative
for
z
pages
across
the
languages?
I.
B
Don't
know
I
actually
have
this
question
for
openclimate.net,
so
we're
releasing
open
telemetry
done
at
beta,
and
I
have
a
thing
called
open,
telemetry,
the
pages
exporter
where,
in
this
repo
we
have
discussed
that
we
shouldn't
call
that
exposure.
So
I
want
to
know
if
there's
like
overall
decision,
I
think
yes
back
people
decided
it
is
not
the
next
quarter
right,
but
who's
in
charge
of
the
overall.
Like
decision
of
the
invaders.
F
I
think
sergey
is
driving
the
sea
pages.
Currently
it's
more
of
a
project
in
different
languages.
There
is
another
initiative
in
the
java
open,
telemetry
package
and
the
other
one
is
currency,
plus
plus,
and
I
think,
based
on
those
results,
there
will
be
some
progress
of
maybe
more
holistic
solution,
but
I'm
not
sure.
B
Okay,
so
did
you
follow
up
with
sergey
to
get
two
things?
One
is:
is
there
a
way
that
different
language
projects
can
share
the
same
set
of
the
javascript
and
style
sheet,
bring
more
consistency
and
find
another
place?
So
we
don't
have
to
check
in
similar
code
in
different
places
and
the
second
one
is
like:
have
we
decided
to
call
that
expeller
or
something
else,
and
I
think
we
have
decided
here?
It
is
not
expoller
and
like
yeah.
F
F
I'm
not
sure
if
a
central
place
for
the
javascript
will
be
ready
by
end
of
internship.
What
would
you
recommend
for
doing
in
this
case.
B
F
Cool,
I
agree
I'll
double
check
with
sir
again
get
back
to
you.
Okay,.
F
B
Download
permit
this
exponent.
N
N
B
So
my
understanding,
I
I've,
marked
two
prses
right.
Please
merge,
I
believe.
B
It
should
be
good
enough
for
this
one
sam,
I
remember
I
mentioned
there's
some
small
issue
that
you
plan
to
work
later
today
and
after
that
you
should
be
ready
to
merge
right.
G
B
Yeah
so
sam,
can
you
still
hear
us.
B
We
can
probably
move
on
any
other
prs
that
you
believe
is
ready
for
merge.
F
G
General
remark
here:
the
prs
we
saw
that
I
think
there
was
a
post
on
the
github
channel
that
the
pr
builder
is
failing
currently
with
some
built-in
windows,
and
what
we
see
is
that,
like
vc
package,
fails
to
fetch
the
protobuf
stuff-
and
I
wonder
if
maybe
max-
could
have
a
look
at
this.
L
Is
this
the
circle
ci
issue
or
the
the
the
other
one
I
think.
Last
week
we
discussed
the
lib
event
and
the
event
I
tried
with
vc
package
just
by
default.
It
compiles
and
built
successfully.
So
that
was
blocking,
I
think,
ryan's
pr,
but
for
this
one
I'm
wondering
if
we
should
just
remove
circle
ci.
Can
we
because.
B
L
I
Yeah
I
shared
the
github
actions
link
as
well
in
the
chat.
L
Yes,
I
have
one
topic
that
I
wanted
to
discuss
with
you
guys
I
just
loved
an
issue
githubish
it's
about
visual
studio,
2015
support.
Let
me
provide
some
context
here:
I'm
actually
working
with
bing
bing.com
right
now
to
integrate
open
telemeter
for
atw
atwm
windows
right
now,
I'm
just
taking
the
api
headers
and
implementing
atw
sdk
purely
based
on
the
api
and
seems
like
for
now
for
the
next
few
months.
At
least.
L
So
I
was
wondering
if
we
should
add
a
ci
loop
now
I
realize
that
maybe
it's
a
temporary
investment
like
in
a
year
from
now
most
customers
may
not
need
these.
But
my
argument
here
is
that,
since
we
already
invest
in
like
things
like
gcc
for
supporting
linux
linux
and
we
keep
compiling
it
and
we
keep
patching
things
for
gcc,
for
we
might
as
well
support
visual
studio
15
because
it
has
c
plus
plus
11
support.
L
Things
seem
to
work
with
few
minor
patches
and
maybe
we
should
add
a
ci
loop
to
make
sure
that
any
new
code
we
add
also
compose
with
that
it
does
come
at
a
certain
price
because
sometimes
there
might
be
some,
you
know
legacy
quirks.
L
G
L
L
B
L
We
already
have
loops
for
17,
I
think
17
and
19.
By
default.
There
are,
there
are
environments,
only
productions,
15
is
more
like
maybe
a
custom,
and
maybe
only
cmake
would
be
sufficient,
because
I
wanted
to
initially
add
some
make,
I'm
not
an
expert
in
bazel
at
this
point,
but
functionality-wise
we
can
verify
that
the
compiler
itself
and
all
the
headers
are,
you
know
operational
and
to
avoid
regressions.
L
Maybe
we
can
set
up
a
timeline
like
we
can
phrase
it
out
in
a
year
from
now,
but
for
now
like
for
today,
my
customers
would
need
it.
G
So
it's
gonna.
Yes,
first
of
interest
is
microsoft.
What's
the
deprecation
part
police
for
visual
studio
versions,
like
I
guess,
2015.
L
It's
a
great
question,
so
in
general,
all
security
patches
are
going
to
be
in
17
and
there
are
17
and
above,
like
visual
studio,
17
visual
studio
19,
and
there
are
other
reasons,
practical
reasons
why
most
customers
should
be
moving
on.
L
L
Usually
all
of
this
was
supported
in
13
and
15.
2012
server
radiation
is
windows.
8.1
we'd
need
a
few
years,
maybe
two
years.
B
When
I
take
a
step
back
and
ask
what's
the
general
support
policy
we
have
so
in
donette,
the
story
was
quite
messy
previously,
like
you
need
to
draw
like
several
big
tables
to
extend
to
people
like
which
version
which
business
you
support
and
later
we
change
that
to
a
very
simple
message,
whatever
officially
supported
by
the
dotnet
team
for
the
donated
time,
it
will
be
supported
by
open
telemetry
and
we
don't
have
to
explain
that
messy
story
over
and
over
again,
so
for
c
plus
plus.
L
I
think
we
were
saying
c,
plus,
plus
11
and
above
right,
like
as
a
genetic
general
message.
That's
why
we
backported
all
these
noise
3d
classes
also
and
visual
studio.
2015
is
the
first
one
that
has
decent
near
complete
support
of
c
plus
plus
11,
and
it
should
work
with
like
98
confidence
should
work,
and
then
maybe
we
just
need
to
be
careful
in
terms
of
might
be
quirks
when
new
language
feature
is
added,
might
not
be
exactly
treated
well
by
an
older
compiler.
We
can
set
up
some
duplication
policy.
L
B
Okay,
so
having
to
see
a
pr,
my
suggestion
would
be
when
you
send
the
pr.
It
would
be
great
if
you
can
clarify
like
the
support
policy.
L
B
G
That
would
be
great
having
kind
of
a
deprecation
policy.
I
mean
the
same
for
like
lineup
stuff
like
365.
Yes,
yes,
yes,.