►
From YouTube: 2020-07-13 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
B
B
B
B
B
B
B
B
There
are
some
some
common
questions
on
different
work
streams,
so
examples
are
regarding
the
ABI
stability
or
regarding
the
span
processor
or
exporter
I
in
general.
I
think
it'll
be
helpful.
If
folks
will
come
like
one
particular
work
stream
can
take
a
look
at
the
other
PRS
and
put
your
comment
there.
D
C
B
B
Think
on
the
Templar
side,
I'm
seeing
good
progress,
I've
seen
like
like
the
always
resolve,
and
also
they
follow
the
parent
and
also
the
probability
when
they're
in
you're,
in
pretty
good
shape
for
the
span
contact,
I'm
saying
the
initial
proposal
that
makes
sense.
I
have
my
minor
question
regarding
the
API
stability
I.
Think
there's
some
comment
but
trying
to
address
my
question,
but
I
need
to
understand
how
how
that
solve
the
problem
and
for
the
Z
pages,
I
think
we're
seeing
a
similar
thing
as
the
matrix
X
polar
design.
B
C
B
E
Yeah
the
probability
sampler
is
ready.
Now
it
was
mock.
It
was
blocked
by
the
span
context
PR
that
emerged
earlier
today,
but
it's
ready
to
go
now.
Yeah.
F
Job
yeah,
so
for
for
me,
no
tracer
integration
and
parent
or
LPR
were
all
merged
and
spent
start
spam.
Pr
is
under
review
and
I.
Think
I
think
I
can
merge.
It
like
I
can
complete
it
within
a
day.
So
there's
just
a
minor
issues:
I've
been
working
on
the
integration
tests
like
Rich's
are
like
unit
unit
tests
with
larger
scope,
but
actually
I,
don't
thinking
about
the
scope
of
the
test,
because
currently
we
have
limit
very
limited,
like
function.
F
So
I
think
my
thinking
is
I
can't
add
a
like
a
big
pass
in
both
tracer
provider
and
tracer,
and
test
functionality
from
creating
a
tracer
to
like
ending
a
summer
spent
and
check
the
the
data
is
right.
So
that's
my
current
pan
I'm,
not
sure.
If,
if
we
can
like
test
it
on
longer
scope,
yeah,
that's
my
current
progress,
but
there's
are
still
a
question
like
what
I
do.
Let's
go,
should
we
cast
our
like
Kemper.
B
F
So
current
the
current
unit
testing
uses
a
mark
exporter
which
saves
data
in
a
fan
like
say
a
sent
expanding
structure,
so
I
can
use
the
same
way
and
but
I
think
so.
If
I
write,
many
of
the
functions
will
be
like
well
overlap
with
other
testing,
so
yeah
I'm
just
wondering
if,
if
it
is
necessary
to
put
them
together
or
yeah,
something
like
that,
because
I
feel
like
I'm
rewriting
some
like
like
many
similar
code.
B
F
I
mean
yeah
I'm,
using
a
memory
exporter,
and
but
some
of
the
like
many
of
the
function
has
been
tested
in
the
unit
testing
and
if
I
write
a
integration
testing
which
combines
all
the
features
I'm,
essentially
combining
all
those
test
cases
right,
missus,
so
I'm
wondering
if
that's
necessary
or
like
can
we
do
it
in
another
way?
So.
F
H
When
I've
done
this
Oliver,
so
we
talked
about
a
bit
about
this
last
week
and
I
think
it's
it's
a
bit
confusing
talking
about
integration
test
because
in
the
in
the
open
democracy
plus
last
project,
if
unit
tests-
but
we
don't
have
any
new
equation
test
framework,
yet
right
integration
test.
We
basically
only
have
like
the
three
test
unit
tests
and
I
think
that
idea
will
follow,
is
for
like
getting
up
something
like
integration.
H
F
B
F
B
H
Okay,
we
already
or
Nick
and
Oliver
already
already
wrote
tests
like
that.
So
we
have
the
moxx.
It's
been
exported
basically
in
memory
exporter
and
already
are
those
tests
that
basically
test.
If
spans
arrived
there
so
I
think,
then
we
don't
need
much
integration
tests,
basically,
don't
need
to
do
or
they
don't
need
to
do
man
not
from
the
integration
test
from
the
ball.
Okay,.
C
G
C
H
One
thing
more
much
more
to
sampling
API,
because
that
at
one
point
it's
really
do
the
probability
simpler
and
what
we
agreed
on
with
Nick
for
now
is
that
it
just
implant
implemented
like
the
many
more
like
scope,
the
to
spec
defines
because
the
spec
defines
some
maths
and
several
shoots,
and
we
left
two
footprints
out
for
now,
because
I
saw
also
the
other
languages.
I
looked
bison
go
in
trouble,
they
don't
implement
those
results.
Yeah.
B
F
B
F
H
Look
just
back
first
time
for
probabilities.
It
defines
like
whatever
it
is,
I'm
prefer
do
and
then
it
says
so
I
but
you're
sure
to
be
configuration
to
change
the
route
spends
only
or
all
spends
it's
this
shuts.
Basically,
this
back
defines
that
we
left
out
for
now
from
the
probabilities,
an
emperor
there's
certain
kind
of
configuration
options
for
abilities
and
rotate
the
defined
in
respects,
but
they
don't
notify.
It
is
a
must,
but
is
afoot
and
yeah.
B
Is
okay
to
go
with
the
minimum
version
for
now?
Okay,
yeah,
because
that
would
be
simpler,
like
I,
I've,
cracked,
some
spec
issue,
probably
like
a
month
ago
and
I,
know
a
lot
of
folks
from
Python.
They
also
the
ropm
prototyping
like
implement
the
latest
back
and
they
have
concerns
so
I
expect
there
might
be
changes,
it
don't
make
sense.
Okay,
so
next
one
for
the
metrics
yeah.
A
Here
I,
so
we
have
open
PRS
for
the
metrics
classes,
so
meter
provider,
meter
and
metric
instruments,
we're
waiting
for
feedback
on
the
meter
and
metric
instruments.
I
think
the
the
meter
provider
is
relatively
uncontroversial,
there's
also
a
design
doc
out
for
the
exporter,
which
I
think
many
people
have
commented
on
and
we're
currently
working
on
the
SDK.
You
should
see
PRS
for
some
of
that
later
this
week,
yeah.
B
So
I
I
think
I
proved
one
like
spag
PR.
So
my
question
is:
do
you
expect
us
to
merge
that
markdown
file-
or
this
is
just
for
discussion?
So
it's
more
like
I'm
getting
feedback
from
all
of
your
guys,
like
some
of
the
design,
is
much
bigger
than
a
github
comment,
so
instead
we
use
markdown
file.
You
know
I'm
open
to
use
whatever
format,
but
the
question
is
once
we're.
Okay,
with
the
design
approach,
like
I
I
tend
to
signal
that
explicitly
so
I
will
I
will
sign
off.
B
I'll,
say:
I
approve
this
PR,
but
my
question
is:
should
we
merge
that
markdown
file
into
our
report,
all
I've,
already
merged
that
that
means
later
on?
We
need
to
maintain
that
someone
won't
want
to
make
the
change
they
need
to
maintain
the
markdown
file
and
the
question
is
who's
going
to
maintain
the
markdown
file
and
do
we
plan
to
publish
that
or
is
this
is
just
the
intermediate
alchemy
that
we
use
for
discussion
perverse
right.
A
I
mean
it's
so
there
are
a
lot
of
interns
working
on
this,
so
our
thought
process
was
kind
of
like
if
we,
if
we
leave
the
project
in
an
unfinished
state,
someone
else
should
be
able
to
look
at
our
design,
darkens
and
kind
of
know.
What's
going
on
yeah.
So
that's
that's
kind
of
what
we're
thinking
I.
What
do
you
think
the
best
way
to
make
that
design
doc
available
for
everyone
would
be
I.
B
G
I
A
A
B
Okay
and
I
think
I
have
a
penny
item
that
had
to
work
with
the
spec
sig
to
clarify
whether
the
exporter,
when
used
with
span
processor,
will
happen
concurrently
or
it'll,
be
middle
exclusive,
yeah.
The
way
I
look
at
the
spec
I
think
there's
some
some
issues:
oh
oh
I
I'll
get
back
tomorrow
and
now
King
your
guys
on
beta
Channel
perfect.
Thank
you.
J
So
poor
exporters
we
have
implemented
CPU
and
time
which
mark
tests
and
nurse
them
and
for
GCP
exporters.
Here
now
works
on
the
batching
processor,
which
is
related
and
for
GOP
we
have
some
minor
stuff
left
to
finish
some
implemented
stuff,
but
actually
looking
for
the
next
fix
piece
of
York
to
work
on,
we
think
potentially
we
may
work
on
the
context
propagation,
but
I,
don't
know
what
the
status
of
it
is
there
in
New
York.
Who
can
take
there
or
is
anything
else?
Oh
you
just
ask
your
phone.
B
Yeah
I
think
they're
few
contacts
see
ours
are
there
and
the
work
country
is
split
between
like
two
interns.
One
is
on
the
the
in
process.
Mobile
basically
define
the
contacts
API
and
implement
a
basic
thread-local
storage
like
contacts
carrier.
The
second
part
is
related
to
the
integration
with
the
like
HTTP,
where
we
take
the
actual,
like
span
contacts
from
the
contacts
API,
and
we
call
a
formatter
to
turn
that
into
the
w3c
trace
contacts
formatted,
and
then
we
pick
the
HTTP
client
and
the
server
associate
the
actual
HTTP
header
names.
B
J
B
B
H
H
K
Sorry
I'm,
actually
working
on
the
this
is
Nijo
I'm.
Actually
working
on
the
batch,
batching,
processor
and
I
had
like
a
follow-up
question
for
the
dispatcher
PR,
so
it
still
yet
to
be
merged.
I
realize
that
there's
some
checks
failing
and
I
was
looking
into
making
my
own
timer
and
it
seems
like
you
can
do
that
like
in
like
a
background
thread.
So
it's
working
if
I
was
just
wondering
if
this
was
like
a
hard
requirement.
H
Was
looking
a
bit
into
murdering
a
rides
to
PR
and
I
know
what
got
all
tests
to
pass?
I
did
some
Ripper
racing
and
did
some
record
orders
to
pass
except
like
based
on
Windows
there's
some,
it's
a
pretty
complex
little
train
there.
It's
kind
of
downloading
you
event
and
then
using
basil
on
Windows
to
compile
it
by
own
externals.
You
make
comment
integrated
into
basil
and
if
it
fails
currently
in
one
test
and
yeah
I
didn't
have
the
time
yet
to
look
deep
into
that
and
get
that
fix.
H
C
Do
have
one
concern
about
live
event,
I'm,
not
sure
about
the
practical
reasons,
but
chromium
of
which
actually
heavenly
uses
the
event.
My
understanding
it
doesn't
use
it
and
windows,
so
that
was
the
only
like
sure
that
I
have
in
mind
right
now
that
maybe
there
are
some
in
a
practical
consideration.
So
it
may
not
be
very
viable
to
use
that
library
in
Windows.
C
Like
I
can
certainly
check
on
Windows
with
Visual
Studio
17
and
19
haven't
merged
that
stage
yet
ball
in
my
local
build
tools,
I
actually
very
fine,
run
oh
go
on
at
least
two
compilers.
We
can
probably
also
run
it
on
15,
like
on
older,
a
visual
studio
as
well
to
cover
the
full
current.
A
set
of
supported
register,
diverse
yeah
I
can
take
over
and
help
you
harness.
If
you
have
branch
or
something
where
you're
currently
working
I
can
move
the
changes
from
there.
You.
H
Can't
just
take
to
the
PR
branch
like
I
committed
to
rinds.
Oh,
so
that's
that's
up
to
date
and
it's
just
one
place
living
those
tests.
It's
fading
and
yeah.
It's
I
think
some
are
related,
actually
I
think
every
it
works
on
Windows
to
code.
It's
just
the
integration.
Hardly
payment
is
built
on
Windows
I
said
that
doesn't
currently
work
right,
Oh,
interesting.
H
To
the
thing
that
is
failing
is
the
Burano
test
like
that
builds
the
whole
project
with
based
on
Windows
and
the
event
is
integrated
in
this
build
file.
As
an
external
see
make
project
so
pay,
it
will
basically
course
you
make
to
build,
live
event
and
somebody
doesn't
work
there.
Vista
version
detection
for
Visual
Studio,
so
the
build
fails
and
I
didn't
figure
out.
C
I
have
some
thoughts.
Maybe
we
can
adjust
this
so
instead
we
already
use
VC
package
in
a
few
sports,
for
example
fishing
like
COI.
We
use
VC
package
to
maybe
instead
of
manually
cloning
and
cooling
cement
directly,
we
can
just
call
y
jc--
package
install
live
event,
and
then
we
rely
on
whatever
maintain,
and
so
the
VC
package
to
provide
us
was
the
proper
build
recipe
for
that
I
can
try
like
I,
can
take
a
look.
L
So,
for
me,
this
week
is
maybe
about
its
many
about
updating
APR,
to
enable
transforming
the
STD
map
to
the
templated
method,
to
the
key
value
iterator
for
a
by
safety.
So,
after
discussion
with
max,
he
told
me
it's
it's
safe
to
use.
You
said
this
way,
so
I
I'm,
not
a
hundred
percent
sure
how
this
work,
but
I
I
try
to
update
the
PR
and
they
should
be
up
for
now
for
a
further
view.
And
another
thing
is
about
the
next
step:
I'm,
not
too
sure
about
the
request
framework
for
propagation.
L
So
in
the
chat
in
the
general
chat
max
mention
about
the
HTTP.
Only
simple
cross-platform
web
server,
like
I,
think
personally
I.
Look
more
into
that
and
discover
what
how
we
can
do
about
this
and
and
and
also
I,
have
made
a
cup
of
tests
for
the
HTTP
trace
context,
and
it's
a
relatively
complex
class
and
there
and
does
a
lot
of
complex
things.
So
I'm
not
sure
so.
I
have
made
a
couple
simple
tests
and
I'm
not
sure
if
they
will
do
the
job.
L
L
B
L
By
the
way,
I'm,
not
too
sure
so,
apart
from
the
propagation
folder
I
have
for
now,
I
also
read
about
the
propagator
API,
not
sure
if
I
mean
I.
Definitely
can't
do
that,
but
I
don't
know
if
there's
anyone
else
working
on
that,
so
I'm
not
sure.
If
that's
my
duty
to
implement
those
or
it's
some
other
team's
job.
L
D
D
L
B
B
So
there's
the
test
folder
it
gives
you
the
instruction
how
how
could
you
validate
your
service?
Assuming
you
have
a
simple
like
HTTP
server
implemented
and
a
client
implemented
both
using
the
propagator
that
you
wrote
for
the
SIP
pass
pass
and
if
they
are
conforming
to
the
trace
context
back,
you
should
be
able
to
get
a
hundred
percent
pass
and
many
of
the
open
clannish,
because
people
are
eating
this
for
complete
coverage.
In
this
way,
you
don't
have
to
write
individual
test
cases
in
your
report.
B
L
B
M
B
Do
they
all
fit
together?
Okay,
oh
so
I
answered
that
several
times,
probably
like,
like
like
20
times
this
year,
I'm
happy
to
do
that
here
again.
If
people
want
to
understand
more,
just
want
to
say
like
what
was
the
right
use
of
the
time
or
probably
we
can
go
with
the
zip
pages
and
later
we
had
time
I'm
having
to
expand
here
so
I
don't
have
to
explain
that
to
individuals
every
time
or
like
what
do
you
guys
think
or
like
in
studying
understanding
the
contacts
API
I
I.
C
J
M
M
B
B
Span
processors
will
run
concurrently
like
whenever
they're
suspended
and
processor
will
get
called,
and
given
that
that
thing
can
get
coffee,
multiple
threats
want
to
make
sure
all
the
operations
I
like
like
at
least
they're
not
going
to
break
the
customers
application.
So
they
shouldn't
crash
the
internal
states.
B
If
someone
is
going
to
take
this,
it
might
crash
their
application
or,
like
define
to
be
sure
they're.
Okay,.
D
C
D
B
So
I
can
briefly
explained
make
sure
we're
on
the
same
page.
So
this
twist,
like
twist
based
and
processor,
arrives
from
the
span
processor
like
it
has
charged
her
wishes
and
STR
type
and
any
icy
out
I'd
like
collection,
is
not
thread
safe
unless
mentioned,
so
it
has
a
running
STR
collector,
and
this
is
where,
when
we
tried
you
see,
where
is
my
common,
when
we
try
to
like
start
has
span,
this
function
will
get
called
and
it
will
insert
into
the
SQL
container,
and
this
is
where
the
structure
can
get.
B
D
I
I,
like
added
some
tests
and
I,
could
see
it
felt
like
when
I
do
the
the
big
available
to
teeth
on
yeah
the
issues
when
running
like
on
start
on
and
our
concurrently
run
so
I
like
tried
some
different
measures,
including
like
a
basic
block
card
from
the
SPL,
and
it's
they're
still
like
some
issue.
So
I'm
still
trying
to
work.
I
see.
B
So
it
like
you
can
leave
that
as
a
comment.
It's
not
like
threat
shifting
matter.
Basically,
this
part
is
broken.
It's
not
stressed
if
it's
going
to
low
your
application
away
and
we
can
make
that
X
place
it.
Some
scary
comment:
I
mean
people
like
you
should
not
use
this
one
and
find
words.
This
is
my
worries.
If
someone
is
taking
this
and
that
that
crash
their
application,
that'll
be
a
something
very
hard
fast
to
manage.
B
B
D
I
actually
tried
adding
you
lob
cards
on
both
the
Oscar
on
and
on
the
same
you
text
and
then
I
had
some
test
reading
like
those
functions
concurrently
and
there's
still
gives
me
some
flags.
That's
there's
some
like
issues
so
I,
don't
know
if
I
need
a
other
solutions.
I
should
look
into
because
I
was
also
been
trying
to
combination
so
blocks.
H
B
D
D
B
G
G
Another
thing
something
that
I'm
trying
to
do
is
I'm
trying
to
at
the
moment
the
data
aggregator
it
advocates
fan
when
a
request
for
requests
I'm
trying
to
aggregate
these
fans
at
regular
intervals.
So
there
aren't
too
many
stunts
coming
in
at
the
same
time
having
some
trouble
testing.
This
approach,
I'm
not
sure
how
you
guys
have
like
a
system
in
place
that
is
time-based
functions,
I.
B
Don't
think
so,
I
have
seen
the
like
the
test
infrastructure
that
that
can
cover
the
climate
thing
in
Python,
how
they
can
take
a
look
there
and
I.
Think
recently
the
maintainer
from
down
at
the
DI
did
not
probably
just
like
four
days
ago.
So
you
see
pass
pass
high,
I
think
probably
I
was
I
would
say
the
same
thing
so
follow
what
has
been
doing.
G
B
G
B
I
should
be
able
to
yeah
and
I
mean
done
at
high
I.
Think
folks
in
my
team,
they
they
were
initially
thinking
if
they
should
implement
a
marked
timer
and
just
to
use
a
predictable
like
very
predictable
sequence,
and
later
they
figured
it's
very
hard
maintenance,
so
I
would
say
like
a
DLA
here
we
can
use
math
and
everything
is
like.
B
Like
reproducible,
we
have
our
fake,
like
timer,
that
always
give
you
the
expected
series
of
events,
but
if
the
engineering
cost
is
too
high,
I
think
it's
fine
to
just
use
some
like
random,
like
behavior
timer
and
leave
that
has
a
like
a
reasonable
timeout
value.
For
as
long
as
like,
like
99%
of
the
cases,
we
have
very
reliable
results.
I
think
this
something
you
can
live
this.
G
B
N
Okay,
we
filed
a
PR
this
morning
for
the
design
document
and
we
are
waiting
for
comments
and
replies.
I
also
included
the
link
here
and
at
the
same
time,
we
started
defining
the
header
files
and
at
the
classes
we
also
reviewed
our
test
design
document
last
week.
So
we
are
also
translating
the
tear
pseudocode
into
C++
code
yeah.
That's
all
for
us.
Okay,
thanks.
O
B
Okay,
thanks
yeah,
now
coming
back
to
the
the
contacts
part.
So
the
first
thing
I
want
I
want
to
start
is
I
won't
explain
the
like
the
contacts
API,
so
they
talk
about.
Actually,
this
is
a.
We
were
open
time
and
we
try
to
stab
away
from
just
being
a
telemetry
solution,
because
a
lot
of
telemetry
solution
do
the
there,
depending
on
the
contacts
API
to
achieve
certain
things
from
how
I've
been
able
to
put
something
at
the
current
trace
and
then
I
did
receive
that
within
the
same
process.
B
So
we've
seen
like
in
different
language
or
different
runtime,
some
of
them
have
contacted.
Yeah
like
in
donĂt,
have
a
synchronous.
They
have
in
Python
also
have
two
different
contacts:
API
thing
goal:
we
have
contacts,
but
they
don't
have
the
contact
yeah.
So
that's
the
moment.
Open
telemetry
decided,
okay,
we're
going
to
standardize
the
contacts,
API
community,
all
consistent,
look
and
feel,
but
the
idea
is,
if
there's
already
a
language
we
probably
should
just
like
in
Python
I've
been
going
forward.
B
B
From
home
and
saying,
like
I'm,
doing
all
the
work
for
particular
guy,
this
is
the
user
ID,
because
like
Riley
and
here
when
I
actually
do
something
at
the
end
for
them
how
I'm
going
to
send
the
actual
seco
command
on
behalf
of
delay?
And
then
the
question
is:
how
do
you
propagate
this
information
comes
with
a
deep
stack?
It
could
be
a
hosta,
it
could
be
sound
like
workflow,
but
you
need
a
way
to
ultimately
propagate
this
information
to
the
very,
very
low-level
operation.
B
There
two
ways
of
doing
this:
the
first
one
is:
he
passed
either
ID
explicitly
for
every
single
step.
So
here
you
figure
out
how
you
have
a
function.
You
got
to
pass
in
the
user
ID
because
to
something
and
imagine
you
have
a
system
where
you
keep
coming
up
with
ideas.
Oh
I'm,
probably
at
this
or
even
have
the
energy
to
change
all
the
API
is
to
take
all
the
information
every
time
you
have
no
idea
in
this
way.
You
can
lead
to
do
something
instead
of
saying
this
is
the
other
idea.
B
Are
they
come
concept
or
contacts
and
all
the
subsequent
costs
will
take
the
input
how
they
contact,
but
still
this
is
hard
for
them.
Hope
if
you're
doing
the
work
here,
some
framework
like
here
you
got
the
framework
of
this
framework.
Api
is
already
defined
the
question
swirling
that
context
object
and
pass
you
into
the
framework.
B
You
have
no
answer
unless
the
framework
gates
your
way,
so
people
start
think
about
how
should
I
have
a
generic
way
of
putting
something
somewhere
and
it
hurt
be
able
to
achieve
that
deep
in
my
workflow
or
the
how
static
and
still
be
able
to
retrieve
the
same
value-
and
the
answer
is
very
simple:
you
can
use
a
global
variable.
You
can
put
some
sense,
a
global
variable
and
later
receive
that.
B
But
the
downside
is
obvious
because
it's
global
variable,
whatever
it
could
air,
if
you
have
another
workload,
trying
to
access
and
override
the
global
variable,
since
we
got
through
that,
so
you
see
that
global
variable
is
not
going
to
work
unless
you're
running
a
single
workload
each
time
in
your
application
and
that's
where
people
start
eating
house
right:
local
starvation.
So
that's
where
the
term
comes
from.
This
is
and
I
know
it's
probably
like
20
years
ago,
when
every
single
work
load
is
kind
of
a
dedicated
this
rad.
B
So
things
are
very
straightforward,
so
you
cut
some
work,
you
create
a
new
thread,
you
take
the
work
and
do
whatever
you
want
and
after
you
finish,
the
work
you
kill
that
thread
in
this
way.
You
can
imagine
when
you
start
a
new
thread.
It
put
something
associated
with
the
trial
and
that's
normally
backed
by
the
city
register.
B
Read
hell:
this
is
my
particular
context.
Object
associated
with
this
thread.
This
way,
that's
right
is
able
to
get
value
whenever
it
wants.
So
you
don't
have
to
pass
that
dummy
any
of
the
explicit
manner
and
if
you
have
multiple
threads,
each
of
them
will
have
a
context.
Object
believer
as
the
season
becoming
more
complex.
Thread-Local
storage
is
not
working
because
people
are
paw-riffic
consumption.
That
every
thread
is
only
used
for
paper
tasks,
so
you
have
thread
pool
where
rather
share
you
take
the
thread
to
some
work
and
later
the
thread
can
be
taken.
B
B
B
M
B
Plan
that
I
need
to
explain
the
contacts
API.
First,
okay,
okay,
it's
like
a
dictionary
interface,
which
allows
you
to
set
something
a
little
retrieve
something
and
that's
associated
with
your
workflow
and
that's
what
I
refer
to
as
contacts
API,
and
that
is
exactly
the
same
thing
as
what
open
time
she
refers
to
as
contacts
idea
and
now
comes
to
the
span
context
and
whatever
that
correlation
connects
distributed.
C
B
So,
although
those
things
are
mutable,
but
the
underlying
implementation
could
be
immutable,
I
I
think
I
hear
what
you're
saying
so
for
people
who
won't
make
that
a
meatball.
This
is
more
for
concurrency
control
and
regarding
why
they
want
to
return
that
value.
I
think
this
is
proposed
by
the
golden
team,
where
they
don't
have
in
place
it
contacts
propagation.
The
only
way
they
have
is
exit
place
it
away
you
can
see
in
go.
Every
function
will
take
to
the
contacts
as
the
last
argument
right.
C
So
and
a
bottom
immutability
requirement,
another
thing
like
I
was
gonna:
ask
if
we
think
of
what's
running
in
the
app
like
how
the
app
prepares
the
object,
populates,
some
key
values,
whatever
we're
multi-threaded
and
adds
things
to
that
property.
But
then
there
has
to
be
some
SDK
code
or
something
that
actually
in
takes
that
object
and
handles
it
in
a
way
appended
to
telemetry,
and
all
this,
like.
C
My
question
is
more
like
where
they
can
creat
usage
examples.
For
example,
if
there's
some
flow
in
my
app
before
I
call
in
to
open
telemetry
SDK
I
still
operate
on
some
kind
of
header,
only
implementation
of
whatever
context.
It
is
whether
it
is
quant,
expand,
context
or
correlation
context.
Then,
at
some
point,
I'd
have
to
pass
it
on
to
my
pre-built.
Open
telemetry
is
the
key.
So
it's
like
I'm
more
interested
in
like
step
one
to
three
as
you
described,
and
then
framework
or
framework
all
and
I.
B
And
what
do
you
do
is
in
the
HTTP
server
code
you
have
the
propagator
and
the
probably
Gator
will
leverage
component
called
the
the
HTTP
formatter
and
this
one
implement
implements
the
actual
like
logic,
how
you
convert
the
header
into
phantom?
Has
this
one
takes
the
headers
and
generates
the
span?
Contacts
structure
and
span
connects
is
just
a
structure,
and
once
you
got
that
thing,
this
is
the
the
propagator
is
doing.
B
With
the
current
contacts,
so
there
should
be
some
like
contacts
API
cop
contacts
like
set
value,
and
this
should
be
the
current
span.
Compacts,
and
this
is
the
data
structure
you
have.
This
is
the
data
you
have
and
after
that
the
work
is
done
and
all
the
subsequent
propagation,
for
example.
Here
you
use
the
tracer,
don't
create
a
new
span.