►
From YouTube: 2020-07-07 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
A
B
C
C
C
C
C
I
just
shared
my
screen,
let
me
know
you
can
see
that
I
think
most
of
the
folks
should
be
already
ramped
up
on
this
project.
I've
seen
my
PRS
pretty
active
for
almost
all
the
work
streams
we
had
so
like
they're
all
turning
things.
I
want
to
do
a
like
called
here,
otherwise
I
I
think.
Probably
we
can
spend
more
time
going
through
the
PR
and
see
what's
the
status
and
I
probably
need
to
catch
up.
C
E
E
Sure
I'm
just
trying
to
reconcile-
and
you
know
or
reconstruct
the
picture
in
my
head
right
now.
So
would
you
see
if
I'm
saying
some
dumb
things
here
like
let's
say
in
a
case
where
we
have
loadable
tracers
and
the
API
currently
allows
runtime
loading
or
the
plug-in
like
different
tracers,
stable,
a
bi-stable
interface
and
the
API
for
that
for
the
span
processor
row,
I
assume
that
the
page
is
supposed
to
be
this
spanned
processor?
E
What
are
the
bindings?
How
do
we
ensure
that
for
all
libraries
loaded,
we
pass
the
necessary
information
to
the
C
pages
implementation,
or
is
it
that
the
G
pages
must
be
compiled
in
and
concrete,
tracer
library?
That's
something
that
I'm
not
entirely
understanding
like
is
different.
Vendor
libraries,
you
see
what
I'm
saying
it's
like
different
exporters
pre-built
who
installed
it
and
then
I
want
to
use
the
pages
to
monitor
my
flows.
E
D
D
E
Like
maybe
we
should
keep
that
in
mind,
but
still
do
the
actual,
concrete
implementation.
First,
let's
say
just
one
tracer
and
this
tracer
implementation
has
some
way
of
binding
to
Z
pages
and
cover
this.
But
we
have
to
keep
that
question
in
mind
like
for
the,
because
one
of
the
goals
has
been
always
to
provide
that
API
I
can
pass.
Then,
if
that's
goal,
and
if
we
are
planning
to
distribute
some
pre-built
packages
like
enviros
linux
distros,
my
question
is
going
to
be.
I
am
running
my
server
with
a
few
exporters.
E
How
do
I
benefit
from
zip
by
edges
in
that
runtime
loader
bo
environment?
Maybe
it's
not
even
for
interns
to
solve
right
now
you
see
what
I'm
saying.
Maybe
it's
a
challenging
work
item
that
we
need
to
discuss
and
describe
how
that's
like
I,
don't
have
an
answer?
That's
why?
Maybe
we
should
just
you
know,
discuss
and
figure
out.
What's
the
right
path.
D
A
So
I
spoke
to
Sergey
and
he
said
that
the
three
pages
should
not
have
a
dedicated
exporter,
so
kind
of
the
span.
Processor
is
responsible
for
interfacing,
with
the
app
and
kind
of
making
the
recordable,
which
is
the
purpose
of
the
exporter.
I
didn't
really
I'm
still
talking
to
him
about
it,
but
I'm
not
quite
sure
what
he
has
in
mind.
E
So
there's
a
bit
of
a
catch
here
so
say:
let's
say
we
have
some
description
of
the
exporter
interface,
but
when
we
went
down
to
report
about
and
all
these
internal
SDK
implementation
details,
we
I
think
we
said
that
we're
not
gonna
provide
like
a
bi-stable
interface
for
that,
like
it's
an
internal
implementation
detail
within
the
treasure
within
the
concrete,
SDK
implementer.
Who
decides
how
it's
done
now
that
immediately,
if
we
don't
have
stable
spec
for
that?
E
That
means
we
cannot
make
span
processor
interrupt
with
runtime
loadable
tracers
from
different
vendors
and
I,
see
that
it's
an
important
feature
to
have,
because,
when
I'm
observing
a
system,
I
want
to
see
everything
what
that
system
is
doing,
not
just
what
one
of
the
exporters
but
the
full
set,
but
for
now
like
I
guess,
the
first
stage
is
once
we
let's
say
forget
about
ABI.
Let's
focus
on
spun
processor
interface
within
the
SDK,
make
it
work
like
a
man
decide.
E
F
When
I
think
to
spend
processor
has
a
defined
interface
between
died,
speaker
because
process
or
interface
and
recording
api
compatibility,
I
mean
it
starts
out.
I.
Remember
what
we
agreed
on
balls
that
we
assumed
that
the
application
engineer
is
able
to
ensure
that
the
SDK
and
all
exporters
or
spam
processors
basically
share
the
same
API.
So
we
don't
care
about
api
compatibility
on
this
level,
which.
E
Like
we
have
the
stable
classes
right,
whatever
we
describe
them
under
the
API
folder
is
fine,
but
the
spawn
processor
is
not
in
there.
One
processor
isn't
the
SDK
if
I
load
a
dot
a
so
and
be
the
so
for
the
two
tracers.
How
do
I
make
sure
that
the
two
tracers
that
only
followed
the
contract
on
API
surface
would
be
able
to
interrupt
with
the
same
as,
if
a
giant
processor,
what
does
it
in
the
booth
to
have
their
own
GPS,
which
is
gonna.
F
Be
a
bit
messy
I,
don't
understand
because
they
talk
about
traces
to
trace
or
implementation
is
part
of
the
SDK
and
I
think
we
have
to
be
so
I
think
they
don't
support
having
different
SDKs
it
once
we
only
support
having
one
SDK
at
once
initialized.
So
we
can
kind
of
assume
that
our
tracers
that
are
neutralized
are
of
the
same,
are
instantiated
but
from
the
same
SDK
or
may
be
missing
something.
C
C
The
answer
is
that
processor
thing
is
backed
out
in
the
ice
DK
and
for
what
happens
backed
out
in
the
SDK.
We
don't
bring
the
binary
Interop
with
otherwise
decays.
This
is
the
istick
head.
Fisher
is
bringing
as
a
sdk
package
and
you'd
only
expect
a
component
from
the
IC
hi
freedom
products
call
her
we'll
walk
with
another
sdk
from
a
different
vendor.
E
E
Okay,
I'm
just
trying
to
understand
the
value
of
the
processor,
then,
because,
if
I'm,
building
my
own
SDK
right
and
there's
a
limited
set
of
API
calls
on
that
surface,
I
can
then
proxy
all
calls
through
my
own
API
surface
to
other
I,
create
implementation
like
what's
the
use
case
for
this
span.
Processor,
then,
if
I
can
intercept,
Oh
API
calls
and
provide
some
aggregate
statistics
on
on
everything.
That's
been
done
like
through
that
anyway.
So
it's
full
up,
just
maybe
I'm
missing
something
yeah.
E
That's
what
I'm
trying
to
tell,
but
since
we
don't
have
a
bi-stable
spec
for
this
one
processor,
this
becomes
an
engine
always
decayed
and
if
both
of
the
kids
go
with
their
own
information,
it's
a
bit
nice.
You
like
how
they
share
and
it's
a
separate
view
anyway.
So
let's
follow
up
on
this
offline
yeah.
C
E
D
E
C
And
so
far,
my
understanding
is
that
things
in
the
API
gauge
is
only
like
a
very
small
scope
or
instrumentation.
We
want
to
make
the
API
as
small
as
possible
just
for
customers
who
simply
want
to
instrument
their
library
or
a
cold
anything
else
for
consumption
for
troubleshooting
business,
probably
not
in
the
API
stack.
They
want
to
make
the
change.
We
can't
just
go
through
those
back
meeting.
C
Okay,
so
coming
back
any
all
sending
things
I
think
we're
not
going
to
go
through
all
the
details
here.
This
is
for
people
to
do
cross
reference
leave
anything
you
got
blocked
or
you
think
there
is
something
someone
should
follow
up
or
like
I
should
fall
out.
Please
call
for
the
refers
to
like
something
yeah.
B
G
So
we
need
a
way
right
now
to
enforce
bounds
when
constructing
the
probability
sampler,
but
the
basal
flags.
Don't
let
us
throw
any
exceptions
you
Honus
recommended
I
bring
this
up.
Is
there
any
recommendation
on
how
we
should
proceed
regarding
that.
F
I
think
it
some
context
here,
because
we
definitely
agreed
that
our
API
colors,
like
the
API
package,
doesn't
throw
any
exceptions.
I
mean
it's
cute
to
me,
but
then
there
is
this
other
case
that
we
had
for
one
of
the
Assembly
hours.
We
were
basically
on
the
SDK
lever
when
assembly
is
initialized,
Nick
had
a
PR
up
that
kind
of
food,
an
exception
on
sampling,
utilization
and
kind
of
some
arguments
for
out
of
bounds
and
that
kind
of
kills
our
CI
tests.
F
So
it's
kind
of
a
corner
case
to
ask
a
baton,
SDK
level,
whether
we
are
allowed
to
throw
exceptions
there
or
not
and
I
trust,
let's
bring
it
up
in
the
stick
meeting
and
get
papers
because
our
current
CI
tests
are
really
very
strict.
So
we
don't
tolerate
any
exception
at
all
on
any
level,
isn't
just
a
question:
we
want
to
keep
that
up
or
kind
of
losing
that
wasn't.
E
This
kind
of
purpose
that
there's
a
special
macro
that
if
we
throw
we
can
still
build
with
the
no
exceptions
which
results
in
the
process,
determination
and
I
think
we
don't
really
use
exceptions
to
an
API
surface
at
all
per
se.
Then,
within
this
decay
implementation
you
can
decide
what
you're
gonna
do.
If
you're,
building
with
exceptions
you
throw
or
without
it's
a
stability
terminate.
F
E
Have
selfish
reasons,
for
example,
for
Android
there
might
be
a
use
cases
where,
with
preferential
ii
built
without
exception,
okay,
and
for
that
reason,
I
think
it'd
be
nicer
to
have
some
sort
of
for
status
code
API.
Instead
of
relying
on
exceptions
for
the
concrete
SDK
implementation,
I
think
it
might
be
fine
to
have
some
flexibility.
Do
we
absolutely
require
exception,
or
maybe
I
sir?
Where
can
we,
but
should
we
just
use
status
codes
instead?
I
think.
F
On
API
surface,
we
shouldn't
wish
to
avoid
exceptions.
You
know
never
bossy,
but
it's
just
a
question
like
on
the
SDK
side.
Then
instantiate
again
is
the
K
object
by
the
PF
exceptions
there.
So
I
think
we
definitely
can
go
without
exceptions
there.
It
makes
kind
of
the
code
a
bit
more
how
to
say
yeah
just
it
was
it.
Forces
has
to
make
some
implicit
assumptions
like
saying
an
invalid
argument
comes
saying:
they
just
use
something
for
value
there.
C
E
C
Bit
of
gold,
so
my
take
in
the
ice
DK,
we
don't
have
the
requirement
of
like
no
dependency
on
exception
at
all.
We
already
have
dependency
on
like
still
like
vector
or
the
other
types,
and
they
will
give
exception.
I
wonder
if
there's
a
way
we
can
remove
the
dependency
on
exception
entirely
from
Nasik.
My
answer
is:
currently
it's
not
possible.
C
We're
already
taking
dependency
on
exception
are
using
a
scale
types
yeah.
If
you
take
the
penis
out
vector
you
compile
that
you
use
like
no
no
exception
flag
in
compilation,
you
will
get
it
from
hardhome
failure
right
and
this
reverse
the
wrong
exception.
Anyways
and
I.
Think
here
for
the
SDK.
We
should
just
allow
the
the
like
any
of
the
masters
exposed
directly
by
the
is
DK
to
be
able
to
throw
exception.
It
might
be
a
nice
feature
if
we
can
allow
more
like
scenario:
freedom.
C
If
you
specify
a
flag,
then
you
can
get
some
call
back
or
something,
but
that's
too
complex,
so
my
suggestion
would
be
like
API
will
be
unchanged.
We're
giving
options
for
people
to
stay
with
like
the
arrow
and
fought
directly
killed
application,
or
if
you
compile
enable
the
exception,
you
will
get
exception
and
for
is
DP.
If
there's
matters
exposed
directly
and
their
exception,
we
should
just
roll
and
if
the
test
cases
would
be
to
straight,
we
should
just
adjust
the
test.
C
This
is
to
allow
exception
for
my
little
and
what
what
we
can
do,
I
I
think
is
so
the
test
cases
I
would
imagine
there
will
be
a
subset
which
only
tests
against
the
API,
and
we
can
make
sure,
like
those
test,
cases
are
not
getting
any
exception
and
for
the
ICJ
initialization
part
should
be
able
to
tell
the
exception.
Like
you
create
assembler,
you
create
your
own
particular
very
passing
invalid
value
see
them
whole
person
might
have
like
simple
race
or,
like
a
null
pointer.
We
should
cut
exception
directly.
F
Okay,
does
it
make
sense
to
me
to
turn
that
regard
Nick?
Maybe
you
can
rewrite
an
issue
for
that,
so
that
when
somebody
can
look
into
updating
our
CI
tests,
okay,
we
have
more
fine-grained
kind
of
tests
in,
in
the
exception
case,
so
that
we
only
take
the
API
layer
for
basically
being
free
of
exceptions,
but
that
we're
more
tolerable
isn't.
C
F
Think,
yes,
everyone
should
be
able
to
create
issue.
Quite
an
issue
and
I
would
recommend
to
create
an
issue,
and
then
we
can
see
how
text
it
up.
How
takes
the
issue
because
I'm
not
sure
if
that
slows
down
too
much
down
when
they
have
to
take
into
the
busy
I
dance
I'm,
not
sure
how
much
worked
it
would
be
just
to
have
this
more
fang-banger
yeah.
It's
protests
about
thanks,
I,
think
I.
Think
having
a
separate
issue
should
be
helped
any.
B
Here,
yes,
so
let
me.
B
Okay,
yeah
so
I'm
working
on
those
like
the
start
span
integration
so
right
now,
what
I'm
doing
is
what
we
call
start
span
above
tracer
like
a
tracer
instance
like
we
need
a
first
cost
enter
and
to
get
the
decision.
If
we,
you
know,
create
a
span
like
real
span
or
not
so,
but
the
problem
is
the
truth.
Sample
requires
I'm
like
parent
span
and
the
trace
ID,
but
I
I
don't
know
how
to
get
those
like
daddies
like
it's,
not
the
work
anywhere
in
decision.
C
B
C
Think
till
it's
a
are
probably
have
a
placeholder
for
so
that,
okay
with
him
and
see,
if
you
can
take
a
dependency
or
is
it's
going
to
wait
for
a
longer
time,
I
think
you
can
have
a
placeholder
feel
info.
You
can
comment
all
that,
so
your
simpler
is
not
going
to
take
that
dependencies
and
now
but
comment
and
later
when
you
add
that
sometimes
okay.
B
Okay,
so,
like
the
ideal
way
to
use
that
is
like
restore
reference
of
parents
fell
in
the
tracer.
Is
that
correct?
Okay?
So
a
span?
Has
the
parents
ban
is
a
corrected
and
that's
how
we
get
the
context.
C
B
C
C
Hey
known
as
early
so
there
are
the
different
ways:
if
you
pass
that
explicitly,
you
can
do
that,
but
normally
people
would
expect
you
get
the
parent
spend
contacts
from
the
contacts
API.
So
imagine.
If
you're
joining
a
thread,
contacts
situation,
we
take
some
incoming
requests.
That
request
will
have
the
trace,
ID
and
span
ID,
which
will
be
automatically
as
with
local
storage
as
your
contact
and
create
a
house
span,
you
don't
have
to
specify
that
because
the
context
is
like
by
default
using
the
they
local
story.
Oh
so.
B
It's
store
and
maybe
in
the
arena
or
somewhere
in
distress,
yeah.
C
C
F
C
F
Pass
in
an
arrow,
pointed
error
or
wastes,
and
then
maybe
also
they
are
going
to
make
an
issue
so
that
we
don't
forget
to
update
this
once
the
context
API
functionality
is
in
place
and
then
we
plug
that
other
place,
you're
kind
of
when
you
implement
some
kind
of
intermediate
parenting
handling
your
public
hiding
some
work
that
is
longer
square
into
things
that
doesn't
make
sense.
At
this
point.
C
B
F
H
We
had
a
little
quick
thing
surfer
for
batch,
recording
we
kind
of
need
a
way
to
be
able
to
pass
in,
like
pointers
to
instruments,
several
pointers
to
instruments
with
the
values
that
should
be
recorded
to
them
and
so
to
us,
like
the
most
usable
and
logical
way
to
do.
H
This
is
through
some
kind
of
key
value
scheme
where,
like
the
pointer,
is
the
key
and
then
the
values,
the
value
but
I,
think
currently,
the
in
know,
standard
key
value
interval
only
accepts
strings
as
keys,
so
I
just
want
some
advice
on
how
I
should
how
I
should
store
this
or
how
I
should
take
this
in
as
a
parameter.
C
H
It's
it's
definite,
something
that
we
just
need
to
iterate
over.
So
a
linked
list
is
definitely
possible.
I
just
asked
because
in
other
languages
or
I
think
in
all
the
other
languages.
The
way
they
do.
It
is
with
some
kind
of
house
hash
based
structure,
I.
E
H
E
Right
way
of
doing
it
in
a
bi,
safe
manner,
there's
some
test
earlier
in
my
example
that
is
submitted
I'm,
not
sure.
Yet
what?
But
you
can
take
a
look
at
what
I'm
using
it
for
and
I
think.
There
are
two
examples
like
key
value
is
horrible
and
the
other
one
is
passing
like
a
vector
structure
through
the
span.
Mm-Hmm
pretty
much
span
of
spans
in
your
scenario
might
be
the
right
approach.
C
E
Happens
is
like
correct
me
if
I'm
wrong
guys
when
we
are
compiling
a
portion
of
header
on
the
code
and
as
long
as
we
invoke
the
transform
into
Keeble
iterable,
which
provides
you
the
standard,
stable,
ABI
interface,
you
can
still
express
your
structures
in
you
in
the
application
code,
like
the
customer
who
has
map,
but
then
on,
like
on
on
an
API
call.
It
is
already
the
key
power
iterable.
That's
why
we
do
some
of
that
interesting
mumbo-jumbo,
so
I
thought
that
some
examples
would
help.
F
Ok,
I
think
also
have
to
economics.
Next
year
we
have
to
pick
to
separate
a
guy's
decay
concerns
few,
because
I
think
the
crisis
related
to
a
to
the
API
and
thinking
the
API
signatures.
We
should
have
any
voice
that
like
structures,
but
it
still
or
leads
to
the
possibility
that
SDK
implementation
to
kind
of
store
it
in
a
map,
yeah
argument
in
this
kind
of
span
or
key
it
represent.
You
can
standardize
the
case
store
it.
E
Yes,
if
the
key
can
use
whatever
classes,
it's
just
the
API
is
funny
and
I
think
for
a
period
would
help.
If
we
actually
can
approve
the
ABI
stability
release
over
release,
I
was
working
at
some
ABI
compatibility
checkers.
Perhaps
if
we
have
some
way
of
storing
a
previous
implementation
of
the
pre-built
binary,
then
we
can
integrate
some
sort
of
C
I
hope
to
verify
that
the
changes
are
not
breaking
ABI
I
can
but
a
good
release
over
release.
I
Enriching
enriching
that
the
context,
the
propagation
classes,
so
the
main
thing
I
focused
on
was
about
the
implementation
of
some
of
of
the
HTTP
text.
Format,
class
and
basically
I
realized
a
couple
of
function,
basically
most
functions
inside
inside
of
the
text
format,
but
then,
and
then
I
issued
a
PR
on
this
on
github.
But
then
one
developer
told
me
that
you're
not
supposed
to
have
a
detailed
implementation
in
the
API
folder
and
you
shouldn't
have
a
dot
C
C
file
in
the
API
folder.
I
But
then
I'm,
not
yeah,
I,
don't
quite
understand
what
ap,
what
the
API
a
folder,
a
dentist
and
or
if
we
only
have
that
many
implementations
for
those
functions
and
if
we
want
to
really
have
so.
If
we
really
want
to
have
a
functional
project,
then
we
must
have
the
implementations
right.
So
then
should
I
put
those
files
in
the
SDK,
footer
or
I
should
just
straight
out
change
the
thought
CC
file
I
was
into
a
header
files,
but
just
with
a
complete
set
of
implementations,
I.
E
Think
the
original
comment
was
was
from
Emil
and
I
actually
agree
with
what
he
is
saying,
so
it
just
it's
just
quite
a
file.
You'd
need
to
split
it
into
two
parts.
Like
imagine
that
you
have
a
pure
virtual
kind
interface,
which
must
follow
certain
agreed
rules
like
only
no
STD
classes
on
that,
and
then
you
have
a
concrete
implementation
under
the
SDK,
which
will
be
the
actual
job
and
then
I
consider
that
what
if,
for
example,
I
take
just
the
API
directory
and
I
do
a
complete
rewrite
of
the
SDK
from
scratch?
I
I
E
E
C
My
also
will
be
checked
the
Python
contacts
implementation,
so
my
understanding
might
be
outdated,
but
well
I
was
working
on
Python.
This
is
what
we
discussed
so
the
API
package
and
is
DK
there
little
bit
different
from
the
traditional
way
of
how
we
look
at
the
API.
So
traditionally,
when
you
look
at
API,
it
is
something
that
he
used
and
you
shouldn't
be
able
to
just
use
the
API
and
be
able
to
compile
your
product
because
there's
no
implementation,
basically
pure,
like
virtual
function
here
in
open
country.
It's
different.
C
The
API
is
the
minimum
viable
package
that
works
with
your
applications.
You
instrument
in
the
API
and
your
application
compiles
and
runs
without
any
problem.
It
follows
the
the
spec
and
the
only
thing
that
you
don't
have
is
you
don't
have
any
telemetry
out
of
this
application
because
they're
not
collected
and
I
think
it's
going
to
make
the
things
in
real
so
with
the
istick
injected
into
your
application.
It
hooked
up
with
the
API
and
all
of
a
sudden
your
data
started
flowing
to
the
SDK.
You
start
having
like
Spanish
custom
hold.
C
You
start,
have
metrics
being
aggregated
and
worried.
Exporters
with
that
spirit,
we
were
saying
that
the
open
climate
API
package
will
at
least
meet
with
the
w3c
to
his
contacts
means
if
you
use
the
API
to
implement
your
service,
someone
calling
you
using
the
trace,
ID
and
span
ID,
and
you
will
be
able
to
follow
and
react
to
those
ideas,
and
when
you
call
the
next
hop
of
service,
you
should
have
the
right
span.
Id
and
trace
ID.
C
Instead
of
ignoring
everything
in
Harley
and
with
that
said,
I
think
the
latest
suspect
might
have
changes
and
I.
Think
Python
is
always
ahead
of
sleep
as
fast
and
my
suggestion
would
be
take
a
look
at
Python
if
the
Python
HTTP
format
and
the
propagation
budget
is
in
the
API
package,
do
the
same
for
a
sip
has
passed.
It
is
not,
then
still
do
the
same
for
C
pass
pass
and
regarding,
if
like
by
looking
at
the
Python,
you
realized
Python
is
doing
the
information
in
the
API.
What
are
you
going
to
do?
C
The
answer
will
be:
do
the
same
thing
in
the
sip
has
passed
product,
but
keeping
in
mind
that
all
the
implementation
logic
inside
the
API
package
should
be
Heather
only
approach
you
don't
have
like
CC
file,
it
has
to
be
header
and
included
in
the
customers.
Application
got
compiled
at
their
compression
unit
instead
of
having
your
own
collection.
C
I
C
I
E
Riley
have
a
questions
here,
a
little
bit
about
like
say
if
we
have
a
new
object
pattern,
tracer
and
no
object
pattern,
implementation
of
everything
why
he
got
the
food
with
the
aspect.
We
could
have
separated
that
into
a
separate
header
file
inside
the
API,
where
those
who
just
needed
to
test
how
it
operates
was
perfect
performance,
no
clog
where
they
can
just
use
that
one.
E
So
I
think
we
might
want
to
go
with
a
bit
more
clear,
a
definition
of
let's
say
no
object,
dot
age,
and
that
gives
you
like
the
full
set
of
tracers
and
spans
in
there
and
for
that
context,
question
I.
Think
one
of
the
issues
is
in
the
proposed
TR.
They
were
classes
like
reg,
X
vector
and
a
few
others
stuff
that
we
generally
don't
exactly,
except
in
the
API
right
now.
So
when
we
go
with
pure
virtual
interface,
it
should
not
have
unfertile
things
yeah
well,
the
concrete
implementation
may
have.
C
C
C
I
C
C
Part
is
not
clearly
documented
in
this
bag
and
that's
a
reality.
I
think
working
at
least
people
can
agree
on.
Is
your
application
is
going
to
work
in
production
without
giving
you
like
random
exception
or
error,
whether
it
conforms
to
the
the
trace
contacts?
Spec
is
something
we
have
been
debating
and
I
in
mistake,
yeah,
but.
C
C
I
I
think
if
the
contacts
are
indeed
being
passed
from
one
process
to
another,
then
that
is
basically
the
entire
at
the
entire,
the
entire
job
I'm
supposed
to
do
so,
I
guess
and
since
that
part
is
going
to
be
done
in
SDK,
then
I
simply
is
just
a
compatible
interface.
It
is
a
collection
of
comparable
interfaces.
I
E
My
understanding
is
that
when
we
have
something
in
the
API,
it's
a
no
object
pattern.
Implementation
that
efficiently
does
nothing
he's
doing
nothing
conforming
to
spec.
I
cannot
comment
on
that.
Like
I,
see
that
I
can
at
any
given
moment
of
time
replace
my
can
create
this
decay.
Implementation
with
no
and
my
service
is
still
fully
operational.
E
C
C
J
C
E
C
E
E
Right
so
my
worry
here
is
that
if
say,
we
provide
a
normal
object
implementation
and
we
spell
it
out,
no
log
doesn't
actually
store
anything.
Maybe
we
should
discuss
this
during
the
break.
See
ya!
Okay,
let's
discuss
is
during
this
I'm,
just
trying
to
see
how
we
can
unblock
them
from
like
to
get
some
to
something
functional.
E
F
Up
offline
untouched
from
here
anything
some
time
ago
we
talked
about
it-
was
this
requirement
that
we
should
be
able
to
probably
context
like
the
WCC
context,
even
for
the
bino
SDKs
initialized.
Is
this
still
a
requirement?
Do
we
need
to
get
you
to
do
that
I?
Think
so?
Okay,
because
then,
if
that's
a
requirement,
when
you
think
we
have
to
provide
and
working
context
in
orientation
in
the
API,
otherwise
we
will
not
be
able
to
propagate
w3c.
Yes,.
F
So
I
think
with
that
mr.
primont
in
mind.
I
think
we
need
to
provide
implementation
in
the
API
I
think
and
provide
a
way
to
override
it.
So
I
think
the
implantation
API
it
doesn't
need
to
be
the
most
performance.
So
we
can
let
the
SDKs
over
right
type,
your
implementation,
but
we
need
to
provide
a
working
default
implementation.
Yet
in
time
and.
E
Then
maybe
we
shouldn't
try
to
add,
store
additional
storage
crisis
to
noise
today,
like
if
you
am
saying,
maybe
if
it's
all
header
only
and
it's
a
build
in
one
spot.
Maybe
we
just
go
and
say
this
squint
text
implementation
uses
standard
library
available
in
your
compiler
and
you
still
get
no
implementation
with
context
propagation
without
having
to
take
on
that
hassle
of
maintaining
our
own
clone,
like
I'm,
just
trying
to
avoid
the
unnecessary
effort
here.
E
F
I'll
need
to
think
about
this,
like
this,
like
I,
think
sacrificing
API
stability.
That's
a
big
thing,
I
think
maybe
an
other
route
for
Peters
to
scale
down
on
the
like
no
standard
immunization
I
mean
we
don't
need
to
have
a
vector
object.
We
can
just
have
like
the
context
of
be
a
linked
list,
so
each
context
kind
of
is
a
link
to
its
parent
context,
and
that
can
be
how
we
manage
kind
of
the
context
hierarchy
with
that,
just
as
an
alternative
it
year.
F
A
C
K
K
B
L
L
According
to
that
issue,
it
says
that
for
the
timer
functionality
there's
needs
to
be
this
dispatcher
that
has
to
be
used,
but
once
you
open
the
PR,
it
hasn't
actually
been
merged.
Yet
yeah,
so
I
was
wondering
whether
this
could
be
approved
and
yeah
I.
Think
I
think
it's
that
one
yeah
and
I
spoke
with
Ryan.
He
said
that
everything
is
done
yeah.
He
just
produced
all
the
few
conflicts
and
little
bit
good
to
go.
Yeah.