►
From YouTube: 2020-08-24 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).
B
A
B
A
A
Before
so,
we'll
give
a
like
folks
couple
minutes
to
to
put
their
update
there,
and
if
you
have
time,
please
take
your
time
and
sign
your
name
here.
A
I
think,
as
we're
like
close
to
the
the
summer,
like
seasons,
probably
like
well,
like
interns
will
start
to
go
back
to
school
and
we'll
see
less
activities
here,
and
I
heard
from
johannes
and
also
like
from
microsoft,
work
we're
going
to
add
more
dedicated,
like
full-time
employees
on
this
product.
So
I
think,
like
the
work
we
have
done
here
during
the
summer,
internship
has
set
a
good
starting
point.
A
And
it
seems
like
we're
getting
close
to
get
the
initial
like
tracing
part,
walk
into
it.
Now
we
have
the
context
part
we
have
the
like,
those
trace
id
span,
ids
generated,
and
we
have
the
parent
child
relationship
connected
and
now
we're
getting
the
across
wire.
Following
the
trace
contacts,
it
seems,
like
all
the
pieces
are
coming.
D
D
We
get
your
advice
of
making
a
container
to
do
the
test.
We
actually
plan
to
do
it
later.
D
What
eric
and
I
were
doing
this
afternoon
was
we
were
trying
to
add
another
option
of
ci
job
for
prometheus
test,
because
only
only
our
job
depends
on
the
promises,
client,
and
I
also
added
some
options
to
the
cmake
list
file
to
turn
on
and
off,
building
the
prometheus
exporter
and
the
promises
client.
D
I
just
got
the
idea
from
what
we
have
for
the
otlp
protocol
test,
so
this
method
actually
works.
It
works
locally,
but
when
we
add
this
this
job
as
a
subtask
under
the
cmake
test
job
and
trying
to
install
the
promises
client,
we
got
some
permission
problems
and
we
were
trying
to
figure
this
out.
If
this
method
didn't
work,
we
will
try
to
solve
the
problem
with
the
container
tonight.
Yeah.
A
Yeah-
and
I
would
suggest,
go
with
the
container
approach
at
the
beginning,
like
that's
what
the
other
six
are
doing
like
in
python
and
donna
they're
all
using
the
containerized
approach.
This
gives
you
better
way
to
reproduce
that
on
your
local
machine
and
also
it's
in
general,
the
normal
way
that
that
people
would
do.
D
A
Yeah,
so
so
my
suggestion
like
try
to
get
like
pr
clean,
like
you
see
like
I,
I
send
you
the
instruction
like
how
how
do
you
start
to
receive
active
reviews?
You
need
to
get
the
ca
clean,
otherwise,
after
review
you're
going
to
make
significant
change
and
that
will
require
another
full
run
of
review
and
that's
like
big
waste
all
the
time
here
so
make
sure
you
got
the
basic
things
running
and
if
you
figure
this
is
like
the
integration
test.
Probably
is
something
too
much
for
this
pr.
A
C
Hi
so
last
week
I
was
working
on
setting
the
the
span
and
trace
ids
and
I
basically
had
to
work
around
the
fact
that
we
don't
have
a
span
context
added
to
a
span
yet
but
it'll
be
added,
I
think
by
tan
lin's
pr
with
the
http
trace
context,
and
then
once
that's
done,
we
can
actually
properly
populate
the
parent
span
id
and
the
trace
id
using
the
parent
span
context,
and
I
have
an
issue
for
that
as
well.
I
commented
on
the
issue
for
setting
ids
and
then
other
than
that.
C
So
this
week
is
my
last
week
and
so
I'm
going
to
be
looking
at
wrapping
up
some
things,
so
one
of
them
would
be
like
for
the
exporter
configs,
so
the
actual
spec
was
merged,
so
it's
all
defined
now,
and
so
basically,
I
think
what
I'm
gonna
do
is
add,
like
a
config
struct
to
the
exporter
like
this
is
something
that
was
used.
C
I
think
in
a
lot
of
the
other
languages
and
also
an
open
census,
they
have
they
just
define
like
a
struct
that
can
be
passed
into
the
exporter
constructor,
and
then
it
can
have
things
like
the
endpoint
and
the
timeout.
But
then
one
question
I
had
for
that
is.
I
was
speaking
with
my
host
ilya
and
we
were
wondering
what
the
best
way
would
be
for
someone
to
set
those
options
outside
of
the
binary.
So
would
it
be
like
environment
variables
or
like
what?
C
A
Yeah,
so
so
in
general,
in
open
time
tree
the
way
people
think
about
this
is
they
want
to
expose
the
configuration
through
an
interface
and
the
interface
has
abstract.
In
that
way,
people
have
different
ways
of
doing
configuration.
For
example,
some
people
would
prefer
environment
variables.
Some
people
think
this
is
a
security
concern.
They
want
to
use
some
backend
service.
That
is
like
authorized
authenticated
to
give
the
configuration,
and
some
folks
would
say,
like
the
configuration
can
only
be
provided
when
the
sdk
is
trying
to
do
initialization.
A
Some
folks
would
think
some
of
the
configuration
could
change
on
the
fly
where
the
application
is
running
like
the
sample
it
could
change.
So
in
this
way,
I
I
think
the
design,
although
it's
not
like
totally
spacked
out
in
the
open,
timely
specification,
but
I
think
the
general
practice
is
they
leave
this
as
an
abstract,
something
like
eye
configuration
service
where
folks
can
get
notified
on
the
configuration
change
and
they
can
compare
what's
the
minimum
change
they
need
to
do.
A
For
example,
if
you
have
like
a
thousand
different
configuration
values,
the
subscribers
should
be
able
to
get
notification.
If
only
one
thing
changed
and
they
don't
have
to
go
through
the
entire
change
to
figure
out
what
got
changed
so
there
should
be
some
algorithm
to
improve
that.
But
that's
not
the
the
must
like
have
thing.
I
I
think,
having
that
abstraction
and
then
in
open
telemetry
overall
there's
a
thinking
that
some
of
the
most
significant
configurations
should
be
toast
through
environment
variable.
A
But
you
can
see
there
is
always
a
balance,
so
environment
variable
is
something
that
doesn't
give
you
a
strong
type.
So
you
can
only
specify
some
string
and
you
can
potentially
squeeze
a
lot
of
things
like
computers
like
json,
you
can
escape
different
things,
but
ultimately
you
got
limitation
on
how
many
stuff
you
can
squeeze
into
that
configuration
so
right.
I
would
definitely
imagine
there
will
be
some
some
balance.
A
For
example,
you
can
even
have
a
scenario
where
you
have
three
exposures
and
each
exposure
would
expect
them
to
have
different
endpoints
than
do
you
expect
to
have
different
environment
variable
or
a
giant
environment
variable
stream
to
have
multiple
things.
So
I
think
that
that
might
change
in
the
future,
something
that
I
I
don't
think
will
will
change,
is
to
have
that
abstraction
layer.
A
So
if
you
want
to
make
the
configuration
happen,
first
focus
on
the
abstraction
layer
having
that
interface
and
then
for
environment
variable
you
can,
based
on
your
best
knowledge
like
currently
they're
already
some
proposal
in
the
open,
telemetry
spec.
Some
of
them
are
not
merged
yet,
but
it
gives
you
a
common
sense.
What
people
try
to
do
and
no
matter
what
we
decided
to
do
it
or
not?
The
requirement
will
be
there.
Otherwise,
people
wouldn't
be
crazy
and
talk
about
that
in
a
stack.
So
the
world
will
need
that.
A
A
We
got
a
default
standard
so
folks
who
just
got
the
binary?
They
don't
need
to
compile
anything
or
change
any
instrumentation.
The
worst
case
is
the
spec
decided
not
to
do
that.
Then
we
can
decide
whether
we
need
to
take
extra
steps
in
this
sdk
or
we
just
leave
that
as
one
example
to
give
you
a
best
practice.
C
I
see
so
basically
what
I'm
my
understanding
from
what
you
said
is
that
we
don't
we're
not
sure
yet
how
people
will
configure
it
like.
It
might
be
environment
variables.
We
might
have
more
flexibility
so
for
now,
in
the
abstraction
layer
that
you're
talking
about
that
sounds
basically
like
this
struct
right
that
gets
passed
into
the
constructor
and
that
can
be
set
in
a
lot
of
different
ways,
and
I
don't
have
to
worry
about
how
it's
set
quite
yet
yeah.
A
A
Yeah
one
example
is
out
that,
like
you
can
configure
individual
linux
machine,
but
if
you
have
a
lot
of
linux
machines,
they
can
form
a
group,
and
you
can
start
to
push
group
policies
through
that.
If
you
have
a
big
augmentation
or
you
have
different
technologies,
they
can
have
a
federated
model
where
necessarily
like
they
don't
necessarily
trust
each
other,
but
they
can
label
hey.
I
want
to
get
this
configuration
from
out
that
this
way
you
can
build
more
scalable
software
and
even
open
time.
A
Tree
is
a
project
that
will
work
like
towards
multiple,
like
integration,
so
I
don't
expect
we'll
build
a
monolithic
sdk
that
that
is
just
a
black
box.
We'll
have
that
visibility
so
that
aligned
with
like
the
overall
open
telemetry
direction.
I
believe.
C
C
A
A
Yeah
cool,
definitely
and
one
one
interesting
thing,
so
we're
we're
not
encountering
that
today,
but
we
might
encounter
that
soon
in
in
this
project,
is
think
about
like
otl
tx
powder
or
any
like
zip
king
jaeger
exploder.
They
expected
us
through
the
normal,
like
rpc
channel.
It's
either
like
vr,
you
say
http
udp,
but
basically
it's
a
networking
layer
and
for
a
lot
of
folks.
They
they
do
care
about
automatic
instrumentation
for
those
layers,
for
example,
if
I'm
a
user
I
use
like
http
client
like
live
curl.
A
I
don't
expect
myself
to
write
a
lot
of
code
just
to
instrument
later
I
would
expect
either
lip
car
comes
from
the
like
with
pre-instrumented
or
there's
an
adapter,
where
I
can
simply
enable
say
I
want
all
the
label
car
outgoing
cost
to
be
to
be
tracked
like
generating
like
traces
from
in
this
way,
we
might
join
into
a
live
loop
where
we
start
to
export
something
to
the
http
client,
and
then
we
generate
those
spans
and
feedback
into
our
queue
and
then
like
that
runs
just
forever.
A
A
Here
the
problem
has
has
been
like
almost
in
every
single
sdk.
Remember,
like
folks
solved
that
probably
like
two
years
ago,
in
open
census
pipeline
and
later
they
saw
that
in
open,
telemetry
pipe
and
now
we're
coming
back
again
and
we'll
come
so
it
seems
like
a
problem
that
we
keep
revisiting
and
trying
to
solve
that
every
few
months
and
hopefully
like
we
can,
we
can
finish
that
game
here.
C
Oh
sorry,
so
what
you
were
just
talking
about
with
the
refraining
from
instrumentation
is
that
related
to
the
example.
A
Not
related
to
the
example,
but
more
like
when
you
work
on
the
example,
you
might
hit
issue
where
okay
well
like
if
I'm
using
the
oti
opx
powder
and
the
underlying
rpc
client
is
auto
instrumented.
Should
I
do
so.
This
is
not
a
big
issue
or
you
can.
You
can
basically
say
I
don't
worry
about
that
now,
because
we
don't
have
any
auto
instrumentation
in
c
plus
plus
project,
but
this
is
more
when
you
think
further.
A
This
might
be
an
interesting
area
for
you
to
explore,
not
necessarily
that
that's
a
that's
a
interesting
so
anything
related
to
the
context
propagation
and
how
to
avoid
that
we
like.
Basically,
we
should
ourselves,
is
a
hard
issue,
we're
monitoring
and
we
need
to
interact
with
the
others
like
we
need
to
export
it.
We
need
to
receive
data,
and
yet
no,
the
underlying
library
that
we
depend
on,
might
also
be
monitored
and
feedback
to
our
system.
This
is
a
another
example.
A
C
A
That
brings
the
question:
if
the
sdk
is
not
working
correctly,
they
need
internal
logging,
and
yet
the
internal
login
cannot
depend
on
the
sdk,
because
the
ick
is
messed
up
right.
So
it's
like
that
problem,
so
you
can
see
a
lot
of
things
in
such
like
monitoring
or
telemetry
space
where
like
they're
circular
and
you
need
basically,
you
need
to
build
something
simpler,
that
this
ic
can
rely
on.
So
if
the
ick
is
not
working
correctly,
you
cannot
call
it
sdk
logging,
but
you
can
call
something
like
meta
sdk
logging.
A
C
All
right
and
then
other
than
that
I
had
a
quick
question
about.
I
think
we
had
talked
earlier
about
adding
like
some
memory
benchmarking
tests
to
the
project.
So
I
was
wondering
if
that's
still
something
that
we're
interested
in
like
we
could
use
valgrind
or
something.
A
Yeah
definitely
yeah,
so
I
I
think
again.
This
is
not
a
blocker
for
now.
I
I
think
having
some
performance
benchmark
like
the
cpu
and
memory
would
be
critical
if
we
want
to
get
the
sdk
to
like
beat
our
ga
status,
not
for
now,
but
it's
definitely
nice
to
have
them
and
for
memory.
I
think
the
scope
is
smaller
because
we
run
for
cpu
stress,
then
you
need
to
have
some
dedicated
machine.
C
Okay,
yeah
so
in
terms
of
the
actual
library
that
we'd
want
to
use
for
this,
so
is
valgrind,
okay
or
like.
A
Is
there
something
yeah
and
I
think,
like
we
already
have
something
related
to
that
so
johannes
had
had
some
initial
work
on
that
to
test
the
at
least
the
memory
leakage?
Okay,
probably
it
can
extend
his
work.
So
if
you
search
for
vagrant
in
the
code
base,
I
believe
you
can
find
something
and
if
it's
not
merged
yet
I
believe,
there's
open
pr
about
that.
C
C
At
the
end
of
the
week,
okay
cool
yeah
other
than
that-
I
guess
since
this
is
my
last
sig
meeting,
I
just
wanted
to
say
thank
you
so
much
to
all
of
the
maintainers
and
everyone
working
on
this
project.
I
had
a
really
incredible
summer,
working
with
all
of
you
and
I
learned
a
lot
so
yeah
thanks
a
lot.
A
B
B
Hi
everybody
so
yeah.
I
just
had
a
couple
like
non-fucking
functional
prs,
so
just
the
readme
and
then
the
xd
fuller,
restructure.
A
B
I
know
max
you
were
saying
that
you
wanted
the
hsp
server
to
be
less
nested.
It's
still
pretty
nested,
I'm
not
sure
about.
B
If
this
is
the
kind
of
solution
you
would
have
wanted
for
the
hdb
server
under
the
ext
folder
but
yeah,
maybe
you
can
get
more
of
a
look
at
it
later
and
I'm
also
working
on
benchmarks
for
the
back-end
components
for
z
pages,
and
that
should
be
done
in
a
couple
days.
I'm
currently
getting
reviewed
by
my
host
and
co-host.
B
A
B
Oh
all
right
yeah,
I
think
I've
seen
the
channel.
I
guess
I
was
just
like
wondering
because
I
don't
know
if
you
wanted,
like
a
dedicated
link
just
for
the
z
pages,
but
yeah
I
get.
The
logistics
would
be
a
little
weird
to
handle
like
kind
of
do.
A
Not
I
I
think
we
can
potentially
leverage
leverage
this
like
meeting
link.
We
just
need
to
find
and
like
open
slot.
If
you
open
the
the
open,
telemetry
public
calendar,
you
can
see,
there
are
only
three
different
bridges.
So
every
time
like
you
join
this
meeting,
we'll
have
the
same
id
here
as
the
other
things.
A
We
only
have
three
like
we
purchased
three
channels
and
they
will
be
automatically
recorded.
So
if
you
can
find
a
free
slot,
no
one
can
potentially
work
with
morgan
is
the
the
pm
from
google
to
create
another
slot
in
the
public
calendar,
and
we
just
join
that
meeting,
pick
a
recording
and
that
got
published
to
youtube
and
that's
a
dedicated
session
for
you.
So
you
can
put
the
link
in
the
document.
B
A
Yeah,
so
we
can
probably
follow
up
offline
yeah,
like
you
can
ping
me
on
get
your
channel
I'll
connect
you
with
more
than.
B
Oh
okay,
cool!
Thank
you,
but
yeah.
It's
also
my
last
week
I've
just
been
working
on
some
spec
things
in
the
blog
post,
so
I
haven't
been
able
to
do
much
else,
but
yeah
kind
of
echoing
what
nadia
was
saying
earlier.
I
really
appreciate
all
of
your
help
and
guidance
throughout
my
internship.
I
also
feel
like
I
learned
a
lot,
and
I
really
appreciated
how
helpful
everyone
was
and
kind
of
getting
us
up
to
speed
with.
B
What's
going
on,
though
the
repo
and
giving
us
feedback
on
our
code,
it
was
very
cool
experience
for
doing
open
source
for
first
time.
A
Yeah,
thank
you.
I
appreciate
your
help
and
I
still
remember
like,
like,
like
the
demo,
you
did
like
it's
great,
to
see
something
like
intune
and
has
like
real
stuff
helps
us
to
attract
more
folks
from
the
community.
Thank
you.
B
A
So
I'll
probably
start
from
from
here-
I
I
think
anything
older
than
seven
days.
We
probably
already
covered
them
last
time.
So,
if,
like
folks,
are
making
progress,
they
want
to
get
things
merged.
Please
make
sure,
like
you
cover
all
the
all
the
ci
failures
and
get
things
in
green
and
if
things
are
in
green
and
marked
as
please
merge
I'll
I'll
merge
that
immediately
if
things
are
stuck
somewhere
either
you
need
help
from
the
others
or
you
need
my
input.
Please
bring
us
on
guitar
channel
this
week.
A
I
should
have
more
time
comparing
to
previous
two
weeks.
So
now,
let's
look
at
this,
I
I
I
think
this
one
will
cover
so
there's
some
ci
issue
and
I
believe.
D
Yes,
actually,
I
just
ran
the
the
ci
test.
I
I
modified
just
now,
and
could
you
click
to
include
that
pr,
yeah
yeah.
A
D
D
Yeah,
I'm
not
sure
whether,
like
I'm
doing
the
correct
thing,
but
I
could
you,
like
click
into
the
the
checks
of
the
of
this
pr.
D
Yeah,
so
I
actually
yeah,
you
can
see
the
cmake
test
here
that
there
is
a
news:
subtask
called
the
room
from
easiest
exporter
tests.
D
A
A
Okay
and
this
one
is
still
a
draft.
A
Okay,
so
I'll
take
a
quick
start
after
the
meeting-
and
this
one,
I
think,
is
some
refactor
work.
You
won't
have
some
component
better
shared
among
different
components,
right.
A
E
Oh
yes,
I'll
be
good,
and
I
also
have
some
follow-up
on
cmake
builds
like
don't
worry
about
this
I'll.
C
I
have
one
thing
so
for
tianlin's
pr
143,
it's
called
added
http
trace
contacts
yeah.
I
think
it
has
approval
from
the
google
side.
It
just
needs
approval
from
one
of
you
guys,
yeah.
A
Yeah
then
I
want
to
thank
you
say
thank
you
for
the
interns
who
have
contributed
so
far
and
we're
not
going
to
join
us
next
week.
So
thank
you
for
your
contribution
and
we
can
still
keep
in
touch
on
github.