►
From YouTube: IETF101-JMAP-20180322-1550
Description
JMAP meeting session at IETF101
2018/03/22 1550
https://datatracker.ietf.org/meeting/101/proceedings/
C
C
Okay,
hi
everyone:
let's
get
started
so
Ron
Jory.
You
should
be
noting
the
no
yeah
yeah
note
that
no
well,
it
has
been
noted,
good,
okay,
quick
overview
of
the
agenda
I'm
just
going
to
go
briefly
over
the
major
changes.
Since
idea,
100
in
Singapore
been
a
lot
of
smaller
editorial
fixes
and
stuff.
Obviously
you
can
go
see
for
get
repository
and
log
of
changes.
If
you
would
like
or
compare
the
different
versions,
then
we'll
have
a
quick
look
at
maining
issues.
C
There
will
be
some
discussion
of
a
few
extensions.
New
drafts
that
been
proposed
drawn
might
talk
a
bit
about
test
suite
I'm,
not
sure
how
much
that's
gonna
happen.
So
if
there's
any
other
business,
probably
actually
not
gonna
need
as
much
time
as
is
on
here.
I
think
this
is
gonna
be
shorter
than
we
originally
thought.
So
big
changes
since
before
one
was
just
the
big
rename.
This
was
where
we
renamed
all
the
methods,
names
and
simplify
things.
C
So
the
response
names
now
match
the
request
names
and
you
don't
need
to
pluralize
anything
everything
singular
and
you
don't
need
to
conjugate
anything.
So
it
makes
it
much
easier
if
it's
special,
if
you're
not
a
native
English
speaker
to
to
add
new
methods
and
not
get
confused
with
what's
going
on,
and
we
also
renamed
message
object,
which
represents
the
RFC
5v2
to
message
in
JSON
to
email,
because
each
object
in
je
map
has
its
own
ID
and
it
was
confusing
a
lot
of
people.
C
So
quick
example:
you
can
see
yeah
so
request
and
response
request.
Name
is
now
the
same
as
the
response
name:
email
get,
we're
getting
email
objects
now
then
verb
getting
some
IDs
and
we're
only
fetching
the
third
ID
property
and
then
we're
getting
back
the
data
that
we
requested.
So
that's
that
sort
collation
algorithms.
This
is
simply
the
sorting
on
queries
is
you
can
now
specify
which
collation
algorithm
to
use
in
the
server
can
advertise
which
ones
its
ports
and
we're
using
the
identifiers
from
the
I
on
a
registry
established
in
4790?
C
And
if
you
don't
specify
one,
you
get
a
default
and
the
default
is
reck.
We
don't
mandate
a
default,
but
we
recommend
it
the
case
insensitive
and
Unicode
aware
and
there's
two
options
that
we
suggest
that
both
fulfill
those
requirements,
one
of
which
is
the
Unicode
regulation,
algorithm,
come
from
which
one
anyway
see
the
spec.
C
So
again,
an
example.
This
the
sort
property
which
is
an
array
of
properties
to
sort
by.
If
two
things
have
the
same.
You
move
on
to
the
next
property.
We're
gonna
got
one
here.
Instead
of
having
we
have
string,
interpolation
is
now
an
array
of
objects
which
is
easier
to
process.
So
we're
talking
about
the
subject:
property
we're
sorting
in
a
descending
order,
because
it's
a
something
is
false,
and
here
we're
using
the
Unicode
case
map
insulation,
which,
assuming
the
service
ports,
that
that
should
all
be
good.
C
And
then,
yes,
by
far
the
biggest
change
is
the
new
email
object.
As
I
said,
the
represents
the
message
in
JSON.
This
is
way
more
powerful
and
it
was
before
it
can
give
you
the
full
my
instruction
now,
whereas
before
you
had
a
really
simplified
version,
but
we've
and
you
can
access
arbitrary
headers
in
much
more
powerful
ways
than
before.
But
we've
also
kept
a
lot
of
the
simpler
ways
of
accessing
the
same
data
so
that
clients
can
kind
of
scale
their
needs.
C
Simple
clients
are
not
going
to
be
overwhelmed
by
huge
amounts
of
complexity,
but,
as
you
get
more
advanced,
the
power
is
there
without
the
clients
having
to
resort
to
passing
the
full.
My
instruction,
the
cell
from
the
raw
message,
which
also
means
they
can
be
much
more
efficient
in
terms
of
the
data
transmitted
over
the
wire,
because
you
can
request
just
what
you
want,
including
now
you
can
request
a
maximum
size
for
a
body
value.
So
this
is
this.
C
Was
the
I
think
the
last
place
where
you
could
get
flooded
with
data
when
like
you're
quiet,
but
your
client
might
request
ten
messages,
but
it
turns
out
one
of
those
messages
has
a
50
megabyte
text,
part
you'd,
get
that
full
text
by
Pat.
Well
now
you
can
say
I
want
to
limit
that
to
a
maximum
of
100
kilobytes,
for
example,
and
the
server
will
truncate
that
value
when
when
it
returns
it
lets,
you
know,
there's
been
truncated
so
that
again
you
can't
get
flooded
with
data.
C
C
D
Should
send
you
more
detailed
comments,
one
thing
which
is
a
nit
really
but
I
think
you
took
out
utf-8
reference
as
normative
and
I.
D
C
D
C
I
think
I've
got
it.
Oh
no
I
don't
actually
have
an
example
for
that,
so
that's
yeah,
that's
mine.
The
only
other
thing
I
think
really
to
mention
about
it
is
the
so
you
can
request
that
the
server
give
you
a
list
of
parts
that
you
would
display
as
an
HTML
body
or
as
text
body,
and
that
is
now
a
list
instead
of
a
single
value
which
lets
it
more
easily
handle.
When
you
have
something
like
mailing
list
that
adds
its
own
signature
as
a
separate
mine
part
afterwards,.
D
D
C
D
C
Not
it's
it's
not
that
you
generator,
but
yes,
I
often
Apple
Mac
male
in
particular,
likes
to
send
messages
like
that.
So
how
are
you
gonna
represent?
That,
basically,
is
the
question
you
could
just
treat
all
of
those
images
as
attachments
and
drop
the
fact
that
they're
meant
to
be
in
line,
but
that
loses
some
of
the
the
information.
That's
there.
Obviously
you
could
try
and
convert
the
whole
thing
to
HTML,
but
that's
way
more
complicated.
D
E
C
C
B
D
C
Okay,
so
there
aren't
too
many
remaining
issues.
Let's
start
with
core
and
I
think
this
one
actually
may
have
been
resolved.
At
least,
we've
got
a
good
idea
capability
identifier,
so
each
each
spec,
including
core
and
male
and
any
future
ones,
and
including
any
vendor
extensions
which
are
nice
and
Di's
need
an
identifier.
This
is
used
in
the
capabilities
object,
so
the
service
and
advertise
I
support
this
capability.
The
client
uses
this
identifier
when
it
sends
an
API
request
to
say:
I
want
to
use
this
capability.
C
You
have
to
opt
in
to
use
any
of
the
capabilities,
so
it
needs
to
be
unique,
perfectly
short
and
I
think
you'll
be
quite
nice
if
it
maybe
even
this
is
just
a
recommendation
that
if
it's
something
that
can
help
you
find
documentation
about
what
this
capability
is
and
I'm
thinking
here,
just
if
you're
a
developer
and
you
talk
to
new
server-
and
you
see
it-
has
some
capability
don't
recognize
if
that
identify
in
some
way
helps
you
find
the
information
about
that
that
we
good.
So
here's
a
few
examples
now.
C
This
is
what
we
actually
now
currently
have
in
the
spiritual
Chris
humans
gestured.
So
this
is
for
cause
an
example.
J
map
call
another
option
is
we
use
like
whatever
RC
is
issued,
RFC
and
number
as
the
identifier,
because,
again
that
it's
easy
to
Google
and
you'll
find
a
spec.
Although
J
map
call
you'd
probably
find
the
spec
as
well
as
various
options
here,
I
think
we
definitely
want
to
recommend
for
vendor
extensions.
It's
a
uriah
to
domain
that
you
are
in
control
of,
and
we
would
recommend
that
you
helped
make
that
URI
point
documentation.
C
D
Mean
urine
is
compelling
because
it's
you
know
it's
still
your
ISO,
but
I
understand
it
can
get
their
bows.
I
mean
yeah.
C
A
C
A
B
B
F
C
E
C
A
Good
point:
yeah,
yeah
I
think
you
should
either
go
with
the
first
option
or
the
last
okay
and
oh
yeah.
The
only
thing
you
have
in
those
dot
dots
there
is
params,
it's
just
literally
the
string
parameter
s,
so
it
doesn't
make
it
I
mean
it's
not
as
short
as
J
map
core,
but
it's
not
that
long
and
we're
not
talking
about
something
like
collab,
where
you
want
to
save
every
bit.
That's.
B
E
A
G
G
A
A
D
C
C
C
C
So
here
we
are
creating
a
draft
message.
We're
doing
emails
are
set
create.
This
is
the
creation
ID.
This
is
object,
string
from
the
client
and
then
here
we're
sending
that
draft
and
we're
doing
this
actually
in
a
single
API
request
and
so
sending
is
creating
an
email,
submission
object,
and
you
can
see
here
we
are
referencing
that
email
object,
email
ID
with
its
creation
ID,
so
this
needs
to
be
able
to
resolve
that
to
whatever
ID
was
created
by
the
server
when
it
gets
to
that
point.
C
So
that's
that's
the
that's
the
issue,
so
one
possible
solution,
which
I
think
is
quite
simple
but
also
quite
neat,
is
we
now
have
this
top-level
object
that
for
our
API
requests,
because
we
we
passed
the
capabilities
that
we're
using
as
well
as
list
of
method,
calls
and
I.
Think
if
you
could
optionally
pass
a
existing
hash
of
created,
IDs
a
mapping
of
what
creation
ID
you
had
and
what
serve
the
ID
so
you
created,
then
you
can
just
basically
pass
that
around
to
each
place
you
proxy
to
and
the
when
you
pass
it.
C
If
you
pass
one
in
you
get
one
pass
back
out.
If
you
don't
pass
it
in,
he
doesn't
want
a
passing
one
back
out
again
and
then,
when
it
gets
passed
back
out
in
the
response
you
get,
any
additional
ones
has
been
created.
So
you
can
then
just
pass
that
object
long
to
each
place
that
you
proxy
to
does
that
make
sense
to
people.
Do
people
think
that's
a
good
idea.
Any
other
comments
on
this.
This
something
that
belongs
in
the
standard.
C
You
don't
want
the
proxy
to
have
to
understand,
because
the
trouble
with
the
creation
IDs
is
it
has
to
know.
It
would
then
have
to
know
if
I
go
back
a
slide.
So
this
is
a
string
and
to
know
that
that
you
has
to
know
that
that
is
an
ID
in
order
to
be
able
to
know
actually
I
need
to
check
if
I
should
replace
that
string
with
the
the
ID.
If
you
were,
it
has
to
know
the
structure
of
this
API
call.
That's
the
issue
for
the
property.
C
C
D
C
D
General
I'm,
a
big
proponent
of
you,
know,
implementation
considerations
and
things
to
do,
but
in
this
particular
case
you
actually
have
an
access
in
toxic
elements.
So
probably
doing
it
up
front
would
be
much
easier
than
retrofitting
or
you
know
explaining
to
people
how
to
do
it
after
the
fact.
So,
actually
it's
sort
of
compelling
but
I
I
think
I
would
like
to
think.
If
there
are
better
ways
of
doing
this,
we.
C
C
C
B
C
It
is
the
simplest
paralyzation
which
would
give
you
by
far
the
most
win
is
to
simply
only
paralyze.
If
you
only
have
slash
get
calls
because
those
must
be
immutable
on
the
data
and
your
model
can't
modify
the
data
so
then
you
obviously
you're
safe
to
paralyze
them
and
put
the
results
back
in
the
end.
Well,
actually,
that's
not
entirely
true,
that's
safe!
As
long
as
you
don't
you
don't
have
a
concurrent
process.
That's
writing
stuff,
I!
Guess
so
say
if
you
can
get
some
kind
of
read
lock
so.
B
C
Yeah,
that
was
a
another
suggestion
that
Rob
had
a
different
Rob,
that
the
client
has
to
declare
upfront
all
the
creation.
It's
gonna
bank
reference
I
as
a
client,
author
add-on.
That
would
be
really
tricky
for
our
particular
notation
anyway,
where
all
this
stuff
gets
handled
by
magic
somewhere
further
down
this
solution
was
a
lot
easier.
C
C
C
We
could
still
have
that,
but
that
would
mean
passing
in
a
separate
map
for
every
single
type
that
might
have
been
so
far
so
I'd
probably
change
that
to
be
creation.
Ids
are
just
scoped
to
the
request
in
general,
so
it's
up
to
the
client
not
to
reuse
the
same
key
for
two
different
types
within
the
same
request,
which
shouldn't
be
too
hard
for
it
to
do.
C
Okay,
all
right-
this
was
raised
on
the
list
last
week,
or
so
it's
kind
of
a
small
thing,
but
might
as
well
just
see
if
there's
any
opinions
while
we're
here
when
you
do
a
query
on
like
a
set
of
email,
you
can
either
just
give
it
an
index
say
I
want
the
ten
messages
starting
index
nine
or
something
weird
or
you
can
give
it
a
anchor
an
ID.
That's
gonna
be
the
list
and
say
whatever
position.
That's
that
that's
to
treat
that
as
my
position.
So
you
can.
They
also
give
an
offset
from
that.
C
So
the
question
is
which
directions
the
offset
go.
So
if
I
have
this
anchor
message,
one
two
three
an
anchor
offset
one.
Does
that
mean
my
results
should
start
at
the
first
ID
after
message,
one
two:
three
or
does
it
mean
that
message?
One
two
three
should
be
at
position,
one
in
my
results,
which
would
mean
that
results
are
the
one
before
because
we
index
from
zero,
which
of
those.
B
If
we
look
at
this
sorry,
I
should
have
actually
brought
up
this
back
to
look
at
what,
if
you're,
using
rather
than
using.
Thank
you
using
a
position
in
the
key
is
called
position.
Isn't
it
correct?
Yes,
okay,
and
so,
if
you
have
an
anchor
and
then
you
have
a
position
relative
to
the
anchor,
the
anchor
is
virtually
considered
to
be
at
zero.
If
you
have
no
anger
and
your
position
is
from
that,
could
you
have
anchor
and
then
position
and
have
the
list
start
at
the
anchor
rather
than
start
at
zero.
C
C
C
C
C
So
if
you
client
forgets
to
escape
it,
they're
not
gonna
have
any
issues
which
is
nice
because
they
will
does
anyone
have
any
objection
to
us
just
adopting
that
as
the
restriction
I
think
we'll
just
go
with
that
from
extra
time?
Great!
Ok!
So
that's
pretty
much!
All
for
core
core
is
basically
done
at
this
point.
C
Be
consistent
between
Traynham,
okay,
so
mail
remaining
issues,
so
this
kind
of
feeds
into
what
you
were
saying
earlier,
Lexi
on
the
main
list,
Neal
to
vary
actually
brought
up
the
idea
should
it
not
only
be
allowed
to
include
image
parts
but
also
video
and
audio
parts.
So
if
you
had
someone
that
had
in
spurts
these
in
a
mime
structure,
should
they
be
returned
as
these
inline
parts,
rather
than
as
attachments.
D
C
D
C
D
D
C
C
C
C
C
A
C
C
If
you
don't
need
any
of
the
extra
power,
so
you
can
just
ask
for
a
from
property
and
you'll
get
an
array
of
objects
that
have
email
and
name
properties,
and
you
don't
have
to
care
exactly
how
its
best
find
underneath,
even
though
it's
now
much
more
strictly
specified
and
and
there's
actually
even
we
use
that
stuff
against
our
headers.
But
for
the
body
part,
it's
is
now
more
complicated,
even
in
the
simplest
case
than
it
was
before.
C
We,
you
can
get
that
array
of
parts
and
you
can
get
the
values
for
the
text
bit
of
those
parts
and
the
bodies
values,
property
which
is
on
the
object.
But
it's
it
it's
just
it's!
It's
not
super
complicated,
but
it
is
if
you're,
just
building
a
simple
client
display
a
message,
and
you
look
at
this
for
the
first
time,
it's
a
little
bit
overwhelming.
Perhaps
so
the
question
is:
is
there
value
in
having
an
even
simpler
property
which
you
could
request
from
the
server?
C
B
C
It
if
there's
multiple
parts
like
this,
you
know
an
array
of
parts,
potentially
it
would
concatenate
those
together.
So
what
you
end
up
with
is
a
single
string
that
you
can
pass
to
a
web
view
and
go
display
this,
and
that
is
that
is
your
message
body.
Is
that
worth
having,
or
is
it
enough
complexity
on
the
server
and
actually
no
use
cases
on
the
client
or
I?
Don't
know
thoughts
in.
B
D
C
D
H
C
C
C
E
Is
there
are
a
lot
of
of
yo
web
applications
that
want
to
have
a
little
mailbox
view
of
an
application-specific
mailbox
in
the
corner
and
they
want
to
have
as
little
code
for
that
as
possible
and
yo?
So
I
think
you
know
I
think
our
classic
yo.
This
group
is
really
biased
towards
you
know
our
classic
view
of
mail
as
this
rich
environment,
where
we're
processing
hundreds
of
mail
but
I.
E
C
D
C
D
B
C
C
D
C
C
Ended,
yes,
you
know
you
ended
text
body
parts,
then
you
could
all
see
make
sure
the
HTML
you
generated
from
them
are
safe.
But
if
you're
taking
are
between
HTML
the
trouble
with
sorry
I,
don't
actually
know
think
we
can
do
some
sanitization,
particularly
because
the
trouble
santa's
ation
is.
You
need
to
know
the
context
it's
going
to
be
inserted
into
in
order
to
iliza.
D
C
C
All
right
so
there's
quarter
connect
considerations
that
need
to
go
into
the
spec
and
Chris
has
very
kindly
volunteer
to
deal
with
these
and
has
already
made
a
good
start.
I
just
thought
I'd
run
quickly
through
what
we
have.
We
need
to
reference
the
mailbox,
a
special
use
registry,
which
is
being
creating
extra
and
barriers
post
draft
today
which-
and
we
think
that's
going
to
get
through
pretty
quickly,
I
think
right.
C
So
we
should
have
that
registry
soon
and
then
it's
straightforward
enough
to
reference
that
we're
gonna
create
a
new
registry
of
arrow
types,
I
think
so,
there's
single
place
to
reference,
the
movie
cool.
C
We
need
to
register
dollar,
sign,
dollar,
flag,
etc
in
the
current
registry
of
IMAP
keywords
to
reserve
them
because
they
mapping
to
the
slash
system,
keywords
and
I'm
at
neutral
registry,
well-known,
forge
a
map
and
a
serve
which
I
forgot.
That
Chris
reminded
me
earlier
as
well
for
auto-discovery
serve
so
sorry,
yeah
is
there
anything
else?
There's
a
lot
of
iron
of
stuff
here.
Have
I
missed
something.
D
C
I'm
hoping
this
is
really
fairly
mechanical,
there's
nothing!
You
know
to
decide
here.
Anything
Chris
can
come
of
that
out.
So
overall
status,
both
corn
male,
are
basically
complete.
You
know,
we've
had
look.
There
was
a
few
minor
issues
which
we've
just
talked
about.
I
think
we
have
resolutions
for,
but
there
are
other
those
I
honor
things.
There's
almost
no
github
issues
open.
We
still
need
some
more
examples,
I
think
in
the
spec
and
obviously
yeah
iterations
to
be
done,
but
I
think
it's
pretty
much
semantically
done
Chris.
You
got
a
comment.
Yeah.
E
We
also
need
work
on
the
security
considerations.
I
just
did
a
poll
request
for
the
submission
security,
so
considerations
which
are
which
yo
Franz
post
earlier
today
reminded
me
I
needed
to
write
yeah,
so
that
was
that
was
sure
good.
We
need
more
security
considerations
about
yo-yo
content,
sanitization
and
whatnot
yep.
C
Your
points
are
I
think
I'm
not
kind
of
ease
into
this.
This
is
a
really
good
time
to
review
and
comments
whether
yes,
security
considerations
that
need
to
be
included
typographic,
you
know
just
typos
or
type
errors
and
the
type
declarations
or
obviously
even
anything,
bigger
issues
that
we
haven't
spotted.
Hopefully
there
aren't
too
many
bigger
issues,
but
yeah
I
think
we'd
like
to
get
this
to
last
call
pretty
soon
thought
might
not
actually
publish
before
IETF
one
or
two,
but
at
least
get
it
to
last
call
and
face
it
basically
done
so
yeah.
C
D
Mom
extension
is
at
the
moment
very
simple:
it's
only
deals
with
verifying
as
my
signatures,
because
it
can,
it
doesn't
require
any
user
specific
state.
You
just
need
to
have
trust
anchors
on
the
server,
which
is,
you
know,
hopefully,
state
Ford,
just
a
single
property
that
returns
verification
status
of
signature
and
whether
it
was
successful
or
not.
D
Access
to
the
users
private
key
through
the
account,
then
you
can
also
do
decryption,
and
you
know
I
didn't
include
in
document,
but
this
is
can
be
added.
The
other
side
of
this
is
we
can
possibly
extend
this
for
submission
to
do.
Signing
and/or
encryption
as
well
again
requires
cooperating
server
that
you
can
access
to
the
private
key
somewhere
I.
D
D
B
D
C
D
C
B
E
B
D
E
E
The
question
is,
you
know,
do
you
know?
Do
we
want
to
try
to
pick
up
and
standardize
ex-client
or
something
like
it,
because
you
know
to
glue
a
J
map
submission
to
a
submission
submission
server?
You
need
it.
You
need
something
like
X
Y
and
if
you
don't
glue
a
J
map
submission
to
a
submission
server,
then
you
probably
want
to
implement
milt
or
in
your
j
map
server,
which
tells
you
why
you
want
to
glue
your
j
map
server
to
a
submission
server.
E
Okay,
so
antivirus
any
spam
and
all
the
any
abuse
technologies
like
to
have
as
much
data
about
the
client
as
possible
to
do
rate
limits
and
whatnot
yep.
So
what
X
client
does
is
it
forwards?
Things
like
IP
addresses
whether
use
SSL
I
can
forward.
It
can
even
forward
assassin
I
dent
ities
things
like
that.
So
it's
basically,
you
know
trying
to
give
all
the
context
all
the
context
that
the
front-end
knows
that
the
submission
server
doesn't
know
that
you
use
to
do
rate
limiting
and
security
evaluation.
This.
E
Not
saying
we
have
to
do
this,
but
you
know
if
we
don't
do
this,
we're
kind
of
in
a
position
of
you
know:
you're
either
you
either
have
to
implement
this
de-facto
or
you
have
to
come
up
with
a
private
arrangement
between
the
J
map
server
and
the
submission
server
for
every
submission
server.
You
want
your
J
map
server
to
support,
I.
Think
that's!
Okay,.
B
And,
for
example,
if
you
were
Microsoft
Exchange
to
pull
something
out
of
a
hat,
you
probably
aren't
using
as
soon
as
you
say
internally,
but
you
still
have
a
way
of
getting
something
from
your
mail
store
to
your
I
found
SMT
of
EQ
and
they
wouldn't
be
using
ex-client.
Anyway.
Probably
it's
one
one
of
many
ways
of
doing
it,
and
this
yeah
as
Neil
said
a
bit
suspect
yeah.
C
This
definitely
doesn't
strike
me
as
something
that
should
be
in
the
current
documents,
because
it
does
not
affect
the
protocol
between
the
client.
The
server,
which
is
what
we
specify
I,
could
certainly
their
being
is
part
of
implementation,
notes
of
recommendations
for
J
map
mail,
otherwise,
I
think
this
would
have
to
be
a
separate
well.
E
E
B
D
F
D
E
Yeah
I
in
terms
of
other
technologies
in
this
space
that
solved
the
similar
problem.
There's
a
de-facto
protocol
called
the
proxy
protocol
version
2.0,
but
it
solves
it
at
a
connection
level.
So
it
actually
sends
a
binary
chunk
at
the
front
of
the
connection
that
passes
this
info
and
since
SMTP
you
might
want
to
leave
the
connection
open
and
reuse
it.
It
makes
sense
to
have
one
in
band
of
the
SMTP
session,
but
yeah
SMTP
isn't
a
JSON
protocol.
A
A
And
to
answer
the
other
question
you
actually
can
do
experimental,
but
it
has
to
be
an
iesg
approved
experimental
to
do
stamp
to
do
a
smtp
extension
says.
Each
service
extension
registered
with
IATA
must
be
defined
in
a
formal
standards
track
or
iesg
approved
experimental
protocol
document.
So
you
can't
use
the
is
the
independence
okay.
My.
D
E
D
A
C
E
B
G
A
C
C
B
Yet
it's
not
really
at
a
point
where
it's
it's
ready
to
show
to
the
world,
but
we
have
the
framework
and
we're
starting
to
use
it
for
testing
on
our
server
implementation,
I'm
hoping
to
bring
it
to
the
hackathon
in
Montreal
because
we'll
be
close
to
the
team,
that's
actually
working
on
it
in
the
US,
so
they
will,
they
will
come
up
and
we
will
have
a
gem
of
tests
which
that
that
can
be
used
against.
At
least
the
service
that
we
have.
We
know
about
will
be
tested.
C
E
B
I
made
it
disappear.
Oh
yes,
we
could
absolutely
go
through
github.
B
C
B
C
Other
thing
that
we
go
there
got
away
from
Firefox
is
so
we
have
one
other
document
there
actually
need,
which
is
the
recommendations
arm
as
an
informational
document
for
implementing
J
map,
with
in
coexistence
with
IMAP,
either
as
a
proxy
or
on
the
same
server,
which
Alexei
I
believe
has
volunteered
to
author
and
I
just
wanted
to
remind
everyone.
That
was
the
case.
D
C
D
C
C
Things
one
of
the
only
thing
on
here
that
we
haven't
really
discussed,
but
it's
it's
pretty
trivial
and
I
think
I'm
just
going
to
do
it.
It's
just
at
the
moment.
You
can
designate
one
primary
account,
but
if
you're,
if
you
have
multiple
bits
of
data
that
you
can
access,
you
might
want
a
different
account
to
be
primary
for
different
data.
C
The
example
for
this
well
in
fastball,
our
file
storage,
is
a
separate
account
to
the
mail
contacts
calendars
because
of
the
semantics
of
accounts
and
you
need
copy
blobs
across
you
can't
just
reference
and
stuff
so
they're
separate
accounts,
but
you
still
need
to
say
which
of
those
is
your
primary
files
account
as
opposed
to
your
primary
email
account?
There's
nothing
conflicting
about
that.
So
I
don't
see
any
issue
in
making
is
primary
per
type.