►
From YouTube: May 2020 OpenZFS Leadership Meeting
Description
At this month's meeting we discussed: OpenZFS conference; Mirror Rebuild PR; Documentation system; POSIX AIO.
Details and meeting notes: https://docs.google.com/document/d/1w2jv2XVYFmBVvG1EGf-9A5HBVsjAYoLIFZAnWHhV-BM/edit#
A
We
have
a
pretty
full
agenda
this
week,
hopefully
a
lot
of
time
to
get
through,
but
if
not
we'll
we'll
tackle
it
next
week
or
offline.
So
the
first
thing
I
wanted
to
mention
was
the
opens.
Dfs
conference
carry
hearings
on
so
Karen.
You
can
correct
me
if
I
see
anything
around
here
or
jump
on.
Do
you
want
to
go
ahead
and
and
just
kind
of
overview,
what
we're
going
to
be
doing.
B
B
Obviously,
if
people
aren't
traveling,
it
kind
of
opens
things
up
to
be
whatever
we
want
during
the
year,
but
that's
like
a
reasonably
good
time
for
folks.
So
that's
what
we're
planning
still
working
through
the
details
of
what
it
will
look
like
the
platform's
all
of
that
other
stuff,
but
we'll
be
kicking
off
their
request
for
talks
pretty
soon
in
the
next
couple
of
weeks.
A
A
So
I'll
look
for
an
email
from
me
in
the
next
couple
of
weeks.
Looking
for
talks,
I'd
love
to
hear
about
I
and
I.
Think
the
audience
in
general
would
love
to
hear
about.
You
know
new
features
that
people
are
working
on
or
work
that
you've
done
in
the
past.
You
know
year
or
two
that
we
haven't
covered
at
previous
conferences
as
well,
as
you
know,
talks
from
people
using
ZFS
or
deploying
ZFS
in
interesting
ways.
A
You
know
given
ya,
given
relative
to
their
financial
situations,
so
I
think
that
could
be
really
good
and
we're
gonna
try
really
hard
to
replicate
the
social
and
101
aspects
of
the
conference
because
I
know
that's
you
know.
That's
really.
A
lot
of
the
draw
for
people
is
to
is
to
get
to
chat
with
and
other
people
in
the
community
and
also
kind
of
overhear
conversations
in
here.
I
think
I
heard
some
so
I
was
talking
about
such
and
such
that's.
Cool
I
was
interested
in
that
too
I
didn't
know.
A
A
B
Just
going
to
say,
I
know
that
some
folks
are
attending,
vsd
can
and
some
other
conferences.
Certainly
people
who
are
you
know
mad
and
Paul,
and
others
who
I
work
with
all
the
time
will
be
getting
information
from
them
about
what
went
well
and
what
didn't
go
well
with
the
tools,
but
definitely
we
would
love
to
hear
from
folks
about
like.
Oh,
this
was
a
show,
or
you
know
like
this
was
amazing.
You
should
try
to
replicate.
This
I
would
be
super
helpful,
yeah.
A
A
A
You
know
prepared
talks
with
QA
with
some
kind
of
hallway
interaction
and
then
the
second
day
being
mostly
hackathon
and
I,
think
that
the
hackathon
will
be
you
know
we
need
to
have
a
little
more
structure
or
run
have
been
helping
people
to
organize
it,
but
people
working
in
kind
of
groups
of
1,
&
2
that
you
know
we
probably
all
know
sort
of
how
to
do
that
online.
At
this
point,
so
hopefully
that
part
at
least
will
feel
familiar.
B
A
D
This
is
actually
a
bit
of
work
that
came
out
of
the
D
braid
pull
request.
That's
been
up
for
a
while
now
as
part
of
that
change,
they
were
desire
to
have
a
fully
sequential
rebuild
in
ZFS,
more
like
rejected
traditional
rebuild
instead
of
three
silver,
where
you
go
through
the
device
strictly
and
LBA
order.
D
So
this
was
pulled
out
of
that
pull
request
and
we
worked
a
little
bit
as
preparatory
work,
so
we
can
get
it
in
living
advantage.
4D
right
here
is.
We
can
get
that
significantly
faster
rebuild
times
with
the
D
braid
work.
If
we
know
which
devices
we
need
to
rebuild
right,
we
don't
have
to
rebuild
the
whole
pool,
but,
as
it
turns
out,
we
can
also
use
the
same
logic
for
mirrors.
D
So
the
pull
request
as
it
stands
is
a
implementation
of
that
that
adds
a
new
command
line,
option
to
Z
pool
replace
as
equal
attach,
where
you
can
request
this
sequential
rebuild,
and
it
only
works
for
mirrors
like
I
said,
but
it'll
go
through
and
basically
looking
just
at
the
space
maps.
Sequentially
walked.
The
entire
allocated
space
and
issue
reads
to
force
rebuilds
for
the
mirror
devices.
The
pull
request
is
up
now.
There's
some
performance
numbers
in
it.
D
The
bottom
line
is
it
performs
as
well
as
reso,
prefer
larch
blocks,
and
you
see
a
pretty
nice
performance
bump
if
you've
got
a
pool
full
of
small
blocks
and
they're
reasonably
well,
packed
right
that
I
write
the
spaces
between
them,
so
you
can
get
large
iOS,
so
it'll
issue
larger
iOS
that
span
multiple
blocks
without
the
need
for
like
three
silver,
this
went
to
resilvered
work.
That
was
done
a
couple
of
years
ago.
There's
a
lot
of
this
sorting
in
memory
did
achieve
a
similar
effect.
D
This
works
the
same
way
except
it
doesn't
need
the
in-memory
sorting
because
we're
just
walking
in
space
on
disk,
so
everything's
up
there
now
we're
just
looking
for
reviewers
and
feedback
on
it.
So
any
feedback
on
that
would
be
welcome.
I,
don't
have
the
pull
reference
number
handy,
I
think
I'll
pop
my
head,
but
everything's
there
cool,
I,
don't
know
3,
4
9,
don't
know
3
4
9
any
feedback
will
be
welcome.
D
There
are
a
couple
things
a
little
different
about
it
compared
to
a
normal
or
a
silver
I
guess
the
big
one
is
that
because
it
only
operates
at
the
top
level
v-dubs,
you
can
actually
have
multiple
of
these
going
on
at
once,
unlike
a
normal
reach
silver,
which
is
a
pool
wide
operation,
so
you
want
to
have
rebuilds
going
on
different
top
level.
B-Dubs,
that's
fine!
It
supports
that
kind
of
thing,
but
other,
but
it's
still
reports
as
if
only
one
burn
going
on
so
from
the
users
based
perspective.
A
D
The
original
version
of
the
code
integrated
with
the
existing
stuff,
but
it
was
reworked
to
basically
be
at
own
standalone
infrastructure,
so
the
code
is
entirely
separate
from
the
DSL
scan
code.
At
this
point,
there's
some
minor
things
that
touch
the
same
code,
about
updating
like
the
ETL
but
as
necessary.
D
It
just
adds
the
command-line
option.
Does
equal
replace
a
zpool
attached?
Okay,
you
can
give
it
the
data
our
option.
Better
options
are
welcome.
I'm
not
totally
excited
are,
but
you
just
asked
for
it
by
default.
If
you
give
it
no
options,
attached,
Andrey,
silver
or
attach
and
replace
do
the
normal
or
silver
thing.
If
you
do
the
rebuild
option,
then
you
just
request
with
the
command-line
option:
oh
hey,
the
other.
D
Sorry
good
as
I
say,
the
other
big
got
you
with
this
is
because
it's
not
doing
a
traditional
every
silver,
no
checksum,
there
verified
as
part
of
the
rebuild
process.
So
that's
the
other
big
caveat
with
this.
So
after
you
do
a
rebuild,
the
documentation
is
updated
as
50,
you
strongly
run
a
free
silver
or
a
scrub
of
the
pool
afterwards
to
verify
the
check
sums.
E
D
So
the
testing
shows
it
depends.
So
if
the
block
sizes
are
over
about
128
K
and
performs
as
Bob
as
well
as
every
silver
does,
maybe
a
little
bit
better.
Actually
so
if
the
block
sizes
are
on
average
larger
your
pool,
it
doesn't
buy
you
much
if
they're
much
smaller,
then
it
buys
you
quite
a
lot
like
a
the
testing
showed
for
like
under
16
K
block
size,
it'll
be
silver
like
twice
Azure
rebuild
about
twice
as
fast
as
a
result
for
so,
if
you
knew
what
the
average
block
size
was.
F
D
D
It
went
well,
and
it
should
I
mean
it's
not
doing
using
all
the
same
rebuild
code
that
the
normal
ZFS
rebuild
does
so
it
or
resolver
does
the
only
thing
it's
not
doing
is
verifying
the
checksum
at
the
end
they
re
trying
it,
so
it
should
be
no
less
reliable
than
every
silver.
You
just
don't
get
the
cross-check
until
later.
G
A
H
A
Well,
yeah
that
might
be
right
well,
what
I
was
going
to
mention
is
on
it.
The
getting
the
rebuild
it
gets
you
almost
as
good
as
you're
gonna
get
with
two-way
mirrors.
The
the
really
great
thing
about
having
a
checksum
is
that,
if
you
have
let's
say
three-way
mirrors,
you
lose
one
of
them,
then,
with
the
rebuild
object,
it's
just
going
to
be
getting
the
data
from
like,
basically
one
of
the
other
one
of
the
remaining
two
devices
in
the
mirror
chosen
randomly.
A
A
A
The
so
I
think
it's
kind
of
interesting
that
this,
like
a
is
technically
infeasible
to
do
with
raid-z
but
B,
even
if
it
was
technically
feasible,
it
would
be
less
well
advised
to
do
on
grade
Z
because
you
know
raid-z,
200,
Z
3
are
very
frequently
used
and
like
with
rids
e
2.
If
you,
if
you
could,
you
know,
rebuild
one
device
from
the
other
ones,
but
you
don't
know
which
other
wants
to
use.
D
E
A
A
Faster
and
I
feel
like
if
it's
recommended
might
as
well
kick
it
off.
If
somebody,
if
somebody
thinks
like
the
the
performance
objective
of
doing
that,
scrub,
I,
don't
like
it,
they
can
always
cancel
it,
they
can
or
they
can
add
a
flag.
You
know
we
can
have
a
ZFS
or
a
zpool
replace
you
know,
rebuild.
E
A
D
D
A
D
A
H
Okay,
I
all
started,
so
we
have
a
problem
in
our
project
that
our
documentation
is
very
fragmented
now,
so
we
we
had
as
it
affects
wiki.
We
have
open
city
first,
but
oak
media
week
and
other
things.
In
addition,
we
have
Monday
main
pages
now
in
Windhoek
format,
so
we
have
a
fragmented
documentation.
We
have
from
end
documentation
in
different
formats.
So
what
I
have
done
in
last
week?
I.
H
E
A
So,
just
so,
I
can
understand
about
what
the
goals
are
of
the
changes
that
you're
making
the
in
the
short
term
you're
talking
about
replacing
the
ZFS
Linux
wiki
with
this
new
thing
and
is
all
the
content
from
the
wiki
already
moved
over?
Yes,
all
of
the
content,
and
then
what
are
your
plans
for
documentation
dominance
of
the
open
ZFS
world
like
alright?
Do
you
envision
that
this
will
eventually
will
move
other
content
on
to
this
and
it'll
be
able
to
replace?
H
H
D
F
Like
Mandark
works
everywhere
for
manual
pages
feels
like
it
would
be
good
to
keep
the
manual
pages
in
the
man
doc
sort
of
format
as
much
as
possible
and
like
the
trough,
stuff
move
towards
man.
Doc
I
mean
that
I
mean
the
original
pages
were
in
trough
because
they
came
from
doc
book
I
think
a
million
years
ago
it
was
a
mechanical
conversion
to
trough
that
didn't
work
very
well
and
we've
been
cleaning
that
up
ever
since,
and
the
man
Docs.
The
new
stuff
is
man
doc,
which
I
believe
works
everywhere.
F
H
To
convert
to
man
doc
everything
and
live
with
it,
convert
it
to
HTML
and
built
it
nearly
and
with
our
other
documentation
in
rst.
The
other
way
is
to
look
forward
to
convert
on
to
our
steep
and
generate
on
man
documentation
from
from
it,
because
our
state
format
converts
into
main
pages
easily
by
a
sphinx.
H
A
H
A
A
A
F
A
semantic
tags
that
man
doc
has
like
the
the
mando
package,
for
instance,
like
has
specific
tags
for
like
this,
is
a
library
that
you
would
link
against
or
something
in
an
interface
document.
It
has
some
pretty
specific
tags,
specifically
for
manual
pages,
things
like
cross
references
to
all
the
pages
and
they're
in
band
page-specific
sort
of
things
in
it
yeah
part
of
what
makes
it
complicated,
but,
on
the
other
hand,
like
that's
the
the
full
richness
of
the
document.
A
A
F
H
C
H
H
So
to
come
with
something
I
I
think
that
and
now
we
should
not
do
any
fast
moves,
of
course,
but
I
want
to
look
in
this
question.
I
would
if
I,
okay,
if
things
I,
can
give
us
everything
that
man
doc
gives
us
after
a
conversation,
two
men
doc.
Is
it
okay
for
us
to
store
our
ste
or
something
else
or
not?
I.
F
Mean
to
Alan's
point
we're
interested
in
getting
the
common
reaper
building
on
illumise,
for
instance,
and
to
eventually
replace
the
edifice
that
we
have
in
the
core
of
our
in
our
core
repository
with.
Like
a
snapshot
of
this
one,
it
would
be
unfortunate
to
require
an
additional
documentation
tool
chain
to
build
the
manual
pages
that
worship
so
like
for
us.
We
had
man
talk
already
like
freebsd
does
in
vase
and
like
it
would
be
ashamed
to
have
to
in
order
to
build
the
operating
system.
D
D
C
Think
the
OpenBSD
people
that
maintain
man
doc
have
a
tool
that
they
use
for
all
their
main
pages
to
generate
HTML
and
it
seems
pretty
nice
I'm,
not
I'm,
not
dug
into
it.
That
much
but
I
think
there's
information.
The
link
Dexter
put
in
the
chat
there
from
the
man
doc
website
that
goes
more
into
the
HTML
conversion
tool.
But
yeah
like
Josh
was
saying
you
know
when
you
write
it,
you
can
really
markup
a
lot
like
you
know.
C
This
is
an
optional
argument
versus
a
required
argument,
and
so
on
and
like
Matt
was
saying,
though
you
know,
unless
you're
experienced
with
reading
man
pages
on
the
terminal,
you
don't
necessarily
notice
that
the
angle
bracket
is
versus
the
square
brackets
mean
something,
whereas
in
HTML
you
know
it
can
be
different,
colors
and
fonts
and
and
things
that
make
it
more
obvious
that
you
know
this
argument
is
different
than
that
argument.
I
mean.
A
That
I
think
that
information
is
I
mean
it
may
not
be
pretty,
but
it
is
in
there
in
the
man
doc
and
it
is
rendered
in
the
text
with
like
tango
versus
square
brackets.
Like
you
said,
I
mean
you
might
not
know
that
you
might
get
confused,
but
at
least
the
information
is
semantically
there,
which
seems
like
you
could
in
theory,
render
that
more
prettily
in
the
way
in.
C
J
A
Pretty
easy,
so
it
sounds
like
the
four
man
pages.
There's
really
a
lot
of
things
to
consider,
especially
you
know,
since
we
already
have
something
that
we
have,
that
mostly
works
that
some
people,
love
in
a
lot
of
people
are
familiar
with.
Maybe
nobody
loves
it,
but
at
least
a
lot
of
people
are
familiar
with
it.
C
H
A
So
you
know
my
biggest
concern
with
it
is
the
content
not
with
the
mechanism,
but
if
you
know
searching
to
this
other
mechanism
and
formatting
helps
us
to
like
get
more
contributions
and
help
us
to
keep
the
content.
Up-To-Date,
I
think
that'd,
be
great.
I,
think
like
using
github
and
pull
requests,
regardless
of
the
actual
format,
would
be
wonderful,
because
I
think
we
will
get
more
people
contributing
then
on
the
media
wiki.
So.
C
E
C
F
A
brief
experience
report:
we
did
something
very
similar
with
illumise.
We
had
a
confluence
it
just
I
agreed,
not
focusing
too
much
on
the
mechanism
is
important,
but
I
think
that,
just
if
we
look
at
all
of
the
times,
there's
been
a
wiki.
They
always
end
up
with
just
a
lot
of
spotting
this
and
then
a
lot
of
out-of-date
stuff
and
I
have
to
wonder
if
there
is
not
something
about
them,
the
way
that
people
interact
with
them.
F
That
leads
to
that
and
then
we
converted
to
not
knowing
spent
something
very
similar,
Mike
talks
or
something,
and
so
we
have
we
chucked
out
the
wiki
and
we
copied
just
the
pots
of
the
wiki
that
were
actually
correct
into
a
github
repo
thing
and
that's
how
we
generate
that
pop
of
the
site.
Now
it's
gone
very
well.
Sorry
I
agree
that
that's
a
good
plan.
C
A
C
The
process
of
doing
the
same
thing
replacing
our
old
doc
book
based
documentation
with
a
hugo
based
thing
that
that
builds
documentation,
then
so
that
people
can
edit
it
with
github.
Instead
of
having
to
learn
man
doc
to
contribute
documentation,
and
we
expect
to
get
a
lot
more
contributions
that
way.
Yeah.
A
Thank
you
very
much
Church
for
for
making
this
happen
like
four
foot
having
a
good
idea,
and
especially
for
my
dating
a
good
idea,
because
I
know
this
is
something
like
I
mean
just
to
give
an
example.
How
many
years
did
we
talk
about
how
many
years
went
by
where
we
said?
Hey
really
had
a
hackathon.
We
should
split
the
ZFS
man
pages
out
into
one
man
page
for
sub-command
and,
like
everybody,
thought
it
was
a
good
idea
and
nobody
did
it
for.
A
A
A
A
Well,
you
know,
ZFS
is
kicking
off
its
CIOs
or
whatever,
and
so
this
the
idea
of
the
async
dmu,
is
to
make
those
expose
the
same
kind
of
asynchronous
interfaces
that
we
have
with
the
z
at
the
CIA
level
level
of
like
please
issue
this
call.
This
call
back
when
it's
done,
but
meanwhile
my
thread
can
go,
do
whatever
it
once
exposing
that
kind
of
interface
at
the
d
me
level,
so
you
can
say
you
know.
A
Calls
like
can't
take
advantage
of
this,
because
the
whole
system
call
works
is
like
you
know,
you
do
the
read
system
call
it
sits
there
until
you
have
the
data
and
then
you
return,
but
there's
a
POSIX,
a
I/o
asynchronous
I/o
interface
that
could
be
hooked
up
to
this
async
D
new
interfaces.
So
this
question
was
basically
is
it
worth
it
like?
Is
anybody
using
POSIX
or
interested
in
using
POSIX
a
IO
have
any
experience
with
it
that
could
kind
of
give
him
feedback
as
to
if
it's
like?
A
If
nobody
knows
nobody
cares,
then
I
sounds
like
you:
won't
do
it,
but
if
people
are
are
like
yeah,
that
would
be
great.
If
ZFS,
you
know
worked
right
with
that,
then
I
would
use
it.
Then
you
know
he
would
prioritize
that
so
open
the
floor
to
comments
or
experiences
or
requests
about
POSIX
ai-ya-ya.
G
K
G
K
I
recall,
actually,
can
you
guys
hear
me?
Yes,
we
can
I,
wasn't
actually
trying
on
speaking,
but
so
I'm,
actually
a
guy
whose
fault
it
is
that
we
do
a
I/o
this
way
and
yeah.
Originally,
when
I
was
implementing.
This
I
went
to
the
spec
and
I
was
like
well
I.
Can
do
this
too
two
ways
like
in
just
you
do
compatibility
shim,
which
will
make
certain
database
people
happy
as
they
had
been
talent
going
and
saying,
give
us
a
IO.
K
Please
I
think
it
was
my
sequel
people
in
the
few
others
as
the
default
configuration
required
and
they
had
to
disable.
It
was
a
huge
pain
for
them,
so
I
was
like
I
could
do
it
this
way
or
I
could
hook
into
a
DM
you
and
if,
at
the
time
it's
a
team
to
team,
you
was
so
much
work.
I
was
like.
Let
me
just
do
a
compatibility
or
so
that
does
put
a
current
situation.
K
I
A
I
Sorry,
I
also
wanted
to
mention
that
Pandora
uses
air
yo
for
Samba
excesses
and
basically
it
gives
better
performance
than
using
regular
POSIX
IO,
but
at
this
time,
I
all
in
freebies
DS,
implemented
by
using
a
special
pool
of
threads.
So
I
said
essentially
it's
like
the
threads
are
blocking
their
fasts
and
the
vent.
A
complete
works,
identify
its
actual
consumer
threads.
E
G
That
would
require
that
the
a
IO
library
or
system
calls
do
not
do
it
themselves,
but
instead
know
enough
to
call
down
to
the
file
system.
Most
implementations
I
see
they're
seen
in
fact
just
implemented
as
threads
themselves,
and
you
know
it's
implemented
the
system
call.
They
are
not
at
the
VFS
layer.
D
K
D
K
G
So
again,
the
question
is:
is
there
any
system
where
having
Zetas
Fest
implement,
it
would
actually
be
usable,
I
think
Matt
actually
added
a
VFS
up
for
it.
What
other
systems
be
willing
to
do
that,
or
is
it
best
to
just
say,
stick
with
the
existing
AIO
mechanism,
which
most
systems
do
have,
however,
there
they
implement
them.
A
A
A
C
Yeah
so
I
not
much
questions
on
the
first
export
one.
Yet
it's
kind
of
just
a
heads
up
the
work
hasn't
actually
started
yet,
but
hopefully
we'll
start
in
time
for
there
via
status,
update
on
the
next
meeting.
So
there's
a
feature
we
built
last
year
for
a
customer
to
deal
with
the
case.
They
have
multiple
pools
on
the
same
machine.
C
One
of
them
is
actually
running
off
virtual
devices
that
are
in
the
cloud
remote
in
some
way,
and
so
when
there
is
problems
with
the
connectivity
to
the
the
virtual
disk
there,
that
pool
could
hang
sometimes
while
holding
something
like
the
spa,
namespace,
lock
or
something.
And
then
you
couldn't
operate
on
the
other
pools
and
you
know
once
the
pool
when
suspended,
and
it
was
you
know
you
couldn't
export
it,
because
when
you
do
an
export,
it
wants
to
write
to
the
disks
and
so
on.
C
It'll
be
a
little
bit
of
work,
since
a
number
of
things
have
changed
since
the
original
work
especially
send
and
receive
with
raw
and
redacted
changed
quite
a
bit
of
those
files.
But
that's
we'll
be
working
on
that,
but
yeah
the
one
I
wanted
to
ask
more
about
was
I
guess
it
was
two
weeks
ago
when
Dell
fix-up
streamed
their
bits
for
the
boot
once
feature
or
next
boot
and
they're
actually
confusingly
named.
C
That
just
says
a
text
blob,
but
they
added
a
type
for
an
envy,
less
based
one
and
we're
interested
in
implementing
that,
and
in
particular,
one
of
the
problems
was
the
interface
they
created
in
libs.
Nfs
core
only
takes
a
pointer,
not
an
Envy
list,
or
something,
and
so
we'd
like
to
get
input
from
other
people,
so
that
when
we
add
an
interface,
two
lives
out
of
a
score.
A
C
The
first
question
is,
you
know,
how
committed
are
we
to
the
interface
that
was
committed
two
weeks
ago
and
I
think
because
lives
at
a
fest
core,
rather
than
just
lives
at
a
vest?
Maybe
we
don't
want
to
take
that
away,
although
we've
never
done
a
release
with
it
yep
so
maybe
I
don't
know
it
doesn't
make
more
sense
to
to
make
it
always
be
a
nebula
stand
and
we
can
have
a
nebulous.
Has
you
know
a
string
on
it
or
do
we
just
add
a
second
function
or
what
I
mean
with.
A
F
C
Into
the
ambulance
and
the
idea
for
our
embolus
thing
is
that
we'd
actually
have
a
pair.
That
was
another
MV
list
that
we
died
today.
So
the
I
octal
won't
change,
just
the
libs
edifis
core
interface
and
there's
some
value
of
having
the
the
text.
But
you
know
char
pointer
version
for
the
thing
you
suppose
the
user
space,
because
it
means
the
program
consuming.
It
doesn't
have
to
be
able
to
construct
an
MV
list,
although,
if
they're
dealing
with
libs
at
a
festive
might
already
be
used
to
that.
A
Yeah
I
think
this
is
probably
minor
enough
that
we
could
work
it
out
in
code
review.
I.
Think
that
having
cute
I
I
I've
resisted
this,
but
I
am
totally
fine
with
like
just
having
a
bunch
of
function,
calls
just
you
know
you
can
call
it
under
bar
right.
It's
blah
blah,
blah
and
O'brien.
If
you
like,
store-bought
to.
H
A
C
A
A
H
K
K
Otherwise
it
could
cause
farms
in
theory
with
distributions
or
I'm,
not
sure
if
anyone
is
actually
looking
into
it
for
that
to
be
an
issue,
but
it's
just
something
that
downstream
people
usually
want
to
see,
especially
central
package
manager.
It
depends
on
that
in
order
to
install
backwards
compatibility
in
certain
instances.
A
G
Than
that
I
just
wanted
to
bring
up
again
that
I
have
this
PR
out
there.
We
had
discussed
a
year
or
so
ago
the
issue
with
having
properties
that
are
dependent
or
would
be
interpreted
differently
or
would
need
different
values
depending
on
the
OS
or
other
implementation,
details,
feedback
or
comments.
Please,
that's
it.
A
Very
cool
I
think
that's
I.
Look
forward
to
seeing
that
integrated
I
think
it's
a
it's
a
good
idea.
George
Wilson
I
know
you
had
done
some
work
and
maybe
you
were
the
one
who
propose
that
D.
Is
that
something
that
you
think
you'll
have
time
to
look
at?
Let's
say
before
the
next
meeting:
yeah
I
should
have
time
to
do
that
cool
thanks.
D
A
I'll
just
say:
I'm
working
on
the
problem
with
Ziff
s,
end
where
you
have
an
incremental
that
toggles,
the
capital
L,
which
is
for
sending
large
blocks,
and
you
should
see
a
pull
request
from
me
on
that
hopefully
later
this
week
and
that's
it
so,
let's
we'll
reconvene
four
weeks
from
today,
but
I
think
it'll
be
at
the
earlier
time.
So
it
should
be
the
23rd
of
June.