►
From YouTube: IETF-JSONPATH-20220927-0800
Description
JSONPATH meeting session at IETF
2022/09/27 0800
https://datatracker.ietf.org/meeting//proceedings/
C
F
D
Cool
sorry,
good
morning,
good
evening,
good
afternoon,
this
is
the
Jason
path,
September
2022
interim,
because
this
is
an
official
iatf
meeting.
The
usual
disclaimer
of
sorry
I'm
having
a
room
but
I
have
not
had
my
second
coffee
yet
no
well.
So
all
the
all
the
usual
rules
and
stipulations
of
the
ietf
apply
to
this
meeting,
as
they
do
sort
of
a
lot
of
variety
of
meetings.
Please
vote
well
note.
Takers,
I'm,
terribly
sorry
that
I
haven't
got
around
to
note
taking
the
last
meeting
Glenn.
D
Thank
you
for
sending
the
action
points
over
immediately
after
I
was
on
Wells
and
find
a
big
chunk.
Blue
sheets
are
are
automatic.
We
don't
have
to
worry
about
that.
I'll
keep
an
eye
out
on
the
chat
in
case.
Anybody
else
shows
up
or
willing
to
do
it
through
that
in
terms
of
the
agenda,
I
see,
we've
put
a.
C
F
E
C
D
So,
do
you
want
to
start
with
I
rejects
too.
F
Sure
so
that
one's
I
think
been
ready
for
our
working
group
last
call
for
a
while.
You
haven't
had
the
Cycles
to
to
document
Shepard
it
yet
sorry.
F
D
Yep
in
terms
in
terms
of
my
availability,
it
is
something
that
I
should
be
able
to
do
in
the
next
couple
of
weeks.
D
Okay,
now
that
I've
I've,
most
I'm,
mostly
back
to
normal,
so
I
just
need
to
put
the
time
in
my
calendar
and
and
sit
down,
and
do
it
so
I
think
that
if
you
haven't,
if,
if
I
haven't
done
it
in
the
next
couple
of
weeks,
then
that
might
be
the
best
course
of
action
to
get.
Things
are
involved.
A
G
Ahead,
sorry
go
ahead.
I
just
wanted
to
I
just
want
to
add
in
that
Jason
schema
would
possibly
have
a
a
passing
interest
in
in
this
records
standard
as
well.
Something
to
some
something
normative
to
reference
would
be
fantastic.
F
G
D
F
E
E
The
the
email
from
me
down
towards
the
bottom
of
that
screen,
here's
the
here's,
my
review,
okay,
yeah
so
yeah
I
mean
that's,
that's
kind
of
summarizes
it,
but
just
to
talk
through
it
I
mean
the
the
major
Improvement
that
issue
201
and
pr258
bring
to
the
table
from
her
perspective,
is
it
straightens
out?
E
The
syntax
makes
it
much
easier
to
do
an
implementation
with
a
similar
syntax
input
to
a
parser
generator,
for
example,
and
that's
that's
all
in
place
in
the
pr
and
really
the
rest
is
about
the
way
the
semantics
are
described.
Clearly,
we
want
to
keep
the
semantics
the
same
overall
but
they're
a
different
options
for
describing
the
semantics
and
we've
looked
at
some
of
those
and
I've
got
something
on
the
agenda
to
discuss
that
again
and
also
the
the
naming
and
the
document
structure.
E
E
I,
don't
find
it
quite
as
readable
as
the
current
back
currently,
but
that's
my
personal
position
so
I,
you
know
Greg
and
I
are
both
looking
to
other
people
to
review
it
to
see
what
you
know,
what
other
perspectives
are
and
I've
requested
some
changes
to
actually
finalize
the
terms,
because
I
think
you
know
the
wheels
might
come
off
when
we
finalize
the
terms.
You
know
we'll
see.
F
Okay,
I
put
my
hand
up
and
let
me
say
a
couple
of
things
speaking
I
I
buy
and
so
early
on.
It
seemed
clear
to
me
that
nobody
wanted
to
change
the
semantics
of
Jason
path,
and
this
was
primarily
an
editorial
discussion,
a
large
one,
a
complicated
one,
but
but
primarily
editorial
and
so
I
I
decided
to
to
stand
back.
So
yesterday,
I
I
read
the
draft.
Thank
you
for
somebody,
whoever
it
was
that
formatted
a
copy,
so
that
I
could
do
that.
F
Thank
you,
and,
and
and
I
hadn't
read
a
draft
for
at
least
two
months,
so
you
know
it's
been
a
while,
so
it
wasn't
fresh
in
my
mind
at
all
and
for
what
it's
worth
I
found
the
current
structure
of
the
draft
acceptably
readable
and
clear
and
comprehensible,
and
it
it
made
sense
to
me
so
so
speaking
with
relatively
fresh
eyes,
I
think
that
what
you
have
there
now
is
is
acceptably
comprehensible,
I,
don't
really
have
an
opinion
as
to
whether
it
was
better
or
worse
than
what
was
before,
because
I
can't
remember
that,
but
it's
essentially
Okay.
F
F
B
Okay,
I
would
like
to
say
thank
you
very
much
for
the
incredible
work.
Greg
and
Glenn
also
I
do
like
the
New
Concept
of
the
two
selectors,
and
it
took
some
time
for
me
to
get
used
to
that
new
structure,
but
Carson.
B
B
Thing,
yes,
I
think
it's
very
important
to
get
the
terms
right.
Maybe
we
discuss
it
now.
In
short,
I
read
the
spec.
G
I'd
just
like
to
say
that
the
the
structure
was
the
important
thing
to
me.
As
far
as
naming
I'm
going
to
leave
that
to
everybody
else,
I
I
really
have
no
opinion
on
what
things
are
named
so
long
as
the
the
the
structure
of
the
things
are
somewhat
sensible
to
me
this.
The
way
this
is
structured
in
the
draft
that
I
put
together
makes
more
sense
than
the
the
current
document,
because
it
is
as
as
Stefan
said,
it's
adamite
it's.
G
It's
really
isolating
the
the
the
the
breakdown
of
what's
actually
operating
and
doing
what,
in
my
mind
at
least
but
as
far
as
what
things
are
named,
I
I,
don't
really
care,
you
could
call
them
the
starters,
appenders
and
Pickers
if
you
want,
which
is
which
is
the
the
placeholder
terms
that
that
I
think
Carson
put
together
so
yeah
I'll
leave
naming
of
stuff
to
other
people.
C
A
Ahead,
Kirsten
yeah,
so
I
sent
this
short
note
this
morning
at
2
am
I.
A
Think
that
there's
one
thing
we
should
do
before
doing
furtheratory
work,
which
is
agree
that
we
allow
the
the
white
card
picker
in
all
places,
because
I
allowed,
which
would
simplify
the
the
structure
somewhat,
because
we
we
don't
need
to
have
special
selectors
for
for
those,
because
I
would
try
to
really
separate
the
concepts
of
starter
and
appender,
and
this
means
that
Aquarius
clearly
defined
as
a
single
starter,
followed
by
a
sequence
of
zero
or
more
appenders.
A
Then
we
should
quickly
deal
with
the
shortcut
appenders,
which
are
just
syntactic
sugar,
on
the
two
bracketed
pandas,
the
child
and
The
Descendant
a
Pender,
and
we
shouldn't
even
try
to
to
use
a
common
name
for
starters
and
appenders,
because
they
are
really
different
things.
They
share
a
right
hand,
side
interface
and
outgoing
interface,
which
is
the
node
list
which
we
can
we
come
to
next.
But
this
frees
up
the
drum
selector
and
we
might
actually
want
to
use
the
term
selector
for
what's
called
picker
in.
In
my
preliminary
terminology,.
A
Of
course,
this
is
a
technical
change
in,
in
contrast
to
the
other
things,
but
I
think
it's
a
technical
change,
the
the
need
for
which
has
become
clear
by
by
the
better
structure
we
now
have,
and
this
technical
change
enables
editorial
fixes,
so
I
think
it's
a
reasonable
thing
to
do.
This
first.
A
Yeah
and
then
I
think
we
need
to
spend
some
time
on
on
the
issue
of
sequencing
in
the
node
list,
which
has
these
weird
determinism
constraints
on
it.
I
think
we
we
have
a
pretty
good
grasp
for
those,
but
we
haven't
really
written
them
up
yet
in
a
way
that
I
would
understand
them
so
yeah
this.
A
This
really
is
about
describing
interfaces
between
the
components,
the
components
that
leave
node
lists
and
the
components
that
pick
up
node
lists
are
a
singular
node
and
and
deliver
node
lists
and
how
these
node
lists
are
integrated
for
components
that
pick
up
a
node
list
and
return
a
node
list.
So
that's
number
three,
and
that
is
entirely
a
technical
issue,
but
I
think
one
where
we
mostly
have
consensus
and
just
have
to
make
sure
we
find
the
right
right
words
to
to
describe
the
consensus.
G
Greg
Kirsten
I
think
all
right,
Carson
I
think
you
I
think
you
answered
this
in
your
in
your
little
speech.
G
But
do
you
have
a
preference
on
the
order
of
PRS
for
the
integrating
the
wild
card
and
then
doing
this
editorial
change?
I'd?
Prefer
them
not
be
part
of
the
same
thing.
They're
part
of
the
same
PR
I
would
I
would
prefer
them
to
be
separate.
G
A
Are
less
important
for
for
our
process?
Then
then,
maybe
in
the
process
that
you
are
used
to
so
I
I
wouldn't
mind
mixing
this
stuff
up.
It's
a
little
bit
of
busy
work
to
to
do
a
technical,
only
PR
that
that
enables
the
the
white
card
picker
in
those
places
that
will
be
enabled
once
once
we
simplify
the
current
draft,
so
my
preference
would
be
to
do
it
in
1pl.
A
G
Yeah,
it's
simple
enough
for
me
to
take
the
take.
What
I
have
currently
and
just
take
the
two
I
guess:
it's
a
descendant
wild
card
and
the
Wild
Card
selector
and
then
move
those
I
guess,
combine
them
into
a
single
picker
and
put
that
up
in
the
bigger
section.
So
it
means
it's
trivial
for
me
to
do
that.
I.
Just
like
the
idea
of
the
the
current
document
and
in
this
PR
just
being
an
editorial
an
editorial
thing,
would
you
consider
this
still
an
editorial,
or
is
this
a
technical
change.
G
Right
so
this
would
enable
having
a
wild
card
and
a
single
index
in
the
same
bracketed-
syntax,
yes,
okay,
so
that
would.
E
Yeah
well:
firstly,
I
support
the
the
Wild
Card
item
number
one
I
think
we
should
should
do
that.
I'd
be
very
happy
to
do
that
after
merging
the
pr,
if,
if
that's
more
convenient,
because
I
think
we
can
all
see
what
the
change
is
and
that
it's
very
straightforward.
E
Secondly,
I
think
the
the
terminology
that's
been
proposed
is
good,
the
starter
and
the
Pender
and
I
think.
Possibly
we
could
go
for
selector
as
the
name
for
start.
Picker
I
think
that
works,
but
it
will
need
a
kind
of
editorial
pass
of
the
whole
document
after
that
to
make
sure
that
we're
not
using
I'm
not
misusing
the
term
selector.
Subsequently.
E
And
I
think
the
sequencing
is
okay,
but
there's
a
an
issue
that
will
come
to
later,
which
throws
the
balls
up
in
the
air
slightly
but
I'm
sure
we
can
start
that
out
and
unfold
it
in
later.
So
you
know
essentially,
once
we've
got
the
terminology
straight
I'd
be
happy
to
merge
the
pr.
F
As
a
recent
reader
of
the
draft
I
found
the
terminology
and
current
use,
nice
and
clean
and
and
I'd
worry
a
little
bit
about
having
selector
I
mean
two
different
things.
So
you
know
my
initial
Instinct
would
be
to
just
stay
with
picker,
unless
somebody
has
a
better
thing.
You
know.
G
The
way
that
I
would
do
this
and
I
would
prefer
to
do
this
in
this
PR
I
would
I,
would
Define
starter
and
then
change
over
everywhere.
Selector
appears
to
appender,
so
that
selector
doesn't
appear
in
the
document
anymore.
I'd
commit
that
and
then
I
would
also
have
a
secondary
commit
of
changing
picker
to
selector.
F
So
it
sounds
like
our
editors
have
consensus
on
so
it
doesn't
sounds
like
nobody
has
any
issues
with
Kirsten's
or
with
the
notion
of
adding
the
wild
card,
and
the
editors
are
roughly
pointing
in
the
same
direction
about
how
to
get
there.
B
No
I
think
it's
a
logical
thing
to
add
the
white
car
to
to
the
to
the
list
of,
because
it's
logical,
it
makes
sense.
It
I,
don't
don't
have
a
problem.
I
applaud
them.
C
F
Then
let's
have
a
go
back
to
Kirsten's
email
here
on
the
screen
James.
Could
you
increase
the
font
size
a
bit.
F
F
E
I
think
we
still
have
actually
three
options.
It
turns
out
for
how
to
describe
the
semantics
of
Pickers
and
I'd
just
like
us
to
spend
five
minutes
on
this
call
looking
at
the
options
so
that
we
we're
sure
we're
going
in
the
right
direction.
You
know
I
think
it's
acceptable.
The
way
it
is
which
is
option
one
here
where
the
Picker
is
essentially
a
selector,
because
it
it's
given
a
value
and
it
selects
zero
or
more
children
of
that
value
and
returns
them
in
a
node
list.
E
E
Another
option
which
Greg
and
I
discussed
early
on
which
Greg
didn't
like
and
I
favored,
was
that
the
Picker
should
be
really
a
predicate
and
that
it
determines
whether
any
given
child
of
the
value
is
to
be
selected.
Then
the
selector
would
say
you
know
all
the
children
selected
by
the
Picker
are.
You
know
for
all
the
children
which
are,
of
course
the
Picker
is
true,
appear
in
the
node
list
in
the
order
blah
blah
blah.
E
You
know,
so
that
would
be
more
of
a
in
my
terms,
a
less
implementation,
bias,
kind
of
approach
and
it's
actually
got
a
narrower
interface,
because
the
pickers
then
don't
get
to
choose
the
order
of
the
node
list
and
they
can't
see
the
parents
of
the
children.
So
it's
much
a
much
tighter
description.
E
The
third
option,
which
I
you
know
we,
which
we
moved
away
from
fairly
early
on,
but
I,
think
we
need
to
just
revisit
so
we've
got
the
the
broad
set
of
options
is
to
not
actually
give
Pickers
semantics
on
their
own,
but
only
to
give
Pickers
semantics
in
the
context
of
a
selector,
so
essentially
to
say.
Each
picker
behaves
like
this
when
it's
in
the
child
selector
and
then
to
infer
the
semantics
of
a
child
selector
from
that.
E
Much
as
we
do
with
today's
list,
selector
in
the
current
draft
and
the
benefit
there
is
that
we
don't
then
get
into
implementation
at
all.
Or
you
know
we
don't.
We
don't
separate
out
picker
as
an
implementation
item.
It's
purely
syntactic.
E
So
you
know,
we've
probably
got
different
opinions
on
this,
but
I
just
wanted
to
float
those
three
options
and
say
that
I
prefer
either
of
the
other
two.
You
know
I,
don't
really
favor
the
current
one,
but
I
could
live
with
it,
but
I
don't
really
like
it.
B
Okay,
if
I
would
like
to
comment
on
on
on
on
Glenn's
options,
I
first.
B
Would
come
to
the
terms
I'm,
not
very
happy
with
the
term
picker
pick
I
in
from
my
point
of
view.
Pica
is
a
misleading
term,
as
picking
is
an
explicit
action,
but
a
number
one
or
spring
full,
merely
passive
labels
of
an
array
element
or
an
an
object.
Member.
B
Those
are
only
indicating
to
the
items
in
question
rather
than
actively
picking
which
is
selecting,
and
this
is
from
my
point
of
view,
the
discrepance
between
those
terms
are
they
can
you
exchange
them,
picking
the
Picker
and
selector,
and
this
is
the
current
discussion.
We
we,
we
change
the
meaning-
and
we
already
have
the
term
index
in
the
spec
index-
is
a
non-negative
integer
that
identifies
a
specific
element
in
an
array.
B
Note
that
the
term
indexing
is
also
used
for
accessing
elements
using
negative
integers
and
for
accessing
member
values
in
an
object
using
their
member
name.
This
is
read
from
our
spec,
so
we
already
relaxed
the
meaning
of
index,
and
so
why
not
think
further
and
relax
this
term
more
because
the
new
term,
a
single
item
within
a
bracketed
child
selector
that
matches
the
values
which
are
to
be
selected
so.
B
The
most
implementers
users,
if
they
meet
the
brackets,
would
think
about
an
index
array
index
or
if
it's
a
string,
it's
a
name
if
they
come
from
a
member
name.
If
they
come
from
JavaScript.
G
B
Changed
ticker
to
selector,
so
I
start
with.
The
second
object,
which
reads
now
and
index
is
a
predicate
which
determines
given
child
of
a
value.
A
child
selector
selects
those
children
of
its
input,
value
for
which
an
index
is
a
match.
I
I
changed
the
wording
somewhat
the
first
option:
a
selector
presents
a
single
value
to
an
index.
The
index
selects
zero
or
more
children
of
a
value,
and
the
resultant
note
list
is
used
in
constructing
the
selectors
note
list.
This
simply
makes
no
sense.
B
B
B
These
are
not
selectors,
they
are
only
indicators,
they
are
indicating
to
members
or
array
elements
and
the
selecting
must
be
done
from
the
outer
from
the
selector
or
maybe.
B
So
that's
my
point
of
view
after
thinking
some
time
about
this
wording.
Okay,
thank
you.
A
Yeah
I
think
what
Stefan
just
said
is
really
a
prime
demonstration.
Why
you
shouldn't
mix
up
discussion
of
terms
with
discussion
of
the
technical
Concepts
behind
that,
because
in
the
end,
we
we
have
to
get
the
technical
Concepts
right
and
we
will
find
terms
for
those.
So,
let's
stick
with
the
working
terms
for
now,
and
so
I
I
talked
about
the
appendix
and
and
Pickers
and
to
Glenn
the
fact
that
a
description
technique
is
also
directly
implementable,
does
not
make
it
less
desirable
as
a
description
technique.
A
E
Yeah
but
I
think
it
and
and
naive
implementation
of
the
spec
would
tend
to
make
a
picker
a
a
sub
component.
With
that
description,.
E
A
We
can
we
can.
We
can
add
some
some
verbiage,
that
that
says
that
this
is
really
all
just
description
technique
and
there
are
better
ways
to
to
implement
this,
but
I
think
that
doesn't
have
to
stop
at
us
using
a
really
useful,
really
highly
understandable
description
technique,
just
because
it
also
could
be
used
by
an
implementation.
E
C
A
Exactly
so,
that's
the
way
it
should
be
and
which
we
should
not
make
the
description
more
complicated
by
by
putting
in
stuff
from
the
append
or
into
the
Picker
again.
These
should
nicely
abstract
away
from
each
other
and.
E
G
You
may
have
mistakenly
I
think
swapped
your
option
numbers
in
this
post.
Oh.
G
I
think
option
two
was
was
the
way
that
I
have
it
currently
and
option.
One
was
your
your
predicate.
A
Forget
option
one
and
two:
the
the
question
is:
is
this
picker
or
the
appendo
doing
the
taking
apart
of
the
node
and
presenting
the
children.
E
A
And
the
the
answer
in
my
option
of
whatever
its
number
is,
is
that
an
appendile
presents
one
node
for
the
child,
a
panda
and
a
tree
of
nodes
for
The
Descendants
appendo,
which
are
not
the
child
and
not
The
Descendants,
which
is
why
the
term
is
a
little
bit
confusing.
But
again
we
are
going
to
get
to
terms
later
and
then
it's
the
job
of
the
Picker.
To
actually
take
that
note
apart-
and
there
are
several
good
reasons
and
the
main
reason
to
do
it.
A
So
the
Picker
has
to
have
a
look
at
the
whole
thing
and
not
just
the
child,
because
the
the
Picker
needs
the
index
or
the
member
name
to
decide
which
of
the
nodes
to
which
of
the
child
nodes
to
include
so.
E
Yeah
can
I
just
interrupt
because
I
possibly
didn't
explain,
option
two
well
enough
so
option
two
the
Picker
would
be
given
the
value
of
the
child
and
its
array
index
or
member
name.
A
This
is
seriously
real
to
me
because
you
have
a
component
that
is
doing
some
some
taking
apart,
that
that
is
really
unneeded.
I
mean
if,
if
I
want
to
get
item
number
five,
there's
really
no
point
looking
at
items,
number
zero
and
four
and
then
six
tools.
E
Not
in
not
in
implementation
terms,
I
agree,
but
in
Spec
terms
it's
just
a
for
all.
C
G
So
I'd
like
to
address
something
that
Carson
touched
on
a
little
bit,
Glenn
your
your
argument
that
the
way
that
I'm,
explaining
right
your
option,
one
here-
is
that
it
shows
in
this
issue,
which
is
the
current
explanation.
The
current
descriptive
language
in
the
spec
is
option
one
here.
G
G
A
naive
I
could
also
argue
that
that
a
naive
implementation
could
have
an
offender
take
apart,
all
of
its
take
apart
its
children
and
give
each
child
to
the
the
the
Picker
and
the
Picker
says
yes
or
no,
and
that
would
be
an
implementation.
So
I
don't
know
that.
That's
a
valid
argument.
G
Secondly,
Carson
kind
of
touched
on
this
as
well.
If
I
have
an
array
of
a
thousand
items
with
just
an
index,
picker
of
three,
then
I
want
the
fourth
item.
I,
don't
need
to
I,
don't
need
the
appender
to
present
all
1000
items
to
the
the
index
picker
of
three
in
order
to
just
get
the
fourth
item.
If
you
pre,
if
the
appender
instead
presents
the
parent
to
the
index
picker
the
index
picker
says:
oh
there's.
The
fourth
item.
I'll
give
give
that
back.
E
Yeah,
but
you
see
you
you're
kind
of
supporting
my
view
that
whichever
option
we
go
with,
readers
of
this
or
implementers
are
going
to
be
thinking
of
it
in
terms
of
implementation
that
we
can't.
C
E
G
Whichever
way
we
go,
they
are
going
to
be
thinking
of
implementation,
so
my
bet
would
be
to
have
the
have
the
descriptive
language
be
the
more
efficient
of
the
two.
If
a
naive
implementation
wants
to
follow
the
spec
and
do
that,
then
at
least
they
would
get
something
that
performs
okay-ish
so.
F
So
since,
since
this
is
the
point,
I
wanted
to
address,
I'll
butt
in
so
I
disagree
strongly.
I
think
that
the
most
single
most
important
thing
about
the
language
in
the
spec
is
that
it'd
be
clear
about
the
semantics.
The
result
to
be
achieved
if,
if
the
best
way
to
achieve
that,
Clarity
is
to
describe
an
inefficient
implementation,
I'm
totally.
Okay,
with
that,
what
my
original
reason
for
putting
up
my
hand,
was
to
to
Echo
what
Carson
said
a
couple
of
seconds
ago,
which
is
it's
perfectly
okay
to
use?
F
You
know
an
iterative
pseudocodish
description
of
how
it's
done
to
establish
semantics.
That's
not
that's!
Not
radical
or
new,
however,
I
think
if
you
do
that,
it's
absolutely
necessary
to
have
language
saying
that
you
know
this.
This
algorithm
is
provided
only
for
explanatory
Clarity
and
there
is
no
requirement
whatsoever
that
it
actually
be
used,
which,
which
means
it's
perfectly
okay,
to
do
that.
As
long
as
you
give
that
language,
sorry,
that's
all
Kirsten.
G
A
Difference
between
the
two
approaches
is
essentially
where,
in
in
the
abstract
definition,
We,
Run,
The,
enumerator
and
running
the
enumerator
in
the
appenda
means
that
we
have
to
to
supply
everything
the
the
enumerator
find
to
the
the
pickup.
So
the
the
Picker
interface
is
not
a
nodeless
based
interface,
but
it's
an
interface
that
Associates
the
nodes
with
attributes
such
as
array
index
and
and
member
name,
so
that
that
makes
things
slightly
more
complicated
for
me
because
and
I
no
longer
can
think
in
terms
of
notice.
E
A
E
E
You
know
children
with
their
index
or
or
object
name,
translation.
G
Or
you
can
just
give
the
the
Picker,
the
parent,
node
and
or
I.
C
G
That's
what
and
it
selects
which
ones
that
you
want.
G
Are
we
going
to
end
up
with
possibly
extensibility
issues
if
we
go
with
your
option,
two
here,
for
instance,
Carson
just
mentioned
the
the
the
negative
index
issue,
which
added
another
parameter.
G
Are
we
gonna
later
find?
Oh
there's,
another
thing
that
requires
another
parameter.
We
have
to
add
that
now.
E
I
do
think
that's
a
an
issue,
because
these
are
just
spec
words:
they
don't
actually
require
an
implementation
or
suggested
implementation.
Don't
think
in
particular.
G
A
E
Okay,
I'm
coming
around
to
option
one
on
this
basis,
so
yeah
any
any
other
views.
E
Okay,
then
I'll
remove
this
kind
of
concern.
Then
we'll
go
with
the
current
approach.
C
G
So
thanks
for
guiding
us
around
to
that
that
discussion,
Glenn.
A
That's
fair
enough,
thank
you
so
my
view,
but
my
current
mental
image
of
option.
One
means
that
descendants
append
a
presents
the
current
node,
as
well
as
all
structured
descendant
nodes
to
the
Picker
individually,
so
the
Picker
doesn't
get
to
to
operate
on
the
relationship
between
the
descendant
notes,
so
that
that
would
essentially
be
very,
very
simple.
A
By
saying
the
child's
director
just
presents
the
parent
and
The
Descendant
selector
just
presents
the
parent
and
each
descendant
each
structured
descendant
node
to
the
pickup,
but
the
Picker
does
not
know
whether
it's
being
used
in
a
child
selector
or
descendant
selector.
E
That's
that's
true,
but
I
think
the
the
way
the
spec
is
currently
structured
and
the
way
the
pr
is
structured
means
that
the
descendant
selectors
are
actually
defined
in
terms
of
the
child
selector.
So
essentially
we
we
enumerate
a
list
of
descendants
and
then
we
run
the
child
selector
against
each
one.
So
it's
a
kind
of
it
doesn't
really
matter.
You
know,
because
we
end
up
with
the
semantics
of
the
child,
selector
really
driving
things.
A
F
Okay,
so
let's
move
along
to
with
Glenn's
list
sorry.
E
Just
before
we
move
on,
Stefan
is
frowning
and
he
had
points
about
this
index
notion.
So
I
just
want
to
check
he's
comfortable
with
the
option.
One.
B
B
That
you
exchange
the
numbers
in
your
email
and
in
the
so
what
I
preferred
when
reading
your
email,
Wars
number
two
is
a
now.
B
B
Yes,
I
prefer
number
two
I
prefer
number
two
in
here
in
your
comment,
which
is
shown
here.
Yes,
definitely
definitely
number
two
I
I'm
I.
It's
very
unusual
for
me
to
to
see
a
picker
as
an
actively
working
acting
element
or
a
function
or
an
algorithm
which
selects
parts
of
a
node
list
or
a
note
list
except
the
filter
picker.
The
filter
picker
in
fact,
has.
B
B
These
elements
of
of
the
former
selector,
but
by
that
discussion
of
how
do
the
Picker
and
the
selector
interact,
and
it's
it's
hot
for
me
and
I-
think
also
for
potential
implementers
in
future
to
to
get
that
right.
G
I
would
I
would
agree
with
the
the
idea
that
it's
a
a
passive
like
the
a
member
name
on
its
own
is
passive,
but
I
wouldn't
necessarily
apply
that
passivity
to
all
of
the
pickers.
For
instance,
the
the
filter
picture
is
actually
a
delegate.
It
is
actually
iterating
through
all
the
children
and
and
identifying
which
ones
are
selected.
E
Okay,
may
I
make
a
suggestion
that
we
we
go
with
option
one
as
described
here
and
see
how
it
comes
out,
because
I
I
think
based
on
the
discussion,
it
will
actually
come
out.
Okay
and
I
think
it
might
actually
satisfy
Stefan
when
we
get
the
final
text.
F
Yeah,
this
is
at
the
end
of
the
day
in
a
tutorial
discussion
and
the
best
way
to
move
an
editorial
discussion
along
is
to
edit
to
actually,
you
know,
produce
a
subsequent
draft
and
and
see
how
concretely
we
feel
about
it.
C
E
I
mean
if
there's
any
comfort,
Stefan
I,
share
those
concerns
and,
as
I'm
doing,
the
editing
I
will
be
sensitive
to
those
concerns.
So
I'm
trying
to
you
know,
make
the
description
as
clear
as
possible.
A
Yeah
at
some
point
we
will
need
to
have
this
Stone
energy
discussion
and
maybe
picker
really
was
a
pretty
bad
term
to
use
here,
but
yeah.
That's
why
I
chose
it,
so
we
we
can
make
sure
that
we
change
it
before
casting
it.
In
concrete.
G
Definitely
having
a
bad
having
a
bad
option
in
order
to
force
change
is,
is
a
good,
a
good
strategy.
G
E
Yeah
so
I
raised
this
issue
kind
of
separately
from
the
current
descript
discussion.
You
know
the
pr
discussion,
because
I
noticed
that
our
current
spec
current
rap
was
overly
deterministic
such
that
it
would
even
exclude
some
existing
implementations,
so
I
thought
we
needed
to
loosen
it
up
a
bit.
Add
more
non-determinism.
A
So
you
say
that
at
the
start
of
the
ticket
you
say
that
it
cannot
produce
a
specific
result.
I'm
not
sure
I,
know
why
you
think
we
should
be
able
to
reuse.
This
specific
result
he's
going
way
to
the
top.
C
E
Yeah
well
the
that
we've
actually
got
that
example
there
in
the
spec
and
we
can't
produce
that
example,
result.
E
But
that
example
result
is
precisely
what
is
produced,
or
rather
it's
along
the
lines
of
the
consensus
of
the
implementations,
and
we
seem
to
be
removing
that
consensus
for
no
particularly
good
reason.
G
A
No,
the
the
there
is
no
problem
with
the
object
members
here.
It's
just
that
the
the
the
yeah
that
it's
just
that
this
is
a
breadth
first
result
and
I
just
wanted
to
make
sure
that
this
is
what
we
want.
I,
don't
have
a
specific
opinion
here,
but
allowing
breadth
first.
So
it's
a
little
bit
unintuitive
to
me,
but
we
can
certainly
opt
that
we
want
to
do
that.
F
E
G
Would
like
I
would
like
to
say
that
my
my
implementation
is
not
among
the
consensus.
E
Okay,
that
was
that
was
one
item
The
Descendant
ordering
just
to
loosen
it
up
a
bit.
Then
the
the
next
one
was
the
list.
Selector
ordering.
E
Right
so
yeah,
and
this
does
go
along
the
lines
of
the
iteration
order
for
objects.
E
E
But
that's
not
the
case
because
for
for
example,
in
in
golang,
the
iteration
order
can
change
from
one
iteration
to
the
next,
so
I
wanted
to
allow
for
that,
and
that
example
shows
a
possibility.
You
know
that
the
iteration
order
could
could
be
for
that
list.
Selector.
In
current
terms,
two
Pickers,
if
you
like,
could
produce
different
orderings
could
iterate
the
object
in
different
different
orders.
F
You
know
the
one
thing
I
I
think
I
toss
to
comment
on
here
is
that
the
reason
why
this
is
the
case
is
is
not
mysterious
at
all
and
it
falls
directly
out
of
the
definition
of
Json,
which,
which
is
clear
that
you
know
order
is,
cannot
be
expected
to
be
maintained
when
accessing
objects,
members
of
an
object,
so
so
that
falls
out
straightforwardly
but
I
think
it's
important
also
to
emphasize
that
applying
a
Json
path
does
not
reorder
array
elements
so
yeah
whenever
you
say
you
know,
do
not
count
on
ordering
and
objects.
C
E
Yeah
and
so
let's
look
at
descendant
selector
ordering
now,
if
you
go
down
change,
please
yeah!
So
to
reread
this
for
a
moment,
myself.
E
Yeah,
it's
essentially
the
same
issue
that,
as
you
Traverse
a
a
node,
a
value
than
the
order
in
which
you
encounter
the
children
can
vary
from
from
one
one
picker
to
the
next.
A
E
And
then,
lastly,
repeated
runs,
so
if
you
have
quite
straightforward
Jason
path
and
value
which
can
produce
either
order,
we
maybe
should
even
say
that,
and
a
given
implementation
could
produce
different
orders
on
repeated
runs
just
to
make
it
absolutely
clear
that
we're
not
assuming
that
implementations
are
deterministic.
E
Yeah
yeah,
so
some
array
related
ones
can
can
deliver
that.
A
Yeah
and
the
combinations
of
a
raised
descendants
and
and
maps
I
think
it's
getting
pretty
interesting
when
you
actually
run
more
complicated
examples.
E
Yeah
yeah
and
with
multiple
Pickers
in
a
descendant,
node
list
and
yeah,
it
gets
quite
juicy.
A
Well,
I
think
in
in
the
end,
this
is,
to
a
large
extent
about
what
the
descendants,
the
panda
does
to
the
various
note
lists.
It
gets
from
the
various
speakers
at
talks
to
so
for
the
the
child.
A
panda,
the
the
comma
into
the
Pender,
gives
an
ordering
gives
a
partial
ordering
for
the
The
Descendants
selector.
A
We
just
said
we
we
are
allowing
bread
first,
so
that
is
going
to
be
interesting
and
when
the
the
results
of
the
various
Pickers
are
being
put
together,
which
part
of
the
ordering
is
maintained
and
and
which
part
is
is
not
so,
can
you
given
that
that?
Well,
if
you
have
a
a
contains
one,
two
comma
B
contains
three
four
situation:
can
you
have
something
like
A1,
B,
3,
2,
4
and
so
on?
A
I
think
we
need
to
create
some
examples
for
that
and
and
make
sure
that
the
the
merging
step
we
we
have
in
mind
for
The
Descendant
selector
does
the
right
thing.
E
A
E
G
C
A
G
Of
this
is
going
to
be
the
test
Suite
or,
however,
we
manage
testing
of
this
yeah.
Having
listing
out
every
possible
permutation
of
acceptable.
Orderings
is
probably
not
ideal,
so
we
would
want
a
better
way
to
do
that,
but
I
don't
know
what
I'm
not
sure
what
that
is
in
particular,
but
we
we
definitely
need
to
nail
this
down
in
order
to
implement
some
sort
of
test.
Suite.
C
B
I
must
confess
that
I
need
to
read
that
issue
again
much
more
carefully.
I
I
cannot
imagine
situation
where
a
single
application
will.
B
Will
present
with
repeated
runs
different
orders
of
noteless,
but
maybe
I
I
I
have
to
read
it
again
more
carefully.
G
It's
definitely
weird,
but
it
does
happen,
especially
as
as
Glenn
posted
in
here
in
golang,
the
the
the
present
in
Json
objects.
The
presentation
of
the
various
members
can
be
is
non-deterministic
in
in
their
their
sequencing.
F
Many
many
programming
languages
have
a
you
know,
a
Content
addressable
array,
you
know
mechanism,
we
do
call
it
a
map
or
a
hash
map
or
whatever
and
I
I
I
believe
it
is.
It
is
occasionally
the
case
that
there
is
some
Randomness
and
and
many
many
languages
which
which
provide
this
facility
are
clear,
that
they
do
not
guarantee
the
ordering
and,
and
there
may
be
actual
Randomness
in
the
generation
of
hash
sizes,
hash
table
sizes-
and
you
know,
test
chaining
into.
B
Okay,
this
surprises
me
that,
in
in
in
practice
with
repeated
runs,
you
you'll
have
that
non-determinism,
but
I'm
learning,
just
learning.
A
Well,
the
main
reason
why
you
want
this
is
that
hash
tables
have
anomalies
when
you
sort
all
items
into
one
bin
or
the
majority
of
the
items
into
one
bin,
and
you
don't
want
to
and
to
put
an
attacker
in
a
situation
that
it
can
predict
which
bins
are
things
put
in.
So
you
add
some
Randomness
to
to
your
your
the
hash
function
and
some
implementations
add
that
Randomness
per
run
of
of
the
the
language
and
one
added
per
iterator,
and
if
you
do
it
per
per
iterator,
then
it
becomes
interesting.
G
F
Yeah,
okay,
so
I
think.
The
conclusion
of
this
is
that
we
need
to
make
sure
we
have
a
rich
set
of
examples
that
shows
the
range
of
things
that
can
happen
when
you
combine
multi-level
multi-level
expressions.
G
A
D
Do
we
want
to
cover
any
of
the
other
issues
that
that
are
notable
and
outstanding
or.
F
Well,
I
think
you
know:
I
went
through
the
issues
yesterday
and
it's
not
obvious
that
any
of
them
are
still
really,
as
you
said,
notable
and
outstanding.
F
A
A
Was
that
we
agreed
that
The
Descendant
selector
actually
include
self,
like
just
the
side,
charity
selector
includes
only
itself
so
from
an
expats
point
of
view.
It
should
be
called
safe,
selector
and
self
or
descendants
selector.
A
E
C
E
A
Yeah
I
seem
to
be
in
the
rough
year,
so
most
people
seem
to
think
that
we
should
stick
with
singular
paths
for
all
of
filter
expressions.
A
E
Yeah
I
I
didn't
detect
that
that
consensus
personally
and
I
was
kind
of
going
along
with
the
drift
of
the
issue
of
the
initial
proposal
to
include
the
you
know:
food
non-singular
paths,
but
we
can
always
add
it
later
in
the
spec
process.
A
Okay,
yeah
I
mean
it's
a
little
bit
of
a
black
thing,
the
the
of
course
it's
better
to
have
a
simpler
spec
and
and
only
allow
singular
paths
at
all.
But
it's
also
weird
to
have
an
existence
expression
that
cannot
be
applied
in
places
where
you
might
have
multiple
images
so
that
that's
just
my
general
consideration
here.
Maybe
we
should
have
another
look
at
the
the
implementation
consensus
to
to
find
out
whether
we
even.
G
G
And
my
personal
thoughts
on
this
is
that
I
think
it's
fine
for
the
exists
expression,
but
not
within
filter
expressions,
for
example,
I
think
I
think
having,
but
then
at
the
same
time
having
it
having
like
a
a
wild
card
in
an
exist
expression,
but
not
being
able
to
support
that
in
a
filter.
Expression
feels
weird
well.
A
Don't
know,
but
for
a
comparison
you
want
to
have
a
single
video
so
so
having
a
singular
hypothesis
kind
of
logical
for
an
existence,
expression.
B
If
we
allow
multiple
results
within
a
comparison,
we
would
introduce
the
in
operator
through
the
back
door.
I
think
yeah.
C
E
E
E
C
C
D
We
don't
have
any
more
issues
that
we
want
to
cut
through.
Shall
we
agree
on
roughly
when
we
should
have
follow-up
interim
meaning.
F
Yes,
but
but
I'd
like
to
remark
that
we're
a
little
further
out
from
being
finished
than
I
thought,
we
were
well
I'm,
a
little
disappointed,
I
I
thought
we
were
closer
than
we
are
so
ietf
115
is
in
November
and
at
the
moment
it
seems
to
me
unlikely
we're
going
to
be
able
to
do
a.
F
A
Well,
there
isn't
a
proposal
right
now,
but
the
internet
draft
deadline
is
on
October
24th,
so
we
we
have
about
four
weeks
to
get
our
act
together.
If
we
want
to
to
make
that
deadline,
then
we
would
still
have
two
weeks
for
working
with
last
call
before
the
igf,
so
this
is
kind
of
the
most
optimistic
thing.
Can
we
fix
the
remaining
issues
in
four
weeks.
D
Remember
costs
and
we
we
need
to
get
I
regex
through
first,
because
Jason
files
depends
on
them.
E
Okay,
we're
kind
of
serialized
on
the
Greg
finishing
the
pr,
so
it
somewhat
depends
on
his
Cycles
too.
C
D
A
So
would
we
want
to
have
an
interim
in
week
42.
A
But
I
at
the
time
it
starts.
We
already
done.