►
From YouTube: 2021-09-16 meeting
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
B
B
I
see
it's
three
of
us
right
now.
This
group
is
the
easy
one
to
convince,
maybe
as
long
as
we're
here
we
should
start.
B
I
was
hoping
to
bring
a
final,
essentially
a
presentation
or
proposal
to
the
open,
telemetry
sig
spec
sig
on
tuesday,
and
so
I
was
kind
of
trying
to
run
up
run
down
all
the
final
like
questions
that
the
decisions
that
people
want
made
bogdan
has
been
listing,
some
of
them
as
he
came
in
to
finally
review
168.,
and
I
don't
really
expect
him
here
today,
so
we
can't
actually
debate
him,
but
he
raised
an
interesting
question
and
I
think
we
should
have
our
minds
made
up
ourselves
about
it.
B
Sometimes
I
always
take
the
easiest
route
of
trying
to
negotiate
these,
these
sort
of
like
complex
group
decisions,
but
in
this
case
I
think
I
may
have
led
the
audience.
So
it's
about
the
zero
value,
and
I
mean
the
zero
value
for
p,
which
we've
currently
got
expect
as
an
unknown,
and
the
question
has
comes
in
several
forms.
One
is
if
you're
going
to
use
trace
state
to
encode
some
information.
Then,
while
the
span
already
has
a
trace
state
field,
we
already
spec
that
you
will
store
the
trace
state.
B
So
in
theory
you
could
capture
the
trace
state,
store
it
and
then
use
it
to
reconstruct
p
and
r
and
so
on,
and
that
means
we
wouldn't
actually
need
log
head,
adjusted
count
on
the
span.
We
could
just
restore
it
from
trace
state.
That
makes
me
nervous.
I
have
commented
on
why
we
are
hoping
I
am
hoping
that
the
w3c
chase
parent
would
accommodate
this
this
fight
or
so,
and
that
we
wouldn't
need
trace
date.
So
that's
my
first
present
like.
Why
would
we
have
our
dedicated
field
here,
but
it
doesn't
answer
directly.
B
Why
should
we
use
an
offset
of
one
to
accommodate
a
zero
value
when
there
are
several
ways
we
could
work
around?
That
one
has
been
discussed
recently
is
to
use
protocol
buffer
field
presence.
That
means
an
option
in
the
protocol
compiler
that
literally
tells
you
whether
there
was
a
field
or
not,
so
that
you
can
tell
the
difference
between
zero
present
and
zero
not
present.
B
If
we
adopt
field
presence
or
some
other
means
of
knowing
the
difference
between
present
and
non-present,
then
I
can
rewrite
all
these
presentations.
All
these
proposals,
by
like
cut
off
by
one
error
again
and
zero,
becomes
2
to
the
negative
negative
0,
but
just
to
count
so
it's
1.,
and
then
there
will
still
be
questions,
and
I
and
I'm
I
feel
like
each
vendor-
has
their
own
position
on
this.
But
if
nothing,
if
nothing
is
configured,
there's
no
sampler.
B
Do
you
want
an
absent
field,
or
do
you
want
to
know
zero,
which
means
one
or
whatever
it
is
like?
Do
we
want
to
know
the
difference
between
unknown
and
one,
and
do
we
want,
which
is
essentially
saying
that
if
there's
no
sampler
are
we
going
to
set
a
new
field
to
say,
there's
no
sampler
and
when
I
say
no
sampler,
what
I
mean
is
the
default
which
is
always
on
so,
if
you're
always
on,
should
you
record
p
equals
zero
or
p
equals
one
or
whatever
we're
deciding
to
do?
For
that?
B
I
and
and
my
answer
that
depends
on
the
other
defaults,
because
my
most,
the
objective
is
to
be
able
to
count
spans
and
if
one
is
ambiguous,
it's
a
problem
still.
So
I
was
hoping
to
maybe
just
propose
this
or
present
this
issue
about
the
field
presence
and
the
zero
value
being
an
offset.
B
B
B
B
Issues,
but
I
don't
think
we
decide
here,
we
we
present
on
tuesday
any
thoughts
on
that
one.
A
A
Okay,
so
the
reason
for
having
zero
meaning
unknown
is
what
I
thought
is
that
it's
some
kind
of
default
value
right.
So
if
it's,
if
it,
I
don't
know
how
it's
encoded
or
transported,
but
if
it
if
value,
is
missing,
maybe
it's
by
the
fourth
zero.
So
this
would
be
the
only.
B
A
So
if
we
have
other
means
to
detect,
if
it
was
explicitly
set
or
not,
then
I
think
I
would
prefer
starting
meaning
zero
means,
a
sampling
rate
of
100
percent
and
and
yeah,
because
it's
easier
understandable
and
this
offset.
C
B
Then
my
let
me
refine
the
question:
if
zero
is
going
to
be
the
default
and
I
and
I'm
sending
you
a
span
from
a
pre,
a
library,
that's
already
written
that
doesn't
know
about
our
spec,
it's
missing
the
field,
so
you
can
see
it's
missing
the
field,
but
you
can
see
it's
got
a
schema
version.
That
tells
you
it's
an
older
version.
Maybe
so
would
you
want
to
upgrade
that
to
a
unknown
or
upgrade
that
to
a
one?
Is
my
question
essentially.
A
It
should
be
an
unknown,
but
for
for
the
r
value
for
the
p-value.
B
A
A
It's
only
relevant
for
parent-based
samplers
right
if
you're,
using
a
tracer
d
ratio
based
sampler,
you
do
not
use
the
the
the
the
p-value
right.
So
you
only
rely
on
the
r-value.
B
This
bears
some
some
writing,
but
I
don't
think
it's
going
to
help
to
discuss
it
here.
The
it
was
a
related
question
there.
I.
A
Mean
it
also
depends
on
how
the
sampled
flag
is
interpreted
if
only
the
sample
flag
is
is
coming
and
you
do
not
have
a
valid
p
value,
then
there's
also
a
question
how
to
deal
with
that
or
if,
if,
if
the
sampling
or
the
invariant,
I
was
talking
about
that.
If
you
have
pr
and
the
sample
flag,
that
could
be
a
contradiction.
A
B
B
If
sampled
is
set
the
legacy
behavior,
then,
when
not
set
with
regards
to
a
missing
p,
oh
yeah,
that's
something
I'm
just
kind
of
trying
to
skip
what
I
just
heard.
That'll
make
sense
to
me
at
least
a
lot
more.
I
wanted
to
ask
next,
if
my
explanations
in
this
otep
175,
which
updates
170,
have
finally
come
to
a
place
where
we
agree
with
each
other.
B
On
the
same
words,
I
think
I
I
didn't
actually
respond
to
something
you
wrote
and
I
and
I
wanted
to
now
to
acknowledge
that
I
got
it.
It
was
this
statement
about
unknown,
there's
lots
of
different
uses.
We
have
for
unknowns
and
there's
actually
different
uses
for
zeros
that
we
have
that's
what
we're
facing
and
one
of
the
unknowns
that
we
have.
B
That
hasn't
really
been
discussed
here,
because
I
don't
really
want
it
in
our
system
for
lifesteps
perspective,
it's
not
a
good
outcome,
but
it's
where
the
head
defers
its
decision
and
then
the
parents
all
go
unknown,
and
then
you
assemble
your
trace.
The
parent's
unknown
becomes
a
look
up
your
root
and
then
the
root
deferred
decision,
but
eventually
wrote
it
out,
wrote
out.
The
log
had
to
just
count
using
the
kind
of
reservoir
sampling
that
you've
described,
and
this
is
starting
to
talk
about
tail
sampling.
B
A
Jail
sampling
is
something
completely
different,
so
what
a
risk
war
sampling
is
that
you
do
not
know.
If
this
ban
survives
the
reservoir
or
not
so
I
mean
this
is
not
compatible
with
with
head
based
template
parent
based
sampling,
which
I
wrote
so,
but
if
you
are
just
relying
on
tracy
ratio
based,
samplers
and
and
or
maybe
reservoir
samples,
you
can
do
that
in
a
consistent
way.
A
In
this
case,
you
do
not
need
the
sampled
flag,
nor
you
need
the
the
head
probability
right.
So
it's.
A
You're
not
using
any
hat
based
sampling
decisions,
so
if,
if
you
just
use
traceability
ratio
based
samplers,
which
determine
the
sampling
rate
by
their
own-
or
maybe
they
even
use
reservoir
sampling,
so
they
adapt
the
sampling
rate
based
on.
What's
what's
the
incoming
traffic
during
a
minute,
for
example,
yeah
and
and
then
you
do
not
need
the
parent
probability,
and
you
can
also
ignore
the
sampled
flag
right.
I
get
it
no.
B
It
makes
sense
trying
to
figure
out
if
I
should.
I
feel
like
this
is
a
conversation
that
interests
me
and
yet,
if
I
left
it
out,
no
one
would
ask
questions
about
this.
So
I'm
figuring
out
like
what
to
write
if
anything
about
this,
but
now
I
understand
a
little
bit
more
fully
what
you
have
described
and
it
sounds
nice.
I
mean
it's,
it's
it's
not
tail
sampling.
I
get
the
distinction
you're
making.
I
like.
A
It
but,
but
you
have
to
do
it
yeah
still
where
this
the
span
is
processed,
so
you
do
not
send
it
to
a
collector
where
some
sampling
decision
happens,
it's
still
very
close
to
the
span,
but
the
sampling
decision
is
not
made
when
the
span
is
opened.
So
you
cannot
propagate
this
the
sampling
decision
yet
so,
but
if
you're,
just
using
tracid
ratio
based
samplers
or
just
reservoir
samplers
everywhere,
then
you
do
not
need
to
propagate
that,
but
you're
still
getting
full
traces.
A
According
to
this
consistent
sampling,
I've
described.
B
Yeah,
no,
I
get
it.
That's
that's
quiet!
Yeah!
It's
interesting!
I
didn't
see
all
the
way
to
the
end
of
that
story
when
I
got
the
algorithm,
but
I
wasn't
seeing
how
you
do
it,
but
that
roughly
yeah
it's
like
consistent
tail
sampling,
it's
kind
of
cool.
I
like
it.
A
It's
the
tail
of
what
they
are,
because
I
say
it's
not
the
it's.
Basically,
the
the
sampling
decision
is
made
when,
for
example,
the
period
is
over
right
when
you're,
when
you
take
yeah.
B
Yeah,
okay!
Well
anyway,
I'm
trying
to
not
respect
that
tale
anyway.
So
it's
great
the
thing
I
I
want
to
check
if,
if
I'm
on,
if
you're
on
the
same
page
with
me
on
this,
this
stuff,
which
is.
A
But
for.
B
The
record
all
this
stuff
on
the
top
is
really
trying
to
just
lay
out
the
glossary
for
the
spec,
because
we're
gonna
need
it,
and-
and
I
didn't
mean
to
have-
but
I
did
put
in
these-
are
the
four
kind
of
common
examples
that
come
to
mind
of
a
non-probability
sampler
that
people
talk
about
that
are
just
like
full
of
bias,
sort
of
by
definition,
and
that's
why
I
think
of
them
that
way,
and
then
what
I'm
I'm
still
feeling
pretty
strong
about
is
that
this
is
that
we
have
sort
of
three
states
of
a
p-value.
B
It's
the
the
unknown,
the
zero
and
the
the
non-zeros,
and
there
have
been
requests
to
talk
about
composition,
rules
and,
and
I
think
that
we
can
take
some
of
the
ones
that
we
know
about
the
non-probability
samplers
and
compose
them.
With
these
consistent
probability,
samplers.
B
In
a
way
that
gives
users
who
are
familiar
with
these
existing
things,
like
a
leaky
bucket
or
like
a
completely
arbitrary
sampler,
just
keep
doing
what
you're
doing
and
then
it
will
at
least
not
interfere
with
what
the
accounting
that
we're
trying
to
do
and
if
the,
if
the
user
also
wants
to
count,
they
can
also
compose
it
with
a
probability
sampler
and
get
the
results,
and
the
the
biggest
key
here
is
that
it
just
this
encodes,
the
rule
that
you
had
stated
as
well
as
the
rule
that
I
had
stated,
and
this
is
that
minimum.
B
B
If
you
have
two
unknowns,
you
could
keep
it
unknown
and
if
you
have
one
unknown
and
one
not
unknown,
you
just
take
the
p
value
that
you
know
and
you
and
you
combine
the
sample,
and
that
leads
to
the
situation
that
gives
you
known:
zero,
adjusted
count,
which
is
really
just
my
mechanism
that
I'm
trying
to
give
that
lets
us
record
spans
without
counting
them
in
case
they
want
to
use
some
sort
of
auxiliary
signal
that
lets
you
collect
these
spans
that
are
not
sampled.
A
And
I
have
to
read
through
that,
but
I
had
the
question:
why
do
you
wanna
report
a
zero
adjusted
count
or
on
the
span?
So
what's
I
mean
there's
also
unknown
right,
so
I
mean
I
mean
I
understand
to
propagate.
Maybe
the
p
value
so
and-
and
you
know
the
log
adjusted
count
takes
the
p
value,
but
I
am
not
sure
if
the
span
should
finally
end
up
with
a
zero
adjusted
count.
So
I
would
I
do
not
see
the
use
case
yet,
but
I
have
to
read
through
that.
B
I'll,
try
I'll
show
you,
let's,
let's,
let's
hope
that
that's
this
text
works
and,
if
not
we'll
take
it
to
the
review
it.
It
is
really
trying
to
the
the
use
cases
that
we've
confirmed
are
something
like.
B
That's
that's
where
it
becomes
not
probabilistic,
so
you
so
at
the
moment
you
see
a
span.
You
can
make
a
probability
decision
and
the
probability
probability
decision
just
says
no,
no,
no
every
time
you
finally
say
I
just
need
to
record
this
anyway,
even
though
the
probability
was
no
that's
a
zero
and
it
helps
people
with
these
algorithms
that
need
to
see
all
the
information
when
the
when
the
sample
is
not
working,
essentially
so
to
help
with
too
low
sampling
rates.
People
want
to
be
able
to
just
throw.
A
B
A
A
A
If
it's
like
that,
then
it's
fine
here,
but
I
thought
you
agreed
that
using
multiple
samplers
on
the
same.
B
Side
I
I
do,
but
I
I
do,
but
I
mean
it
in
a
more
qualified
way,
and
now
I've
pulled
up
the
next
topic
just
to
help
you
see
where
this.
The
other
conversation
points
are
so
so
nikita
who's,
a
trace,
approver
and
a
generally
smart
guy
wrote
this
comment
saying
well,
this
didn't
answer
all
my
questions
so
what's
going
on
here,
essentially
and
and
the
question
that
he
put
up
in
this
issue
is
about
composing
samplers.
Now
I
think
that
this
is
not
about
sampling
anymore.
B
This
is
about
what
I'm
calling
view
configuration
and
I've
I've
written
that
in
a
lengthy
comment
here,
but
it
is-
and
this
is
so
complicated.
I
really
have
I'm
just
like-
don't
even
want
to
engage
this
conversation
because
we're
so
into
a
technical
corner
right
now,
but
I've
from
the
beginning
of
this
process
been
talking
about.
I
have
tried
to
avoid,
or
or
at
least
points
talked
about,
and
now
I'm
trying
to
avoid
talking
about
multiple
samples.
B
So
I
have
two
independent
samples,
and
I
I
know
what
that
means,
and
I
know
how
to
do
it,
and
I
know
what
I'm
by,
when
I'm
stuck
what
I'm
like
signing
up
for.
When
I
talk
about
that.
But
the
decision
to
record
a
span
there's
always
still
only
one
span
and
so
there's
some
differences
that
are
emerging
between
metrics
and
spans.
And
what
is
that
difference
and
the
difference
is
that
metrics
are
aggregated
and
spans
are
not
essentially.
B
But
when
you
start
to
sample
sampling
is
an
aggregation.
So
we've
talked
about
in
metrics.
Why
would
you
ever
want
more
than
one
view?
It's
because
you're
applying
different
aggregations
to
the
same
event
stream
in
spans,
there's
just
a
single
span,
but
you
could
imagine
applying
more
than
one
sampling
scheme
to
a
span,
and
this
would
result
in
more
than
one
way
to
choose
a
span.
B
Rules
document
is
about
specific,
is
about
combining
things
that
are
not
probability
with
probability,
because
it
leaves
us
with
just
one
probability,
which
is
an
important
case
that
I'm
done
after
and
yet
what
we
want
is
essentially
a
way
to
provide
what
users
are
asking
for
in
this
ticket
here
is
essentially
a
way
to
configure
which
stands.
I
see
when
I
run
my
tracing
library,
which
is
like
configuring
views,
so
I've
just
argued
well
in
metrics.
The
reason
to
configure
more
than
one
view
is
that
you
have
more
than
one
aggregation
in
spans.
A
B
A
B
This
is
this
field
that
we're
speccing
out
now
log
head
just
count:
how
to
just
count
is
only
for
the
entire
population,
which
is
meant
you're
not
supposed
to
apply
filters
to
this,
so
that
that's
why
I
think
there
should
only
be
one
of
them
and
it's
okay
for
us
to
proceed
now
with
just
one
sample,
because
any
anything
more
than
that
is
asking
for
this
tremendous
amount
of
complexity,
and
it
requires
setting
up
filters
in
addition
to
dealing
in
multiple
probabilities.
B
So
that's
what
was
being
asked
for,
and
I
basically
responded
that
this
is
not
a
question
about
composing
samplers.
This
is
not.
This
is
more
about
defining
one
fancy
sampler.
So
this
is
the
kind
of
logic
that
people
want,
and
this
is
what
jager
sampling
does
is
essentially
says.
You
know
in
in
in
sequence,
I'm
going
to
run
a
bunch
of
predicates.
The
first
predicate
that
matches
will
determine
my
probability.
It's
either
going
to
be
zero.
If
I'm
ignoring
expanse
or
it's
going
to
be
some
non-zero
probability
and
the
I
mean.
A
Sample
implementation,
which
exactly
has
this
e-files
and.
B
And
my
point
here
is
that
we're
down
to
a
configuration
problem
now
we're
not
we're
not
dealing
with
sampling
problems
anymore
and
configuration
is
the
biggest
problem
that
open
telemetry
has
today.
So
at
least
we've
reduced
this
to
the
common
problem
and
that's
why
I
think
that
we
don't
need
to
talk
about
combining
samples,
multiple
samples
people
are
probably
asking
for
view
configuration
and
in
the
long
run
it
makes
sense,
but
the
day
that
we
see
multiple
head
probabilities
and
multiple
current
samples-
I
don't
know
it's
like
really
far
out
there.
B
I
don't
care
so,
okay,
I've
talked
through
my
issue
and
we
talked
through
your
valid
use
for
unknown,
which
I
commented
maybe
I'll
write
somewhere
building's
here
now
and
I
want
to
take
his
attention
since
we've
talked
about
field
presence
and
he
raised
it.
Oh
I
unshared
my
screen.
I
meant
to
go
back
to
the
other
document,
hi
bogdan,
so
thank
you
for
your
review
on
168.
It
was
one
of
the
more
useful
pieces
of
feedback
we've
gotten.
B
I
tried
to
respond
to
the
highest
level
point
there,
which
is
like
why?
Why
store
a
field
when
you've
got
trace
date?
And
I-
and
I
hope
I
convinced
you
that
trace8
is
a
really
terrible
solution
for
us
and
would
require
users
to
parse
two
things
and
what
I,
what
I
think,
I'm
hoping
the
outcome
here
is.
B
So
I
think
sticking
to
an
inefficient
solution
is
good,
and-
and
that
means
one
where
that
we
can
change
later.
So
the
idea
that
we
could
not
use
trace
state
in
the
future
is
appealing
to
me,
but
you
ask
all
kinds
of
good
questions
independently
like.
Why
are
we
saving
trace
state
but
we're
not
saving
trace
parent?
Well,
every
field
in
the
transparent
is
saved,
except
for
flags.
We
don't
save
the
flag.
So
if
w3c
traceparent
added
a
flag,
we
wouldn't
have
a
place
to
put
it
in
our
span
data
right
now.
B
So
maybe
we
should
be
storing
trace
parent-
and
I
don't
know,
and
but
but
but
in
any
case
it
seems
as
a
data
model.
I
think
people
are
probably
going
to
be
writing
spans
from
other
places
that
won't
be
sdks,
that
the
hotel
writes
just
having
a
field
in
your
protocol
where
you
say
this
is
how
much
you
count
would
be
a
nice
thing.
B
So
I
I
think
I
would
still
argue,
even
if
we
had
trace
state
or
trace
parent
stored,
that
that
this
belongs
in
its
own
field
and
then
I
think
it
would
make
sense
to
erase
the
information
from
the
span
that's
redundant.
So
we
could
take
the
hotel
trace
state
out
of
the
tray
state
list
before
we
store
the
span.
For
example,
since
every
field
of
the
hotel
trace
state
is
specified
within
some
way,
we
can
just
use
it
as
appropriate
and
then
erase
it.
F
B
F
Correct
the
reason
why
I
say
this
is
once
extra
information
is
added:
we
will
have
for
free
that
extra
information
without
changing
anything,
even
if
you
are
coding
against
the
previous
version
of
the
flag,
if
you
record
it,
you
record
a
new
value,
so
you
are
fine,
it's
gonna
just
work.
I
think
it's
the
right
thing
to
be
backwards,
compatible
based
on
the
definition
of
w3c,
which
says
that
adding
will
not
increase
the
version.
So
we
should
add
the
flags.
F
Second,
one,
the
second
one
with
the
log
head,
adjusted
the
account
my
point
just
for
until
we
determine
better,
I
would
not
add
something
into
this
stable
data
model
because
by
default,
if
you
we
have
the
three
state
already
so
the
information
is
already
there
is
before
we
do
the
simplification
that
you
want
to
remove
and
compute
this.
We
can
easily
have
the
end-to-end
working
testing
and
everything
we
no
changes
to
the
proto.
B
E
B
F
Correct
and
maybe
that
would
be
named
blah
blah
blah
and
I
don't
want
to
do
backwards,
incompatible
changes.
My
point
is
limited.
The
the
the
api
changes,
because
still
that's
an
api
change
from
from
the
protocol
perspective.
I
I
would
limit
that
as
much
as
we
can
to
make
progress
again,
not
want
to
block
the
the
performance
later,
but
I
think
I
think
the
proposal
that
you
have
to
to
extract
this
information
into
a
separate
flag.
F
It's
backwards
compatible
so
or
can
be
implemented
backwards,
compatible,
makes
sense
like
yeah
yeah,
and
I'm
thinking
more
like
an
engineer
here
like
not
like
necessarily
related
to
to
the
sampling,
seek.
B
Yeah,
so
the
advice
is,
we
don't
need
to
add
a
field
for
now
as
long
as
it's
for
a
trace
date
and
we
don't
erase
the
trace
date
or,
as
part
of
some
future
w3c
field
that
we
decide
to
store,
then
we
get
this
sort
of
for
free.
I
think
I,
like
the
advice.
B
Another
thing
I
like
about
this
just
to
I'm
thinking
of
all
the
things
we've
put
off
is
well
earlier
in
my
otep,
I
had
specked
out
the
use
of
a
span
attribute
for
for
sampling
information,
and
that
was
also
not
going
to
require
a
protocol
change.
It
was
also
something
that
I
liked
because
it
gives
me
field
presence
for
for
by,
but
for
free.
B
I
know
whether
there's
an
attribute
or
not,
and
it
also
gave
me
a
way
to
encode
multiple
samples
like
I
could
just
sampler.name.adjustedcount,
and
then
I
can
have
four
of
them.
If
I
want,
which
gives
me
a
path
to
tail
sampling
and
all
that
other
stuff,
there
was
a
point
at
which
yuri
saw
what
I
was
doing
and
proposed.
B
Why
don't
you
just
I
mean
like
why
just
make
a
field
for
this,
because
because
I
I
was
also
proposing
at
that
time,
to
put
a
name
in
that
would
be
like
the
name
of
the
sampler.
As
we
have
now
simplified.
This
is
just
one.
Global
sample
doesn't
need
a
name,
so
we
need
this
optional,
integer
and
yeah
you're
right,
so
field
presence
is
for
free
if
we're
counting
on
trade
state
or
some
other
optional
information
from
from
w3c.
B
Okay.
I
think
I'm
convinced
that
that's
the
easiest
step.
It's
not
the
nicest
outcome
for
the
consumer,
but
they've
got
people
working
for
them.
So
that's
cool.
F
B
F
C
B
All
right,
I
think
that
that
is
practically
it's
not
what
I
want,
but
it's
practically
good
advice,
so
I
will
accept
it.
So
this
is
a
proposal
where
we
can
get
this,
so
we
would
so
carlos
has
a
pr
on
the
spec
about
how
to
use
trace
state.
B
Who
would
need
that
my
otep
number
170
would
be
revised
to
say,
there's
no
field,
it's
just
we're
going
to
store
this
in
trace
state
or
somewhere
else
in
the
future
tbd,
and
it
means
the
same
thing
so
go
looking
inside,
trace
state
for
a
p
value.
If
you
want
your
head
adjusted
count.
B
The
corner
cases
are
coming
out.
I
don't
think
we
should
discuss
them
here
in
the
moment,
but
when
you
s,
when
you
do
record
the
incoming
trace
state
or
the
outgoing
trace
state,
the
result
of
your
sampler
gets
sort
right,
because
the
trace
id
ratio
sample
has
to
return
its
own
p
value
to
be
stored.
F
We
record
the
tray
state
of
the
after
the,
as
that
is
the
result
of
the
start
span.
So,
okay,
nothing
happens
during
the
start
span
that
will
be
recorded,
not
the
parents
one.
So
so
it
will
be
recorded,
the
ones
that
you
produce
okay,
yeah,
because.
F
B
All
right,
I
think
this
is
very
helpful
advice.
It's
good!
It's
good
advice.
I
appreciate
it.
I
mean
I
work
on
it.
The
other
thing.
F
B
F
B
The
change
you've
just
suggested
is
in
the
one
I
already
merged,
so
I'm
going
to
I'm
going
to
edit
it.
I
in
170
about
logcat,
adjust
account
and
then
now
you're
talking
about
168,
which
is
the
propagation
stuff
go
ahead.
F
Yeah,
the
propagation
part
for
me
was
unclear
if,
if
a
system
that
doesn't
need
to
do
any
kind
of
propagation,
they
do
full
always-on
or
other
mechanism
that
does
not
require
these
complicated
things
because,
for
whatever
reasons,
imagine,
for
example,
the
way
how
google
had
it
internally,
they
had
a
guarantee
that
trace
id
is
completely
unique,
random,
sorry,
completely
random.
They
could
use
that
simple
super
simple
algorithm
without
having
to
propagate
too
many
things
correct.
F
So
that
being
said,
I
want
to
facilitate
that
google
can
use
this
or
us
a
splunk
where
we
right
now
we
don't
do
sampling,
and
this
will
not
affect
these
users.
That's
that's!
That's
the
only
thing
that
I
was.
F
It
was
not
clear
to
me
from
that
attempt
that
I
would
have
to
deal
with
this
and
and
and
and
interpret
this
in
specific
ways
or
or
even
errored,
or
something
if,
if
I'm
not
participating
or
again,
it's
a
it's
a
very
slippery
bound,
because
because
I
know
if
the
system
that
works
with
say,
google
communicates
with
the
system
that
works
with
with
somebody
that
uses
this
information,
there
has
to
be
a
minimum
transfer
of
data
between
them.
F
B
B
I
wish
we
had
w3c
already,
so
I
I
would.
I
would
one
idea
that
I'd
like
is
just
have
another
flag.
That
says
all
my
bits
are
random,
and
that
means
you
don't
need
an
r
value.
If
you
have
that
flag
set,
because
you
can,
if
you
count
the
leading
bits
in
the
right
order,
you
can
compute
the
r
value
deterministically
from
the
trace
id.
B
So
that's
one
thought
it
sounded
like
you
wouldn't.
F
F
Of
them
or
63,
maybe
16
64
63,
the
last
63,
the
right
63
in
yeah
in
egyptian
order.
Yeah
are
the
one
random,
for
example.
Let's
assume
these,
then
we
have
that
bit.
I
think
this
is
a
reasonable
change,
which
will
be
much
easier
to
accept
by
wgc.
B
B
B
F
F
B
Is
the
upper
limit
for
us
because
of
java
we
were
at
62
because
we
wanted
to
use
one
value
in
this
expense
and
so
on,
but
it's
either
62
or
63,
not
more.
Okay.
I
was
trying
to
understand
beyond
that
sort
of
what
I
see
here
is
an
optimization,
I'm
trying
to
figure
out
what
you
think
about.
Like
you,
you
don't
care
about
this
feature,
I'm
just
like
playing
devil's
advocate.
You
have
nothing
to
do
with
this
feature.
B
You
are
the
root,
though,
and
yeah,
and
we
would
like
it
if
you
could
inject
this
r
value
so
that
your
children
could
make
a
consistent
sampling
decision.
Let's
say
if
you're
not
generating,
if
you're
and
you're
and
basically
the
two
the
two
positions
are
either
you
have
randomness
and
you'd
rather
not
generate
an
r
value
or
you
have
randomness
and
you'd
rather
infer
the
r
value
or
you
don't
have
randomness
and
you
need
to
inject
the
r
value
or
something
like
that,
and
I
just
too
many
outcomes
here.
B
B
If,
if
the
trace
id
is
randomness
has
enough
randomness,
then
we
we
don't
need
an
r
value.
If,
if
this
trace
is
unsampled,
we
don't
need
a
p
value.
It's
when
we
don't
have
randomness,
we
need
an
r
value,
and
when
we
are
traced
we
need
a
p
value.
A
A
If
you're
using
a
consistent
sampler,
then
the
consistent
sampler
determines
the
the
sampling
probability
by
its
own.
So
it
knows
what's
the
adjusted
count,
but
the
parent-based
sampler
uses
the
sampling
decision
of
the
parent,
and
this
sample
is
not
aware
of
the
sampling
probability
of
the
parent
unless
it's
propagated.
So
that's
why
we
are.
We
need
to
propagate
the
p-value
for
this
case,
so
it's
needed
if
the
parent
is
sampling
and
you're
using
a
parent
based
sampler.
F
At
least
at
least
for
example,
okay,
so
so
you
are
making
p
required.
So,
let's
assume
w3c
accepts
the
beat
for
the
last
32
to
be
random,
63
to
be
random.
You
still
require
p.
B
B
B
F
Yeah
but
it
needs
by
everyone,
because
otherwise
things
are
not
working,
so
it
kind
of
becomes.
We
don't
know
if
you
are
leaving
your
system.
So
let's
assume
you
are
in
a
bubble,
but
if
you
are
leaving
and
talking
to
aws
or
other
services
that
that
uses
something
like
lystep
which
needs
this,
then
to
be
a
nice
player
in
the
community,
we
we
need
to
to
to
set
the
the
p-value
if
we
are
sampled,
because
otherwise
we
will
not
be
nice
with
the
entire
ecosystem.
Let's
put
this
way.
B
Okay,
that's
about
right.
There
seem
to
be
two
essentially
questions
about
opt-in
and
opt-out.
One
is
for
r
one's
for
p
and
we
think
the
r
question
can
be
swept
away
because
hopefully,
one
day
we
just
have
randomness
and
we
don't.
We
don't
have
to
worry
about
it,
but
the
p-value.
It's
like
only
needed
if
you're
sampled,
but
it
is
needed
if
you
want
to
count
and
so
and
it
has
to
be
set
by
the
trace
id
ratio
sampler
in
some
parent
and
so
the
question.
Essentially
the
trace
80
ratio.
B
Sampler
then
begins
having
two
options.
One
is,
do
you?
Do
you
always
set
and
propagate
r,
and
one
is,
do
you
always
set
update
p?
If
sampled
p
have
sampled
always
r
and
I
think
yeah,
you
could
even
have
two
different
opt-ins,
but
I
think
just
one
is
probably
most
sensible.
B
Lightstep
will
accept
opt-in
because
we
don't
want
to
fight,
we
just
want
it
done
and
we
we
tell
customers
how
to
configure
their
tracers.
It's
not
a
big
deal
for
us
to
have
this
off
by
default,
but
I
think
hopefully
it's
just
one
configuration
item
and
I
think
I'm
I'm
leaning
towards
just
trying
to
say
that
this
is
temporary
we'd
like
to
get
w3c
changed
and
then
and
then
it
won't
cost
you
anything
except
a
couple.
Few
bytes
when
you're
sampled,
I
see.
B
So
I
don't
know
the
way
this
is
done
in
the
ghost
sdk
is
the
only
one
I
can
talk
about
and
the
the
the
id
is
generated,
and
then
you
call
the
sampler
and
the
sampler
can
tell
whether
there's
no
parent
context,
it's
a
new
route
and
if
you're
a
new
root,
you
generate
your
r
value
and
attach
it
to
the
trace
context,
but
you're
you're
you're,
hinting
at
a
future
difficulty
that
will
come
out
as
soon
as
we
try
to
improve
with
w3c,
which
is
that
the
is
that
the
trace
parent
depends
on
the
sample
decision,
but
you're,
but
like
the
way,
the
code's
organized.
B
You
generate
the
trace
context
before
the
sample
decision,
and
I
think
it
will
call
for
some.
I
don't
want
to
even
specify
change
to
the
sampler
because
I
feel,
like
you
want
different
call
signature
when
you're
the
root
than
you
do
when
you're
the
child.
But
I
I
this
this
level
of
detail
to
me
is
an
sdk
spec
problem
that
stands
sort
of
separately
from
the
data
model
for
spams
or
for
the
propagation
context
stuff.
F
No,
I
mean
we
can
make
things
we
we
just
need
to
connect
the
trees
id
generator
with
the
sampler
to
see
if
we
are
a
root
as
the
trace
id
generator.
How
how
many
sample
bits
do
you
have
essentially
right
that
yeah?
That's.
F
Oh
go
ahead.
I
have
solutions
for
this,
but
I'm
not
I'm
not.
Can
I
propose
everything,
but
do
you
want
to
propose
the
beat
the
633
beat
to
the
w3c?
I
want
to
have
that
ball
going.
We.
B
Both
atmar
and
I
have
appeared
at
a
w3c
meeting,
I
more
or
less
gave
an
overview
of
everything
we
were
talking
about
and
it
included
the
r
value
and
the
p
value,
because
both
of
them
suggest
changes
in
the
trace
parent.
A
I
explained
them
basically
options
how
to
to
to
have
shear
randomness,
which
is
needed
for
consistent
sampling,
but
I
shared
the
slide
on
the
slides
on
the
on
the
slack
channel
but
which
it
didn't
mention
is
that
it's
possible
to
have
a
trace
id,
which
is
just
partially
random.
So
maybe
this
is
an
option
for
them.
A
fourth
option
which
I
didn't
mention.
So
I
don't
know
what
you're
thinking
about.
F
Officially,
unofficially,
or
whatever,
I'm
not
necessary,
because
I
part
of
well
and
it's
anyway,
it's
different,
but
I
have
influence
on
that
and
I
can
help
so
what
I
would
propose
is
the
following:
I
think
we
should
use
the
next
beat
for
the
following:
one
means
guaranteed
the
last
or
whatever
beats
we.
We
we,
we
choose
the
beats
that
will
not
affect
the
the
x-ray
the
amazon
x3,
because
I
think
they
use
the
first
a
couple
of
bytes
for
for
for
the
timestamp,
so
they
are
not
random.
B
F
I
would
do
that.
I
would
so
let's
propose
this:
let's
propose
a
bit
that
the
second
least
important
beat
as
being
if
it's
zero,
it's
unknown
or
cannot
guarantee
any
randomness.
Essentially
not
not
unknown,
cannot
guarantee
randomness.
One
guarantees
that
the
63
bits
that
that
is
again
matches
with
the
x-ray
are
random
so
that
they
are
happy.
We
are
happy,
everyone
is
happy
and
I
think
we
can
even
simplify
everything
here,
because
then
we
can
drop
the
r
value
correct
completely
and.
B
Your
point
earlier
I
didn't
realize,
is
that
w3c
allows
adding
flags
without
revving
the
version.
So
this
is
a
simpler
change
than
the
one
I
really
want,
which
is
to
say
if
the
sampled
flag
is
set
attached,
two
extra
bytes
right
after
the
flags,
you
might
concatenate
two
more
base:
16
bytes,
but
only.
B
Yes,
I
actually
was
onto
something
new.
I
I
like
what
you're
saying
about
the
flag,
but
that
gives
us
r
to
get
p.
I
think
I
need
version
one
of
trace
parent
and
the
version.
One
extension
is:
if
the
flag
is
set
sampled
flag,
the
first
flag,
then
we
need
two
more
bytes
and
it
would
be
a
context
dependent
parsing.
Now
where
it
says,
if
the
fl,
like
anyway.
F
I
I
have
good
arguments
to
always
send
them.
We
can
chat
about
that
and
we
can
debate
about
that.
But
but
let's,
let's
get
what
we
mean
always
get
key:
okay,
yeah.
But
that's
that's
a
separate
discussion.
I
want
to
help
you
and
want
to
simplify
this.
So,
okay,
let's,
let's
make
this
proposal.
I
I'm
happy
to
write
it
if
you,
but
probably
you
have
a
better
yeah.
F
This
can
you,
you
can
slag
me
and
we
can
work
together,
how
to
propose
this,
but
I'm
super
confident
that
I
I
pushed
okay.
B
So
we're
talking
essentially
about
two
distinct
w3c
proposals.
One
is
easy,
which
is
the
first
one,
which
is
a
bit
that
says
I
got
63
bits
of
randomness.
My
trace
id
correct.
Second,
is
a
way
to
put
p
values
when
sampled
or
always,
which
would
go
in
version
one
I
like
it.
I
think
this
is.
This
is
helpful
bowdoin.
Thank
you.
F
Are
I
can
guarantee
you
that
will
happen
so,
but
we
can
move
forward
with
having
the
p
on
our
side
for
the
mobile
and
and
even
even
if
we
have
the
r,
we
can
do
even
the
the
simple
google
trick
and
give
that
trick
as
well
to
to
to
to
people
if
they
want.
So
even
google
will
be
happy
to
have
that
because
we
can
give
the
trace
id
based
thing,
because
we
have
some
random
number
and
this
p
can
apply
to
only
this
subsection
of
the
the
anyway.
What
I'm
trying
to
say.
F
I
think
this
is
useful,
not
for
you
only
for
for
this
proposal.
For
for
many
many
people
p.
I
think
personally,
I
would
not
do
the
proposal
right
away.
I
would
bake
it
on
our
side
with
a
tray
state,
as
we
have
right
now,
move
it
prove
it
more
understand
better
and
then
and
then
make
the
final
proposal,
maybe
that
power
of
two
that
we
used
is
not
good.
F
Maybe
it
is
also
w3c
uses
hex,
not
base64,
so
you
have
because
everything
is
hex.
Yeah.
D
F
But
but
again
it's
what
I'm
trying
to
say
is:
if
you
have
you,
if
you
have
six.
B
I
think
that
might
be
what
we
do
in
trace
state
that
comes
back
to
carlos's
pr,
okay,
I
think
we
should
end
this
call,
I'm
feeling
weary
I've
got
lots
of
good
information
and
good
advice,
you're
going
to
look
at
that
pr
175
for
me,
I'm
going
to
take
everything
bogan
said
and
try
and
make
it
actionable
by
tuesday.
F
Go
ahead,
the
next
specie
yeah,
the
other
thing.
As
I
said
ping
me
when
you
want
to
to
do
the
proposal,
it's
actually.
F
I
know
it
works
in
w3c
if
you
put
names
for
for
who
are
the
proposers,
so
you
can
put
my
name
and
other
who
wants
to
be
odmar
if
you
want
to
be
in
that
proposal,
it's
good
to
to
have
a
like
support.
People
of
this
proposal
like
and
I'm
supporting
from
the
beginning.
The
other
thing
that
you
discuss
I
want
to
just
do.
F
One
minute
that
we
have
left
is
the
the
proposing
the
aws
schema
for
our
trace
id
in
open
telemetry
that
we
generate
this
way
that
may
later
come
with
another
beat
to
say
that
the
first
some
kind
of
beats
represents
the
timestamp,
but
it
does
not
propose
that
yet
because
I
say
yeah,
so
some
system
may
need
to
know
if
that's
the
timestamp
or
not,
but
let's
not
get
into
too
many
things.
I
think
only
us
suggesting
that
our
trace
id
or
default
trace
id
generator
uses
the
uses
the
the
aws
schema.
F
I
think
it's
it's
a
it's
a
good
suggestion
to
have
it
spec
in
the
specs,
and
I
think
we
will
get
a
lot
of
support
from
from
the
aws
people.
They
will
be
very
happy
to
have
that
as
a
default
and
I'm
actually
thinking
that
is
actually
not
not
that
bad
to
to
for
systems
that
want
to
have
a
kind
of
an
append
only
mode
for
for
things.
So
so
this
this
time
stems
gives
you
some
ordering
of
where,
where
things
will
get
into
your
system.
Yes,
this.
B
Is
great,
I
want
to
have
a
definitive
ordering
on
my
events
that
could
be
done
in
the
trace
id
or
the
span
id.
It
needs
to
be
part
of
the
span
id
too.
B
A
A
B
D
C
Yeah,
I
think
that
when
I
came
from
vacation
you
just
left
for
vacation,
so
we
crossed
ourselves
and
last
week
I
I
went.
I
went
with
my
daughter
that
she
was
moving
also
in
london,
so
I
I
moved
to
london
with
her.
She
was
moving
her
accommodation.
So
yeah
are
you
in
london
now
or
you
were
just
no.
D
C
C
There
there
was
one
topic
I
wanted
to
talk
with
you.
It
was
about
the.
D
Oh
yeah,
let's
take
a
look
at
this
yeah.
I
saw
a
couple
of
issues
I
I
wasn't
didn't
get
a
chance
to
look
at
them
too
closely.
C
C
And
see,
what's
your
opinion
also
because
it's
it's
quite
yeah,
it
was
creating
a
program
for
for
them.
C
C
I
think
we
are
probably
old
now
I
don't
know
right
now,
one
more
will
connect.
So
we
can
start.
C
C
Last
meeting
actions
the
external
dependencies
problem,
but
I
would
like
to
talk
in
a
separate
topic
just
for
that,
because
I
think
it
it
can
be
a
bit
longer.
C
C
I
I
merged
a
pr
that
supports
for
automatic
context,
passing
using
the
local
task
feature
of
the
structure
concurrency
and
it
can
work
at
the
same
time.
Then
it
works
with
that
os
activity.
In
theory,
one
of
the
limitations
of
the
asynchronous
thing
is
that
it
doesn't
work
with
os
activity.
That
was
what
we
use
for
the
context
passing,
so
I
created
a
pr
that
supports
asynchronous
local
task
stuff
and
also
mixing
that
with
the
current
os
activity.
C
I
have
been
trying
that
with
my
stuff
and
it
was
working-
maybe
it's
not
the
perfect
solution,
because
it's
not
there
could
be
some
code
paths
that
might
not
work,
but
I
couldn't
find
any
so
we
can.
It
still
is
it's
in
development,
even
when
they
created
the
rs,
a
release
candidate
for
xcode
13,
but
yeah.
It's
working
then
yeah.
One
of
the
issues
also
with
that
is
that
the
latest
xcode
version,
the
rc1,
doesn't
include
the
concurrency
staff
for
mac
and
fails
building.
C
C
Also
yeah,
I
also
created
a
pr
that
it's
already
merged
to
fix
several
issues
with
the
url
session,
instrumentation,
mainly
with
versions
of
ios
below
13,
so
with
ios
12
11
that
it
was
supported
in
the
library
the
issues
were,
we
were
creating,
sometimes
duplicated
spans,
and
sometimes
we
we
also
weren't
responding
to
the
delegate.
If
there
was
a
session
delegate,
those
calls
were
not
being
made.
C
I
think
that's
quite
serious,
yes
and
it
could
break
the
up
running
with
that
and
also
that
was
mirror
the
capturing
the
payload
was
not
working
as
expected
for
some
of
the
codepaths
for
big
files,
for
example,
it
was
not
copying
the
the
payload
correctly
so
now
now
that
that's
also
merging
appear,
but
I
also
found
after
that
appear
that
there's
a
threadrice
condition
in
the
url
session
instrumentation.
I
created
a
pr
for
that,
but
it
still
has
to
be
approved
and
merged.
C
C
E
Oh
yeah,
no
yeah.
Actually
you
know
I
had
yeah,
I
mean
whatever
we
did
right
a
couple
of
not
last
month
I
have
it.
I
I
had
an
older
laptop,
so
I
I
you
know
commented
on
my
repo,
so
I'm
trying
to
figure
out
where
to
where
do
I
put
it?
Okay,.
C
No
no
problems,
I
mean
the
tlp
example.
Is
it's
nice
to
have
so
the
other,
for
the
tutorial
is
less,
I
mean,
show
the
end
than
the
or
the
lp
example
for
for
any
potential
user.
E
E
D
Yeah
the
service-
I
mean
it
really
depends
it
should
it
should
be
working,
I'm
pretty
sure.
Maybe
the
service
name
might
not
be,
might
not
be
getting
picked
up
by
the
environmental
variables.
E
There
is
there's
two
environment
variables
and
I
think
it
says
the
environment
variable
actually
takes
precedence
and
I,
when
I
tested
it
it
didn't
work
so.
E
D
Are
you
using?
Are
you
using
the
environmental
variable
for
the
service
name,
or
are
you
using
the
hotel
resource
attributes,
environmental
variable.
D
E
Both
I
I
I
put
it
on
the
command
line,
and
I
ran
the
example,
so
it
still
shows
up
as
unknown
resources.
A
E
C
C
The
from
that
yeah,
with
with
the
fixes
I
I
have
put
that
at
the
last,
with
the
the
fixes
of
the
url
section
instrumentation
that
I
think
are
quite
serious
and
the
support
for
using
away
they
wanted
to
create
a
release
version.
C
So
I
just
waiting
for
I
did
some
tests
with
a
previous
version
this
morning.
So
like
six
hours
ago
for
you
in
my
morning,
and
I
saw-
and
I
found
those
two
issues-
it
was
working
more
or
less.
C
I
suspected
everything,
but
I
found
those
choices
that
they
have
prs
for
them
and
when
they
are
done,
I
would
like
to
release
a
version,
so
everyone
can
have
the
fixes,
mainly
for
your
assessment
and
the
other
topic
is
external
dependencies
problem
and
the
hotel
client-side
telemetry
that
maybe
you
can
go
with
that
one.
First,
oh.
D
Yeah
yeah,
I
just
was
wanted
to
mention
that
some
of
the
other
sdks
for
open
telemetry,
like
you,
know,
browser
like
a
browser.
Sdk,
and
I
don't.
I
don't
know
what
else.
But
they
are
a
rum
sdk,
but
they
wanted
to
to
form
a
client-side
telemetry
sig
and
I
invited
you
to
the
channel
nacho.
I'm
not
sure
if
you
saw
it
or
not,
but
yeah.
D
Hasn't
been
a
lot
of
action,
yet
I
don't
know
if
they've
created
a
cig
or
not
yet,
but
I
plan
on
attending
that
whenever
they,
whenever
they
start
doing
that.
But
I
I
think
that
we'll
have
a
lot
of
crossover
in
terms
of
like
our
data
structures
and
that
sort
of
thing
it'll
be
kind
of
relevant.
So
you
know
like
session
identification
and
that
sort
of
stuff
will
probably
fall
under.
C
Because
I
am
not
directly
in
that
team,
but
yeah
they
are
still
not
using
open
telemetry
for
that,
but
maybe
in
the
the
near
future
they
will
move
to
open
the
language
and
also
having
a
common
language
is
important.
So
I
move
that
information
with
them.
I
don't
know
if
they
I
think
they
have
not
joined
the
channel
or
something
like
that,
but
maybe
they
are
monitoring
it
and
will,
like
probably
would
like
to
be
interested
in
journalists
so
yeah
that
one
okay
and
now
about
the
external
dependencies
problem.
D
C
Without
taking
care
of
which
target,
you
are
really
going
gonna
use.
So
even
if
you
are
not
using,
for
example,
the
url
session
instrumentation,
it
will
try
to
solve
the
the
library
that
it
yes,
that
it
uses.
That
is
the
the
I
mean
the
one
you
used
now
it
was
what
was
the
name
yeah,
the
reachability
library.
So
even
if
you
just
try
to
use
open
telemetry
api,
it
solves
the
dependencies
of
the
of
the.
C
Let's
see
sorry.
It
shows,
for
example,
the
dependencies
of
the
network
status
that
depends
on
reachability.
C
Library,
because
it
didn't
say
that
there
were
two
versions
I
mean
on
linking
it.
It's
shows
a
warning,
but
yeah,
mainly
you.
You
cannot
avoid
that
if
it's
provided
that,
so
that
was
the
main
issue
that
they
wanted
to
to
solve,
and-
and
we
we
had
this
thread
here
mainly
one-
was
that
the
having
two
versions
and
if
it
uses
two
versions
of
reachability
it
crosses
and
the
other.
C
Was
that
using
yeah
using
script
package
manager
with
two
versions
it
will
not
solve,
even
when
not
using
that,
so
it.
A
C
Here
that
one
of
the
lever
is
dependent
on
reachability,
even
not
using
reachability
or
the
url
session
instrumentation,
I
think
they
have
their
own
network
instrumentation.
C
We
could
also
have
different
repositories
each
for
each
instrumentation,
but
I
think
that
could
be
crazy
to
maintain.
D
C
Yeah,
I
think
that
we
could
try
to
use.
I
don't
know
yeah.
C
Yeah
and
I
I
don't
know
what
to
do
with
it,
yeah,
because
if
you
have
yeah,
I
think
them.
D
D
Yeah
yeah,
it's
I
think
it's
detecting
what
the
network
connection
is,
if
it's
using
wi-fi
or
cell
and
then
collecting,
like
the
cell
information.
C
D
C
D
Wonder
is
this
a
problem
with
swift
package
manager,
or
is
this
a
problem
with
using
both
at
the
same
time,
cocoapods
and
swift
package
manager?
It
seems
like
that
seems
like
an
anti-pattern
to
be
using
two
different
package
managers.
C
For
example,
we
are
not
only
providing
package
manager
for
open
tournament
swift
because
it
will
never
build
with
cocoapods,
for
example,
right
yeah,
because
we
are,
depending
on
some
apple
libraries
that
wouldn't
work
with
cocoapods.
So
we
need
to
support.
We
we
need
to
distribute
through
spm,
say
that
yeah.
If
they
are
mixing
cocoapods
versions
with
with
spm,
then
that's
a
problem
that
it's
difficult
to
solve,
except
not
linking
with
that.
D
D
C
A
C
C
Before
and
after
four
I
don't
know
and
also
yeah
and
solving
the
cocoapods
one
is
is
very
difficult.
John
blay
was
saying
about
separating
all
the
instrumentation
to
another
repository,
so
we
have
the
main
library
in
one
repository
and
the
instrumentation.
You
know
that
that
would
only
reduce
the
problem.
I
don't
think
it
will
solve
it.
A
D
C
D
D
C
C
C
Yeah
yeah,
that
was
apple,
asked
people
not
to
use
that,
because
that
means
you
are
switching
on
your
network.
D
C
Down
every
time
you
check
for
it,
so
it
was
quite
bad
for
for
the
batteries
but
anyway
anyway,
everyone
still
use
that.
So
so
it's
under
it's
under
some.
D
A
C
Recently,
another
dependency,
that
is,
data
compression.
That
is
also
a
very
standard
library
for
compressing
with
ship.
C
C
E
C
Continue
like
that
yeah,
but
if
you,
what
if
we
move
to
another
repository,
if
just
if
you
want
to
use
one
of
the
instrumentations,
you
would
have
the
same
problem.
C
I
don't,
I
don't
see
that
that
to
be
a
solution.
D
Is
this
like
a
flaw
in
the
swift
package
manager?
How
does
like?
How
does
gradle
handle
this?
How
does
like
you
know
like
another
package
managing
tool
like
for
java
work
like
don't
they
don't
they
kind
of
encapsulate
each
sub
module
that
you
add
like
they
don't
try
to
cross
share
those
dependencies?
Do
they?
I
don't
know.
D
D
Bump
it
yeah,
because
I
mean
this
seems
like
this
seems
like
a
problem
with
the
entire.
You
know.
Kind
of
swift
ecosystem
right,
like
people
are
gonna
wanna
use
dependencies.
Like
that's
the
whole
point
of
a
package
manager.
C
Everyone
has
the
same
problem
with
it.
E
D
C
C
C
D
That's
the
problem
is,
if
you
would
that,
would
that
solve
the
problem,
because
the
we
might
need
to
rename
it
too,
like
it
can't
be
reachability
anymore,
because
if
somebody
imports
it,
wouldn't
it
make
that
like
there's,
no
name
spacing
right
in
swift,
so.
B
C
If
you
have
this
file
inside
your
library,
yeah.
C
C
D
C
If
you
move
it
there,
it
should,
it
should
work
and
it
will
be
covered
by
network
status,
module
name,
so
it
shouldn't
leak
outside
so
yeah.
C
C
Okay,
so
yeah,
maybe
I
I
can
also
create
a
pr
with
that
with
that
thing,
so
we
avoid
the
dependency
and
create
appear
and
we'll
release
the
version
with
that
change.
C
Okay,
so
not
dependencies
program.
C
C
C
D
C
C
In
fact,
it's
the
path,
I
can
say
you
share
undefined
labs
yeah.
Let
me
let
me
show
that
one.
C
C
They
in
fact
two
of
them-
and
I
created
a
pr
mentioned
in
this
project,
trying
to
use
it
and
they
never
made
the
request.
So
what
I
did
was
I
created
another
fork
of
the
repository
with
support
for
it
in
my
previous
company
that
was
acquired
by
datadogs,
but
the
repository
is
still
alive
and
this
code
is
just
good
enough
and
it's
not
going
to
change
so
I
I
have
also
control
of
that
and
and
it's
a
dead
project
now,
it's
just
for
compatibility
with
with
the
penetration
teams.
C
The
other
thrift
is
also
another
library
used
for
jagger
jagger
exporting
tracers
for
the
jager
exporter
only
and
also
had
to
add
some
some
changes
to
the
original
project
to
be
compatible
with
new
versions
of
sweep
and
also
it's
an
in
a
fork.
I
did
and
it's
not
gonna
change,
at
least
while
it
compiles,
but
I
I
and
I
maintain
this
these
other
four
libraries
are
all.
C
Apple
libraries,
apple
control,
libraries,
so
apple,
maintains
them.
C
This
reachability
is
the
one
that
price
added
to
network
status
that
we
are
going
to
copy
inside,
and
this
data
compression
is
the
one
I
added
recently
to
the
data.exporter
but
could
be
used
by
other
projects.
But
if
we're
using
another
library
we
will
have
to
duplicate
now,
instead
of
having
one
common,
just
duplicate
that
and
and
continue
from
there.
So
I
will
really.
I
will
my
idea
is
removing
it
these
two
projects
here
and
adding
those
files
into
the
other,
and
I
will
do
that
just
after
this
meeting.
C
So
so
I
can
create
up
here
and
so.
E
C
Then
then,
these
users
will
need
to
use
spm,
they
won't
use
cocoapods
or
cartridge
or
other
package
manager
that
are
not
compatible
and
we
are
supporting
from
a
very
early
version,
so
they
they.
We
could
work
with
any
newer
version.
Also,
I
think,
but
yeah
it
should
serve
to
the
version
they
use
if
they
need
it.
C
Okay-
and
there
are
no
more
topics,
at
least
here-
do
you
have
any
other.