►
From YouTube: 🖧 IPLD weekly Sync 🙌🏽 2020-03-09
Description
A weekly meeting to sync up on all IPLD (https://ipld.io) related topics. It's open for everyone and recorded. https://github.com/ipld/team-mgmt
A
B
A
Probably
has
always
started
with
myself
so
M
in
the
past.
Oh-
and
please
put
your
name
on
T
on
the
attendees
list.
I
basically
started
also
to
put
basically
the
attendances
from
the
last
week
and
then
basically
ask
everyone
who
is
not
attending
to
remove
himself
and
so
yeah
I
work
on
the
rough
side
of
things,
and
so
the
biggest
news
is
that,
finally,
you
rust,
the
Lea
p2p
is
using
the
upstream
version
of
Braska
multi
hash
and
it
had
merged
after
having
a
fault
for
one
and
a
half
years.
A
So
that's
cool
and
also
I,
so
I
moved
from
rest,
Marty,
Heche
over
to
rust
motivation
and
to
rust,
CID
and
all
of
those
have
been
updated
with
all
the
folks
that
are
out
there
all
the
major
forks
and
I've
merged.
All
the
changes
in
and
I
think
I'm
basically
done,
and
this
also
means
that
the
rust
IPS
team
finally
was
able
to
move
the
rust
IPL
these
stuff
to
the
upstream
versions
and
doesn't
use
any
function
anymore,
because
those
were
all
soft
projects.
Basically,
this
has
been
done
in
the
past.
A
Like
I,
don't
know
eight
weeks
or
something
basically
getting
everything
up,
streamed
and
working
and
so
on,
and
also
by
doing
this,
also
there's
no
proper
CI
and
all
those
repositories
and
so
on,
and
the
next
big
thing
is
kind
of
changing
the
licenses
from
MIT
to
MA
he
and
Apache,
but
that's
like
not
a
high
priority.
It's
just
like
kind
of
in
the
background,
but
I'm
also
working
on
those
ones
and
I
also
had
an
idea
last
Friday
about
how
we
can
support
custom,
multi-core
cables-
and
that's
this
has
been
an
issue
in
JavaScript.
A
So
basically,
people
just
import
a
custom
type,
don't
even
see
that
it
is
a
generic
and
just
use
it,
and
everything
will
be
fine
if
they
use
the
default
tonic
tables
that
draw
and
everything,
because
this
is
what
you'd
normally
would
want
to
do.
But
if
someone
really
wants
to
have
their
custom
query
table,
it
just
used
the
generic
type
directly
pets
in
their
code,
a
table
with
the
implementations
and
everything
it's
kind
of
like
you'd
like
quite
a
bit
of
work.
But
let's
expect
it
like.
A
If
you
want
your
own
custom
coding,
table,
you'd
normally
know
what
you're
doing
and
you
take
the
effort
of
drawing
it
in
and
so
hopefully-
and
this
is
what
I
need
to
check-
is
that,
because,
once
we
compile
from
Russ
to
webassembly,
you
have
cost
on
board.
They
have
like
a
huge
panel
size
or
you
also
might
use
different
hash
libraries,
because
some
of
them
use
I,
don't
know
like
assembler
code
or
something,
and
so
you
want
to
have
like
custom
ones,
but
yeah.
B
A
Don't
want
to
draw,
you
know,
see
faces
in
sauce.
Also
basely,
that's
the
goal
and
it
looks
like
generis.
Might
work,
yeah
I'm
still
looking
into
this,
but
that
looks
promising.
So
that's
cool
okay,
so
that's
all
I
have
next
one.
C
All
right
had
a
fairly
slow
week
between
sick,
kids
and
other
stuff,
but
I
did
get
started.
A
jeaious
implementation
of
the
compiler,
a
pretty
good
progress
on
that
and
finally
got
to
addressing
some
of
the
comments
in
the
PR
and
my
goal
this
week
is
to
finish
the
J's
compiler
turns
out
the
generation
parts
really
easy.
I
just
generate
a
JavaScript
object,
and
then
you
can
pass
that
to
the
block
API
and
since
the
there
one
single
block
it
makes
it
really
easy.
C
So
I
just
need
to
work
through
the
full
parser
and
building
the
AST
and
make
sure
all
that
is
not
too
hard
because
I'm
keeping
in
mind
that
we're
getting
multiple
implementations
of
this.
So
if
the
spec
needs
changing
to
make
it
easier
to
compile
now
would
be
a
good
time
to
change
that
and
that's
it
I
think
I
addressed
most
everyone's
comments
in
the
PR.
So
if
anybody
wants
to
follow
up
on
those
I
think
most
of
them
a
result,
that's
pretty
much
all
I'm
doing
right
now.
D
So
I've
opened
a
couple
of
issues,
one
and
that's
probably
interesting
to
everyone
here
about
thinking
about
our
layering
scheme.
Right
now,
I
think
the
layering
scheme
is
documented
as
the
block
layer,
the
data
model
layer
and
then
the
schema
layer
and
I'm
building
that
we
maybe
want
to
separate
blocks
from
codecs
and
then
do
something
with
the
rest
of
the
low.
D
Or
in
scattered
places,
but
I
think
like,
for
example,
be
trying
to
use
clear
linear
layers
between
the
data
model
and
new
schema
layer.
Sorted
makes
sense,
but,
like
advanced
data,
layouts
aren't
really
fitting
into
this
layer
cake.
It
will
clear
away
right
now.
Its
inclusion
cook
on
it
until
something
a
little
more
cogent
comes
out
for
us
all.
D
I've
also
been
trying
to
draft
a
little
bit
of
a
write-up
about
all
the
things
about
strings
know
something
they
ended
up
talking
about
for
a
while
and
call
last
week
and
there's
a
bunch
of
stuff
where
we're
all
just
kind
of
going
by
our
best
intuition.
What
we
think
is
perfectly
reasonable
so
far
and
probably
time
to
start
nailing
it
down.
D
So
that's
turned
out
to
be
a
kind
of
a
large
group
endeavor,
just
because
there's
so
many
assumptions
that
are
so
pleasing
and
so
easy
to
make
that
identify
and
make
them
sometimes
quite
difficult
right,
so
I'm
trying
to
break
down
a
list
of
like
places
where
the
string
questions
show
up
which,
like
kind
of
applied
question
with
it
like.
Is
this
a
question
about
correctness
of
model,
or
is
it
a
question
about?
How
do
we
maintain
sandy?
D
Loving
standards
are
like
no,
because
there's
a
couple
of
different
things
going
on
I'm,
trying
to
figure
out
what
they
all
are.
If
somebody
has
thoughts
on
that,
let's
just
send
me
their
own
tracks.
Energy
disc
will
be
fun,
otherwise
I'm
hoping
to
have
some
sort
of
writing
about
that.
Come
out
in
the
next
week,
but
first
I
was
going
to
try
to
jump
straight
to
answers
and
geography.
D
Also,
a
man
called
thus
I
think,
although
on
Mitas
yeah
posted
an
issue
about
we're
finding
the
mission
statement
and
vision
statements
and
getting
some
of
our
intro
docs
on
that
stuff
up
today,
there's
some
really
interesting
discussion
going
on
there.
So
anybody
else
listening
wants
to
jump
in
on
that
I
think
it
should
also
go
look
in
the
notes
for
initiating
passive
phone
conversation.
That
is
most.
A
E
Sorry
I
forgot
that
I'll
fix
that
yeah,
so
mostly
still
just
a
big
data
stuff
I'm
now
dealing
with
like
a
lot
of
issues
that
only
happen
after
like
hours
of
processing,
so
that's
super
fun
to
debug
and
see
if
you
fixes
work,
but
it's
coming
along,
dag
DB
the
this
sort
of
side,
cardigan
I'm,
doing
to
keep
myself
sane,
often
that
replicates
now.
So
that
looks
like
a
big
milestone.
It's
actually
quite
nice
at
some
point.
Maybe
you
like
later
this
week.
E
I
learned
a
lot
building
this
about
kind
of
how
you
need
to
use
a
lot
of
our
primitives,
we're
gonna,
build
databases
and
it's
sort
of
reinforced
like
how
important
some
of
the
block,
API
stuff
is,
and
also
a
lot
of
like
how
we
thought
about
how
easy
syncing
will
be
was
a
little
bit
naive,
there's
actually
some
pretty
kind
of
difficult
cases
and
decisions
to
make
that
are
not
so
straightforward,
so
yeah,
it's
a
it'll,
be
fun
and
that's
it.
For
me.
Thank.
F
The
go
course
was
good,
but
it
was
just
too
dense
too
quick,
as
Peter
can
can
probably
attest
to,
and
so
disentangling
that
has
been
a
process
so
catching
up
on
old
PRS
that
I
needed
to
review
and
and
stuff
discussions
has
been
a
little
bit
delay,
but
the
highlights
of
things
I've
gone
through
is
Tim's.
Selector
spec
finally
got
a
chance
to
engage
with
that
and
really
interesting
work.
F
F
Now
it's
it's
still
just
the
same
style
car
file,
but
it's
it's
a
really
good
intermediate
step
to
get
to
this
car
v2,
potentially
where
we
want,
where
one
of
the
ideas
is
to
store
the
selector
in
the
car
files
well,
along
with
the
route
that
the
selector
goes
with
and
so
yeah.
So
that's
propel
that
wasn't
merged
overnight,
but
it's
ready
to
be
merged
and
go
out.
F
So
that's
pretty
nice
I've
been
looking
through
Eric's
a
massive
pull
request,
49
on
like
ble
prime,
which
has
got
the
node
assembler
notes,
dial
stuff,
a
pretty
cool
work.
These
are
the
culmination
of
a
lot
of
time
that
Eric's
been
doing
this
in
the
past.
However,
many
weeks
it's
been
months
so
just
trying
to
be
helpful
there,
but
also
trying
to
understand
it
as
well,
so
you
can
actually
use
it,
but
that's
that's
really
the
whole
to
me,
nothing
other
than
that
major
stains
out.
B
Adding
the
car
import
and
export
go
ipfs
because
of
the
extra
concerns
that
we
have
with
pinning,
with
GC
being
able
to
kick
in
while
a
current
Portus
surroundings,
and
things
like
that,
there
was
a
lot
of
thinking
that
had
to
be
done
on
how
to
actually
you
know
put
all
this
together.
I
am
pretty
sure
that
by
tomorrow,
we'll
have
financier
for
this
and
I'll
tag.
Some
of
you
rod
and
Aiko
as
well
to
take
a
look
over
that
and
on
the
side
of
that.
B
B
A
B
F
B
F
B
Keep
them
in
memory,
get
everything
else
in
and
then
go
back
and
say:
okay
from
what
I
got
in
and
what
I
already
have
in
the
store
are
those
roots.
Actually,
something
that
is
complete
or
I
cannot
actually
pin
them.
Do
I
need
to
error
and
stuff
like
that,
so
even
on
reading
it
seems
that
it's
backwards,
at
least
for
the
import
use
case.
Michael.
F
E
Tail,
isn't
it
wouldn't
work
for
me
because
so
assuming
I
need
the
header
at
any
point
in
time
the
tail
is
open
because
it's
actually
quite
hard
to
read
backwards
out
of
an
s3
bucket
so
and
it's
a
variable
length,
so
it's
a
fix
like
better
I
could
read
it
because
I
can't
look
at
the
file
size
and
then
kind
of
go
backwards,
but
since
I
don't
like
there's
no
way
so
that's
just
kind
of
like
a
non-starter
in
my
opinion
like
if
any
information
is
in
the
header
that
you
need
ahead
of
time,
then
it's
not
gonna
work
in
a
lot
of
cases.
B
Whenever
I
think
things
out,
it's
it's
obvious
that
it's
difficult,
you
know
to
have
to
have
a
header
ahead
of
time.
So
streaming,
our
basic
exporting
is
a
little
bit.
You
know
tricky,
but
I
was
actually
wondering
from
import
like
I,
have
a
ganar
file
and
I'm
importing
it
in
I
still
need
to
look
at
the
header
last,
because
I
need
to
go
to
everything
in
the
car
put
it
in
with
me
into
my
plot
store,
and
only
then
I
can
say,
can
actually
pin
this
stuff
in
the
routes
or
I.
E
B
B
Finishes
it
will
help
you
GC
everything
away
and
that's
why?
If
we
have
multi
multi
part
car,
we
actually
have
to
import
everything
and
only
then
release
the
lock
so
that
that's
why
it
was
a
little
more
involved,
because
if
you
have
car
number
one
that
cares
half
of
the
dag
car
number
two
that
has
the
route
2
p.m.
and
the
rest
of
the
deck
which,
combined
with
the
first
car,
actually
gives
you
a
full
deck.
Only
then
you
can
actually
into
it
and
additionally
enjoy
TFS.
E
B
Yeah,
of
course,
okay,
and
this
what
I
meant?
Basically,
it's
not
a
problem
in
terms
of
like
you
know,
simple
method
of
programming,
but
logically
I
need
the
header
last
anyway,
when
I
am
importing,
it
gives
me
nothing
I,
basically,
just
read
the
header.
The
only
thing
that
I
do
before
actually
started
in
the
car
file.
Is
this
version
one
or
not?
That's
only.
E
The
search
further
like
so
for
so
determinism
deterministic
car
files
aren't
just
the
entire
graph.
It's
the
entire
graph
in
a
specific
order,
because
you
need
to
be
able
to
validate
in
order.
So
what
you
actually
should
be
able
to
do
is
similar
to
like
what
graphs
Inc
does
on
the
client
side
like
when
you
start
import.
You
want
that
header
to
say
what
the
root
note
is
and
then
get
the
root
node
next
validate.
That's
the
next
thing
that
you
get
and
then
all
the
rest
of
blocks
should
come
in
order.
E
So
you
actually
do
need
it
first,
if
you're
doing
the
validation
that
it's
deterministic,
as
you
put
it
in
what
you
probably
want
to
do,
if
you,
if
you
can
only
pin
entire
graphs
and
you
want
to
make
sure
that
you're
not
importing
something
you
can't
keep
around,
you
should
probably
do
the
validation
streaming.
You
can
I.
B
E
You
probably
want
to
run
it
and
that's
their
mode
to
be
honest
like
especially
because
you
don't
have
the
ending
infrastructure
to
be
able
to
pin
a
selector
or
a
partial
graph.
You
have
to
have
the
whole
thing
that
you're
importing.
So
if
that's
the
case,
then
you
might
as
well
just
be
really
strict
about
it
like
most
of
the
car
files
that
are
getting
created
or
deterministic.
E
E
The
only
people
that
are
like
creating
car
files,
that
are
you
feeling,
sick
or
like
me,
and
the
words
yeah
I
mean
like
like
I'm
doing
it,
but
this
is
for
like
a
very
specific
need
and,
like
you
can
just
say,
like
we
don't
support
those
right
now,
although
I
do
have
complete
graphs,
I
didn't
complete
graphs
and
just
the
ordering
is
totally
messed
up,
because
I
have
to
concurrently
write.
The
blocks
are
exactly.
D
D
B
E
B
You
yeah
I'm,
not
even
saying
like
Oh,
you
know
this
block
or
anything
I
more
more
wanted
to
like
the
more
I
think
about
it.
The
more
I'm
like
there
is
no
use
case
aside
from
what
you
said
right
now
with
the
basically
when,
if
you
have
a
deterministic
curveball
that
then
you
can
validate
this,
you
go
any
other
use
case
or
a
bunch
of
blogs,
with
header
kind
of
calls
for
I
want
to
see
the
header
last
as
well
as.
F
One
of
my
news
case
might
be
inspecting
account,
file
and
saying
what
does
it
contain?
It
contains
a
graph
with
these
selectors
that
start
from
this
root
these
roots,
and
if
you,
if
you
have
no
way
to
go
to
the
end
and
come
back
again
like
with
zip
file
which
you
can
do,
then
you
have
to
go
through
the
whole
lot.
You
know
skipping
until
you
find
the
last
block
to
get
that
information,
so
you
can't
do
like
a
go
inspect
on
a
go
on
a
car
file.
E
Also
I'll
say
like
the
the
current
spec
is
basically
locks
like
we
can't
touch
it
and
for
the
much
we're
really
like
when
we're
talking
about
changes,
we'll
talk
about
the
next
version
and
then
the
next
version.
What
we
really
want
to
do
is
have
like
there
not
be
a
header
block
like
there
is
now
and
have
the
header
block,
be
like
an
actual
big
super
block
with
the
CID.
E
So
you
can
really
quickly
read
this
the
idea
and
for
the
header
to
contain
all
the
selector
so
to
contain
the
routes
end
any
selector
to
the
graphs.
That
will
be
incredibly
useful,
because
then
you
can
really
usually
look
at
the
front
of
a
car
file
really
quickly
parse
out
not
CID
and
go
like
Oh
am
I
seen
this
car
file
before
like,
like
that
kind
of
thing.
E
So
now,
like
the
the
car
file
itself,
like
what
we
used
to
call
like
the
route
content
header
is
effectively
the
the
cid
of
the
car
file
right
and
it
like.
It
means
something
a
lot
more
and
in
in
those
use
cases.
You
definitely
want
it
to
be
upfront
right.
It's
just
that
in
the
state
that
the
header
is
and
now
or
it
doesn't
provide
you
very
much
like
yeah.
E
B
F
E
A
F
E
Yeah,
it
needs
to
also
say
if
it's
deterministic
or
not,
because
it
earnest
ik,
then
it
should
always
like
the
whole
car
file
should
be
white
matching
some
other
one
for
that
same
CID,
but
also,
but
it's
like.
If
you
doesn't
say
this
is
deterministic,
then
it
could
be
all
the
same
routes
and
reversals
and
be
like
actually
the
entire.
B
I
guess
one
more
note
about
what
you
raised:
Michael
that
it's
variable
length
and
it's
difficult
to
do.
This
kind
of
you
know
sitting
backwards
for
this
kind
of
stuff.
Normally,
what
some
video
formats
do?
They
basically
define
like
the
last
four
bytes
are
the
length
to
go
bad
and
in
64
that
is
defined
exactly
how
far
back
to
go.
So,
basically,
there
are
three
six
you
seek
once
you,
basically
once
how
long
the
thing
is
you
read
is
four
bytes
and
you
do
another
offset
which
already.
E
F
B
F
Let's
do
that,
let's
take
this
to
github,
because
there's
lots
of
good
thoughts
and
it'd
be
good
to
iron
out
all
of
the
as
many
of
these
things
that
bother
us
about
car
in
the
next
version
and
do
it
ASAP,
so
that
we
don't
have
too
many
car
v1
sitting
around
that
we
have
to
care
about
that
much.
Although
Michael
yours
are
probably
going
to
be
the
big,
the
big
ones.
A
Stored
forever,
hopefully,
okay,
yeah
I,
guess
yeah.
We
passed
the
meeting
so
I
also
think
we
don't
have
anything
else
to
discuss
and
I
forgot
to
say
that
this
week
I
have
limited
availability
because
I
go
to
conference
until
end
of
the
week,
but
yeah
yeah
cool.
So
yes,
you
all
next
week
same
time
same
place.