►
From YouTube: IPLD Overview & 2018 Update - Steven Allen
Description
Originally recorded during the Berlin Developers Meetings from July 9-13, 2018.
A
A
A
So
originally
we
had
something
called
the
Merkel
dag
the
Merkel
dag
had
basically
data
and
links.
Every
node
looks
kind
of
like
this,
where
you
could
have
one
node,
that's
protocol
labs.
It
has
a
data
section.
It
just
has
raw
data
here,
I
form
it
in
Jason
to
be
nice
to
you
in
ipfs.
We
use
part
of
us,
so
this
will
just
look
at
garbage
and
then
said
you
can
have
another
node
Steven.
A
That's
me
with
the
data
section
and
then
a
link
pointing
back
to
the
company
I
worked
for
and
showing
the
relationship
between
me
in
the
company.
This
form
it
was
a
bit
annoying
to
work
with,
so
we
made
up
UD
I
feel
D
is
structured.
So,
instead
of
just
having
this
arbitrary
data
section,
the
entire
thing
is
structured
and
you
can
just
like.
Even
you
don't
know
what
you're
looking
for.
You
can
understand
the
structure
its
self
describing.
So
you
don't
have
external
schemas
here,
like
this
disk
image
actually
built
in
in
ipfs.
A
It's
not
it's
just
binary,
but
in
IP
LD.
It's
actually
fully
just
like
skin
meds
is
what
ideally
would
look
like
go
back
to
the
second,
it's
versioned,
so
you
know.
Ideally,
we
have
this
thing
called
C
IDs
or
content
ideas.
A
Basically,
whenever
you
have
a
pointer
that
points
to
you,
some
piece
of
IPL
D
data,
the
pointer
says
what
version
of
IPL
D
it
is,
and
the
type
of
data
it's
pointing
to
or
the
calcul
interpret
the
data,
so
I
could
say:
oh
you're,
pointing
to
a
get
object,
I'll
get
to
that
in
a
second
but
like,
and
then
that's
like.
This
is
how
you
decode
the
object,
and
this
is
the
version
of
IPL
they
were
using.
A
So
it
means
like
if
we
have
to
change
anything,
we
can
always
just
up
the
version
number
and
happy
oculi
in
the
future.
This
is
hard
to
do
with
to
deal
with
this
with
our
existing
merkel
deck
we
actually
had
to
make
like
we
call
b
c
d
b0.
Would
you
hope
you
never
have
to
you
again?
We're
basically
said
like
well.
If
you
see
something
looks
exactly
like
this:
if
the
CA
d
b0
they're
portrayed
it
this
way,
otherwise
there
has
a
version.
You
can
treat
it
the
normal
way.
A
We
don't
have
to
do
that
again,
because
it's
a
lot
special
casing.
Next,
the
pocket
it's
inclusive,
so
IPL
de
is
really
cool,
because
we
can
point
to
anything
like
any
merkel
link
data
and
basically
say
look
like
this
is
now
part
of
IPL
d.
So
I
think
this
is
the
Borg
or
basically
say
well
get
like
it's
a
Merkel
linked
data
structure.
A
Yeah,
it's
basically
it's
a
list
of
commits
about
the
previous
commit
and
then
each
commit
points
to
tree
objects,
yeah
and
yeah,
but
like
that
means
like
from
IPL
D
itself,
you
can
actually
like
explore
a
get
commit
tree
and,
like
reference,
any
object,
you
get
tree
and
just
treat
it
as
if
it's
any
other
ideally
object.
We
can
do
the
same
thing
with
hiriam.
We
also
have
our
own
custom
formats.
The
uses
C
bore
yeah.
A
Finally,
it
has
past
reversal,
it's
actually
go
to
400
explain
it
first
basically
allows
you
to
name
anything
in
my
building.
That's
the
main
point
in
the
merkel
dag
you
can
kind
of
name
things,
but
the
real
way
you
need
things
like
you
said
like
like
some
node
and
then
link
number
32
and
then
somewhat
like
link
number
41
or
whatever
you
sort
of
like
walk
through
the
link
numbers
and
that's
not
very
friendly,
but
I'll
just
show
you
the
example.
It's
easy
to
understand.
A
So
this
is
again
the
this
example
here,
but
rewritten
in
I
feel
D
where
it's
just
two
objects:
the
company
and
then
me
and
then
my
object
is
a
pointer
of
active
company,
but
the
links
actually
like
I
feel
the
resolve.
So
I
should
I
feel
de
my
Content
ID
/
company,
/
name,
which
just
walks
through
my
object
to
such
company
and
then
walks
through
the
company
object
through
slots
name
and
it
resolves
to
protocol
labs.
It
just
varies
afore,
exploring
data
for
the
team.
That's
working
on
this.
A
The
working
group
is
one
myself
dummy,
Dodd
who's,
not
here
currently
and
Volker.
The
midwives
I
call
the
midwives.
This
is
because,
like
the
people
who
helped
her
birth,
fu
Lee
again
daveed
kwon
nickel,
out
myself,
Jeremy
and
then
we
have
a
bunch
of
contributors.
It's
bit
hard
to
list
them,
because
I
feel
these
very
sort
of
spread
out
and
has
been
built
over
with
many
discussions,
some
of
them
status.
A
Currently
we
have
I
peeled.
The
ffs
runs
an
ideally
and
we
have
C
or
D
T's
and
I
feel
D.
We
are
working
on
new
improved
formats.
Actually
get
is
probably
you
should
go
back
up
to
deployment
now
that
the
next
release
of
ipfs
well,
once
the
next
release-
5f.
Yes,
it
comes
out
yet
we
support
it
by
default
and
go
FS
and
yeah
we're
still
working
on
making
a
little
more
user-friendly.
That
can
actually
also
moved
a
bit
more
towards
deployment,
because
Jas
or
JavaScript
has
been
working
a
lot.
A
Research
I
don't
have
to
get
into
this
too
much,
but
basically
we
were
working
on
introducing
types
to
you
ipfs.
So
currently
we
have
like
on
its
structured
data,
but
it's
not
like
we're
losing
type
information.
It's
like
when
you
break
your
program.
You
may
have
type
information
you
program
and
then
like
when
you,
when
you
sorry
data
sources
and
IPL
de
we
kind
of
just
sort
of
like
we
keep
the
structure,
but
we
throw
away
the
actual,
like
extra
type
information,
we're
hoping
to
keep
that
in
the
future.
A
Javascript
just
would
happen
this
quarter
or
the
past
few
quarters
interfaces
and
formats
and
documentation
is
steadily
improving.
Thanks
to
Volker
go
the
next
go
up,
yes,
released,
look
at
and
include
the
get
IV
plugin,
improve,
support,
decoding
coding
and
a
but
others
mostly
stagnant
due
to
lack
of
time.
So
that's.
A
So
the
question
is:
have
you
considered
using
graph
QL
for
trying
out
BLD?
We
have
we've
looked
at
it.
We
are
still
considering
it
one
issue
there
is.
We
would
like
to
make
a
query
length
with
this
person.
I
would
like
to
make
query
language.
It
has
a
nice
performance
guarantees
where
I
can
send
a
query
to
some
arbitrary
new.
A
That
doesn't
trust
me
and
it
can
execute
the
query
and
know
like
you
won't
like
take
exponential
time
like
that
and
also
know
a
few
other
properties
really
ideally
you'd,
be
able
to
partially
execute
the
query
over
some
subset
of
data
and
then
return
a
certifiable
results.
Look
query:
we're
like
the
the
person,
that's
getting
the
data
back
and
then
verify
that
the
query
succeed
correctly.
I,
don't
know
if
graph
to
entre
provides
these
kinds
of
guarantees.
A
So
we've
been
looking
to
other
selected
languages
that
allow
things
like
that
we
could
implement,
or
we
say
we
could
still
use
graph
2
on
top
of
some
other
system.
We're
like
whiskey,
I,
give
you
a
graphical
query
and
you
give
me
back
everything
I
need
to
actually
flick
answer
the
query
early
to
determine
if
you
did
perform
the
query
correctly.
But
that's
sorry.
It's.