►
From YouTube: Weekly Sync 2022-01-26
Description
Meeting Minutes: https://docs.google.com/document/d/1vKYEPtqKiwsFwhVKPmPub5ebMqN9HteBcbdFAuTXalM/edit#heading=h.yqya2ol73x66
YouTube permanently deleted the audio on part of this :(
A
Recording
now
all
right,
this
is
our
weekly
sync
and
apparently
this
is
partly
the
open
broadcaster
weekly
sync
today,
because
we're
figuring
that
out.
So
let's
here's
the
terminal-
and
here
is
the
oh.
How
am
I
gonna
do
this?
Okay,
I
just
realized
all
my
windows.
All
my
with
my
stuff
is
attached
to
the
one:
that's
recording
our
faces.
A
Okay,
there
we
go
so
leave
the
meet
window
over
here,
move
everything
else
there
section
and
there's
the
chat
okay.
So
this
is
the
one
that
has
no
audio
right.
I
think
this
is
actually
I
tried
to
do
this
and
I
think
we
struck
out
yeah.
We
struck
out
no
audio
there,
all
right,
so
I'm
just
gonna
put
you
know
that
there's
no
audio,
just
just
in
case.
I
don't
think
there
was
anything
really
relevant
there,
but
you
know
maybe
we
can
voice
over
part
of
it
later.
I
actually.
A
A
B
B
So
psychic
stuff
is
fixed
great,
but
one
of
the
cycle
tests
fails
randomly
so
I
have
written
about
it
in
the
relevant
issue
like
why
it
is
failing.
So
it
is
probably
we
are
not
using
the
right
metric
and
it
sometimes
gives
us
a
negative
accuracy.
So
that
is
an
issue,
but
you
know
that
is
completely
random,
so
it
happens
some
of
the
time
it
doesn't
happen.
A
A
C
B
B
A
A
A
B
A
B
B
Get
a
mac
one
so
because
we
have
mostly
windows
people,
so
I
will
first
make
a
windows
video.
I
have
a
windows,
vm
setup
and
my
machine
right
now,
so
I'll
do
videos
for
that.
That
will
help
new
contributors
coming
in
and
also
save
us
much
more
time
than
we
would
give
people
setting
up
their
stuff
and.
C
A
A
A
C
B
A
B
B
A
Have
a
lot
of
intermittent
cv,
number
detection
failures-
I
I've
tried
every
time
it
fails.
I
try
to
go
through
and
update
it
so
that
it's
like
a
greater
than
range
rather
than
a
direct
comparison,
because
there's
a
bunch
of
them
in
there.
But
that's
you
know
yeah
that
that
that
that
we'll
figure
it
out
so
should
I
test
so
all
right
and
then
so
hashem
how
how's
it
going
with
you
and
and
what
are
you?
What
are
you
looking
at.
A
C
A
A
Yeah
this
is
great.
This
looks
great.
I
was
about
to
merge
it.
The
one
thing
is
we
should
probably
we
should
probably
okay.
Now
I
just
have
two
tabs
all
right,
so
the
one
thing
is,
I
noticed
that
this
repeats
in
each
in
each
like
this.
This
section
repeats
in
all
three.
You
know
all
three
times
that
you
do
this
and
I'm
wondering.
A
Should
this
be
something
that's
like
a
method
on
a
model
I
think
it
might.
It
might
be
better
suited
as
a
method
on
a
model.
So
I
think
we
could
open
another
issue
and
do
that
later
just
say:
hey
refactor
this,
because
then
the
model
could
programmatically
declare.
You
know
if
it
has
a
different
predict
feature.
You
know
if
it's
called
something
else
in
the
config,
it
could
return
the
right
thing.
A
The
other
thing
was:
what
was
it?
Okay,
I
think
you
got
it
yeah
you
got
it.
You
had
the
test
yeah.
I
think
that
was
it.
I
think
that
was
the
only
thing
so
we'll
just
file
an
issue
on
this
so
and
then
we'll
merge
this.
Does
that
sound
good?
Is
there
anything
open?
Did
I
miss
anything,
it
looks
good
to
me
as
far
as
I
can.
B
So
while
hashim
is
here,
I
wanted
to
talk
about
that
test.
The
basic
test
which
we
run
like
the
base
test
we
have
in
python,
one
that
does
in
three
point
eight.
It
is
failing
due
to
some
I
python
error,
and
that
is
one
of
the
notebooks
I
I
am
not
sure
what
it
is
why
it
is,
but
you
know
if
you
know
someone
who
has
taken
a
look
at.
A
A
B
A
What
would
we
call
this?
You
know
prediction
feature
prediction
or
feature
list.
Yeah
prediction
feature
list,
okay,
so
and
then
we're
gonna.
Also,
we
also
grab
features.
Okay
features
needs
to
stay.
Consistent
predict
really
needs
to
stay
consistent
like
generally,
but
there
are
reasons
why
you
might
not
do
it
predict.
So
does
this
really
need
to
be
refactored
we'll?
Just
let's
see.
A
A
A
A
A
A
It's
already
in
there.
Okay,
great,
I
think
I,
when
I
looked
at
the
diff,
I
wasn't
sure
if
it
was
in
there
or
not,
I
must
have
missed
that
okay,
so
this
should
be
an
over
rideable
method
and
model
base
class
model
context
or
no
model
yeah.
This
is
a
model,
okay,
yeah
so
and
then
we
we
should
call
it
from
the
locations.
C
A
Yeah,
so
let's
take
a
look
at
that
real,
quick
because
yeah
there
was,
there
was
a
bunch
of
stuff
with
the
accuracy
scores
and
that
finally,
I
think
I
tried
to
rebase
this
at
one
point
and
it's
relatively
smooth,
shockingly
shockingly,
okay,
back
when
the
ci
runs
past
okay,
never
mind,
I
take
all
that
back
all
right.
Look
at
that
not
smooth
was
yeah.
A
A
A
Okay,
so
let's
make
sure
that
all
this
stuff
looks
concise,
you
know,
so
I
wish
they
had
just
top
level
of
weight
in
that,
don't
they
have?
No,
maybe
they
it,
then
I'm
not
sure.
Okay,
we
need
to
do.
This
is
another
thing
I
I
realized
this
recently
while
we're
at
it.
We
need
to
make
sure
that
the
sources
arguments
gets
into
the
beta
release,
because
we
can't
we.
A
A
A
A
A
A
B
B
B
A
A
Yeah
yeah,
so
this
looks
interesting
sort
of
like
little
web
uis
directly
for
your
notebook.
I
think
that
you
know
there's
something
there's
something
to
be
done
here
right.
I
don't
know
what
it
is,
but
all
right
because
thinking
about
you
know
so,
here's
here's
my
here's.
What
I'm
thinking
right
is
that
there's
something
to
be
done
with
okay,
so
so
number
one.
The
notebooks
have
a
bit
of
a
it's
difficult
to
review
them
right
and
it's
difficult
to
manage
diffs
to
them
because
of
their
json
nature
right.
A
The
restructured
texts
are
easier
for
that.
But
they're
not
you
know
you
can't
you
can't
it's
not
a
notebook.
So
that's
not
ideal.
So
there's
some
happy
medium
there
or
maybe
some
translation
that
needs
to
happen.
I
think
so
because
I
know
that
there's
the
review,
nb
and
other
tools,
but
it
doesn't
help
us
when
it's
in
the
kit
repo
right.
So
it
would
be
nice
if
we
could
figure
something
out
there.
Just
this
is
like
you
know,
long
long
term.
You
know
maybe
short
term.
Ideally
this.
A
A
A
So
if
you,
if
we
could
define
the
notebooks
like
if
we
could
use
this
mercury
thing
in
conjunction
with
notebooks
written
as
rst,
then
we
could
do
we.
We
could
essentially
have
documentation
that
can
be
executed
within
data
flows,
so
you
can
chain
documentation
pages
together
right.
So
if
you
do
this
demo
right
and
then
you
do
another
demo,
then
you
can
chain
them
together.
So
if
you
do
a
demo
of
like
some
object
detection,
remember
we
were
going
through
like
that.
One
video
recognition
thing.
A
I
can't
remember
what
it
was,
but
it
required
that
you
run
it
through
several
models
beforehand
right.
So
if
we
were
to
write
a
note
like
a
a
tutorial
for
each
model
like
in
how
you
pre-process
that
data
set,
then
you
could
very
easily
end
up
with
a
very
complex.
You
know
model
chain
by
just
chaining,
the
tutorials
that
already
exist
or
writing
a
new
tutorial.
A
A
B
A
A
A
A
It's
a
stop
gap,
it's
a
good
stop
gap,
but
it's
not
my
favorite
that
we
have
to
apt-get,
install
some
compiled
dependency
right
and
that
it's
not
something
that
you
can
just
pip
install
just
because
it
adds
one
more
setup
step
for
development
right.
So,
ideally
those
where
are
those
people
that
we're
doing
this?
You
guys
know
that
strength,
so
sphinx,
markdown,
sphinx
markdown.
This.
The
one
remark.
A
I
think
it
is
I'm
wondering
if
you
can.
Okay,
I'm
wondering
if
you
can
compile
it.
This
is
gonna
sound
ridiculous,
I'm
wondering
if
you
can
compile
it
to
webassembly
and
then
execute
it
from
webassembly,
because
then
you
could
have
it
natively
execute
via
the
wasm
interface
but
anyways.
This
is
problems
for
another
time,
we'll
get
it
working
first,
it's
just.
It
would
be
nice
to
be
able
to
pip
install
everything.
Obviously
this
is
getting
a
little
bit
complicated.
B
A
B
A
A
Oh,
maybe
that's
what
it
was
yeah.
I
wanted
you
to
guy.
I
I
knew
there
was
something
I,
but
if
you
guys
could
turn
on
two
factor
off,
because
if
I
make
you,
if
I
make
you
maintainers,
then
you
know
if
somebody
takes
over
your
account,
there's
bad
things.
Bad
things
can
happen
right
so
or
owners
whatever
it
was,
but
I
noticed
you
guys
didn't
have
two-factor
authon,
so
I
was
going
to
tell
you
first
and
then
I
forgot.
B
B
A
B
B
B
A
A
A
A
A
A
Okay,
so
the
other
thing
off
this,
so
I
think
we've
decided
yes,
okay,
we've
updated.
I
want
to
make
sure
we
have
reasoning
for
this,
so
this
needs
to
be
in
beta,
great
okay.
So
the
other
thing
on
this
is
I've
been
trying
to
think
about.
You
know
what
makes
a
good
issue
right.
How
do
you
make
sure
that
you
have
a
complete
description
of
the
issue
and
I've
sort
of
I'm
playing
around?
A
This
is
a
longer
discussion,
but
I
just
want
to
throw
this
out
there
for
you
guys,
I'm
documenting
I'm
throwing
random
notes
related
to
data
flows
and
and
just
sort
of
general
things
into
this
issue.
Here.
You'll
see
I'm
updating
this
frequently,
and
so
because
there's
not
a
lot
of
docs
on
data
flows,
and
it's
not
you
know
it
would.
That
needs
to
be
documented
right
and
part
of
that
is
right.
We
know
that
we
don't
have
a
solid
interface
right.
A
No,
it's
not
very
ergonomic
right,
not
not
not
the
most
clear,
so
I've
been
working
on.
I
know
that
you
had
saw
hill.
You
were
going
to
do
like
a
pie,
tour
style
thing
at
some
point.
That
would
be
really
cool.
You
know
whenever,
whenever
I
know,
you're
fixing
ci
failures
is
definitely
the
first
thing
we
need
to
do,
but
I
just
wanted
to
say
that
for
the
recording
you
know
we're
going
to
look
at
that
and
then
now
you
are
still
planning
on
looking
at
that
right.
A
A
Okay,
so
I'll
put
this
in
bold,
for
we
need
to
talk
about
this
all
right.
So,
okay,
we
called
this
one's
done
all
right,
so
yeah.
So
basically,
this
stuff
is,
you
know,
a
bunch
of
ramblings
and
random
notes
that
I'm
going
to
put
together
into
more
stuff,
and
then
there
was
also
like
I'm
playing
around
with
something
that's
sort
of
like
a
a
just,
a
plain
python
implementation.
A
That's
like
a
very
simple
implementation
of
like
what
a
data
flow
is
sort
of
just
it's
basically
like
a
loop
with
some
events
and
and
like
a
dictionary
of
hey
this
event
maps.
This
thing
do
this
thing
right,
trying
to
make
it
like
a
little
more
close
to
like
peak
under
the
hood.
You
know
so
that
people
understand
what
what
what
is
actually
happening
right
when
they,
when
they,
when
they,
you
know
create
this
data
flow
description
right
they
can.
A
They
can
sort
of
you
have
to
peek
under
the
hood
and
hopefully
that
helps
people
understand.
What's
going
on
better,
so
kind
of
take
this
transition,
the
the
idea
for
these
tutorials
right
now
is
sort
of
you
know
you
do
your
non-data
flow
approach
and
then
you
do
your.
You
know
basic
this,
this
sort
of
peak
under
the
hood
style
data
flow
approach
right
and
then
you
do
your.
A
You
know
flow
definition
in
the
you
know
the
way
that
sahil
you're
going
to
propose
ideally
and
then
maybe
we
have
another
one
where
it's
like.
Okay,
now,
here's
the
raw
way
of
of
writing
a
data
flow.
That's
you
know
not
not
very
nice
and
pretty,
but
if
you
want
to
get
down
to
the
nitty
gritty
you
know
here,
you
go
here's
the
here's,
the
data
structure
right
so
kind
of
that
that's
kind
of
like
the
arc
of
the
the
story
of
dataflow
and
and
and
how
we
do
that
tutorial.
A
A
All
right
so
now,
governance
and
jsoc
how
docs
executed
we
covered
the
okay.
So
let's
do
how
our
docs
executed.
So
if
you
wanted
to
execute
a
single
doc
test,
you
could
go
to
contributing
and
not
notebooks
documentation
because
well,
if
you
wanted
to
execute
a
notebook,
then
you'd
go
execute
a
notebook
like
that,
but
this
is
testdoctors
so,
for
example,
good
man,
I
could
have
sworn
that
we
had
what
to
sphinx
extension
now
I
could
have
sworn
that
this.
Oh
maybe
it's
under
testing.
C
A
Here,
yeah,
okay,
see
that's
what
it
was.
Okay
see.
This
is
not.
This
should
also
be
under
the
documentation,
stuff.
Okay,
this
is
the
how
to
run
a
single
rst
test
and
it
doesn't
run
them
unless
you
have
this
environment
variable
set,
because
they
that
way,
if
you
run
all
of
the
unit
tests
it,
you
know
that
running
all
of
the
unit
tests
is
a
you
know.
It
takes
a
long
time.
A
A
A
A
A
You
guys
talked
to
them
great,
so
yeah,
you
guys
should
talk
yeah
and
hashem
you.
You
should
talk
to
them
too.
If
you
haven't
talked
to
them
already,
so
we
are
looking
at
I.
So
I
talked
to
my
legal
department
and
you
know
we
talked
to
yash
and
saksham
and
I
think
there's
some
other
people
involved.
You
know
I'm
sure
they
want
you
guys
to
get
involved
there,
and
you
know
they're
trying
to
build
out
this
org
and
this
set
of
projects
under
this
org
github.com.
A
Bill
tree-
and
you
know
we
had-
we
had
talked
about
moving
things
to
dfml
org.
You
know
we
might
might
do
it's
looking
like,
as
far
as
you
know,
transfer
of
ownership
of
the
repo
legal
is
looking
for
an
organization
with
a
defined
governance
structure.
So
I
I
met
with
legal
and
they
told
me
here's
what
a
governance
structure
looks
like
and
then
I
met
with
yash
and
saksham,
and
I
said
you
know
hey.
If
you
guys
make
a
governance
structure,
then
we
can
transfer
the
repo
to
you
right.
A
They
could
then
transfer
to
repo
to
the
dfml
org
itself,
but
it
you
know
it
would
be
like
a
build
tree.
All
this
stuff
would
be.
You
know
you
all
would
be.
You
know
the
the
main
maintainers
on
it
right.
You
know
I
would
be
involved.
Obviously
too
extend
call
you're
gonna.
Make
me
pay
for
that.
Aren't
you
google,
okay,
so
I
would
be.
You
know,
obviously
involved
too
still,
I'm
not
just
gonna,
like
you
know,
you're.
A
B
A
B
A
B
B
A
I
think
that's
perfect.
We
had
talked
about
that
because
you
know
they
said
you
know
they
want
to
be
a
space
to
create
projects
and
then
match
people
up
with
other
people
who
have
the
rest
of
the
skills
needed,
and
then
you
can
all
go,
build
a
project
right.
So
it
sounds
like
that's
exactly
what
you're
talking
about
so
we'll
communicate
to
them.
You
know
looking
for
formal
project
proposal.
A
Template
you
know
and
we'll
mention
sahil's
ideas.
Are
you
know
the
meeting
recorder
and
all
of
their
project?
They're
they're,
you
know
so
they're
thinking
about
you,
know,
governance
and
then
the
projects
themselves.
So
they
would.
You
know
anybody
who's
a
within
you
know
the
top-level
governors
of
that
org
would
go,
and
then
you
know
help
where
needed
in
the
different
projects.
Right
and
then
people
could
obviously
work
across
projects.
But
you
know
they're.
The
goal
would
be
to
try
to.
A
A
Okay,
great,
I
saw
something
recently,
oh
I'll,
try
to
remember.
Oh,
oh
god,
you
know
this
would
be
so
sweet,
okay,
yeah,
some
kind
of
python
interface
security.
I've
talked
about
that
with
people
before
that
would
be.
Oh
man,
you
could
really.
You
could
really
emulate
some
cpus
or
simulate
some
cpus.
It
would
be
great
so
setting
up
governance
so
and
then
so
I
so
we
have
logo
approval.
I
think
most
people
have
voted
on
the
poll,
we'll
probably
right
now.
A
Jason
long
is,
is
winning
right
now,
yeah,
so
we'll
we'll
see,
you
know
we'll
see
what
what
ends
up
there
on
the
30th
and
then
you
know
we'll
communicate
to
whoever
the
person
was
that
you
know
designed
the
logo
that
that
got
the
most
votes
and
will
ask
for
svg's
right
and
then
yeah
and
then
we'll.
You
know
we'll
use
that
logo.
A
B
A
A
You
know
likely
moving
into
bill
tree
would
end
up
polluting,
build
tree
with
a
million
plug-ins.
There's
also
the
thing
that
you
know
you
don't
really
need
to
split
it
out
into
the
poly
repo
like
sorting
it
out
into
the
poly
repo
structure
would
be
good.
It's
also
how
bill
tree
is
organized
right.
Oh
wow,
they
changed
the
setting
page.
Why
am
I
on
the
settings
page?
Oh,
I
don't
need
to
be
but
yeah,
but
I
think
we.
A
Ci
would
be
much
faster,
exactly
we'll
only
test
what
you
need
to
test.
I
mean
and
plus
not
a
lot
of
people
do
that
poly,
repo
validation
like
in
open
source
projects,
and
so
it
would
be
a
good
example
like
we
would
be
a
good
example
of
an
open
source
project.
Doing
poly
like
a
poly
repo,
ci
setup
right,
that's
very
unique
and
would
be
really
cool
if
we
could
figure
that
out
now-
and
I
think
part
of
all
this
stuff
with
the
mana
all
this
stuff
with
these
docs
and
the
manifest.
A
C
A
A
You
don't
understand:
okay,
yeah
there's
I
mean
I
don't
I
I
mean
I
don't
understand
either
right
like
I
just
know
that
there's
an
opportunity
to
move
it
to
a
place.
You
know
that's
owned
by
students
have
been
who
have
been
a
part
of
this
project
and
who
would
like
you
know
other
students
who
have
been
a
part
of
this
project
and
other
projects
to
come,
govern
a
bunch
of
repos
together.
So
we
can
talk
about
it
more,
but
that
sounds
you
know
great
to
me.
A
A
A
B
A
So
should
I
is
a
is
a
metastatic
analysis
tool.
So
essentially
you
know
a
static
analysis
tool.
Analyzes
some.
You
know
some
something
right
so,
and
these
are
all
most
of
these
are
about
finding
vulnerabilities.
So,
basically,
should
I
be
in
a
metastatic
analysis
tool.
You
point
it
at
a
code
base
and
it
figures
out
what
static
analysis
tools
it
needs
to
run
and
then
you
know
reports
an
aggregate
of
the
results
right.
A
So
if
I
pointed
at
a
code
base
that
had
both
there's
one
in
there,
that
has
both
javascript
dependencies
and
rust
dependencies.
So
your
point:
if
you
point,
should
I
add
the
code
base,
it
will
know
that
it
should
run
both
rust
dependency,
analyzers
and
javascript
dependency
analyzers,
and
it
will
report
back
both
results.
So.
A
A
A
A
Boom
my
faces
are
back
on
the
side:
okay,
okay,
so
this
the
reason
why
I'm
so
skeptical
of
this
is
because
this
javascript
algorithms
is
pinned.
This
is
we're
analyzing
this
repo
and
this
repo
is
pinned
to
this
sha
commit
that's
really
old.
So
if
we
look
at
this
binaries,
so
these
are
all
the
binary.
So
we
did
this
for
caching,
we
put
them
all
on
the
same
thing.
So
if
any
of
these
change,
then
we
reload
the
cache,
because
these
are
expensive
to
download
time
wise.
So
where
is
this
javascript
algorithms?
B
A
A
A
Then
this
is
a
lie,
so
I
don't
know,
but
one
is
not
great
yet
there's
no
way
that
that,
like
almost
three
thousand
vulnerabilities
just
disappeared,
so
we'll
just
make
it
greater
than
one,
but,
let's
just
say,
I'm
thoroughly,
skeptical
of
that.
Okay,
so
good
status
get
pulled,
I'm
probably
on
my
manifest
branch,
all
right.
Okay,.
A
B
A
A
A
A
So
you'll
notice
that
if
you
do
this,
did
we
go,
did
I
get
rid
of
that?
Okay?
Well,
I
guess
I
got
rid
of
the
check.
Maybe
oh
maybe
the
dependency
got
updated.
So
you
just
use
javascript
okay.
So
this
is
the
cli.
So
basically
you
run
the
cli
and
then
it
will.
You
know
appropriately
determine
whether
it
should
use
python
or
use
javascript
or
use
rust,
analyzers,
okay,
so,
okay
commit
sam.
A
B
A
B
A
A
A
A
A
B
A
Hopefully,
they'll
have
more
to
to
tell
us,
and
you
guys
can
talk.
I
would
encourage
you
both
to
talk
to
them
some
more
and
help
them
help
them
write
that
stuff
right,
because
that's
going
to
be
a
big
group
effort.
So
so
should
I
test
my
python
okay,
so
this
was
other.
This
is
okay,
that's
the
issue.
A
B
B
A
B
A
Changes
across
python
versions-
hive-
I
is
your
best
friend
or
a
doctor
container,
but
you
know
those
aren't
always
the
most
fun
to
use.
Okay,
so
you
know
we
might
want
to
document.
So
I
was
also
looking
at
windows
instruction
installation
instructions.
Oh
I
I've.
Since
I've
been
on
windows
now,
all
of
a
sudden,
I'm
fixing
windows
bugs
so
yeah.
So
so
this
one
I'm
yeah.
So
we
might
want
to
put
this
in
the
installation.
Instructions
might
want
to
mention
and
they
install
docs.
A
B
B
A
A
B
A
B
A
C
A
A
What
about
davis
yeah?
This
would
be
sweet.
Okay,
let's
just
like
log
this
as
some
of
these
elixir
people,
these
people-
if
you
guys,
haven't
seen
this
elixir
stuff,
they
don't
they
they're
doing
some
cool
stuff.
Let's
see
it's
very
ruby
on
railsy,
if
you're
you're
a
python
developer,
not
not
doesn't
super
jive
with
you
probably,
but
it
is
very
cool
they're
doing
some
great
stuff.
A
A
A
A
A
Now
that
I
think
about
it,
but
you
know
that's,
so
we
definitely
want
to
do
that
either
way.
But
if
that
exists,
we
should
try
to
look
into
that.
So
if
anybody
sees
anything
so
the
reason
why
auto
sk
learn
is
not
is
is
having
problems
is
because
they
don't
declare
their
build
dependencies
per
the
new.
Well,
it's
not
really
new,
but
it's
pep
pep.
A
A
Yes,
they
don't
declare
them
in
a
pep
17
compliant
format,
if
I
remember
correctly,
and
so
basically,
what
this
is
is
it
says,
give
me
a
pi
project,
dot,
tomml
or
pi
project
is
a
project
hotel.
I
think
so,
and
so
and
then
there's
also
five.
Is
this
five
eighteen,
this
one
is
not
accepted
or
no
is
it
five
eight,
so
517
is
draft
or
no
they're,
both
both
final
now
great.
A
So
basically
they
this
this
by
doing
this
new
format-
and
this
is
all
the
setup
config
stuff
that
we
went
through
that
whole
debacle
to
to
migrate
everything.
So
by
doing
this
new
format,
it
means
that
you
can
reliably
install
things
and
like
it
all
works.
Basically,
you
know
because
the
python
ecosystem
has
just
been
a
bit
of
a
hell
for
packaging
right
and
so
say.
For
example,
you
have
a
compiled
portion
of
your
project
like
auto
sklearn.
Does
then
you
can.
A
You
know,
declare
the
compiler
that
you
need
within
the
pi
project.toml,
and
in
doing
that
you
know
your
compiler
gets
installed
first
and
then,
when
your
package
gets
installed.
Second,
it
actually
has
a
compiler
to
run
with
now.
If,
if
the
the
auto
sklearn
project
last
time,
I
saw
had
not
made
that
switch
yet,
which
means
that
their
compiler
and
their
package
get
installed
at
the
same
time
and
therefore
their
package
doesn't
see
the
compiler,
because
it's
not
installed
yet
because
it's
getting
installed
by
pip
at
the
same
time.
A
So
then
you
get
that
compilation
error,
which
means
you
have
to
reinstall.
I
believe
that's
what's
happening
at
least
that's
what
what
has
been
happening
before.
So,
if
you
look
at
the
blank,
so
basically
you
know
you
guys
know
we
couldn't
create
plugins.
I
think
somehow
you'd
ask
me
like
what,
if
you
added
a
new
package
right,
so
I
added
one
that
was
just
creating
a
blank
package
right.
So
this
is.
This
is
just.
A
So
this
is
just
like
what
does
what
this
is.
What
a
blank
package
looks
like
right,
so
this
is
just
sort
of
like
a
skeleton
python
project
with
packaging
right.
So
if
you
need
to
create
a
python
project,
you
know
this
is
there's
many
utilities
out
there
to
do
it,
but
here's
yet
another
one,
and
it
has
you,
know
your
basic
setup
py
with
this
you
this
there's
a
current
bug
that
it
has
a
workout
around
for
and
then
it
has
your
setup,
config
right
and
yeah.
A
A
This
is
an
example
of
a
project
that
has
release
automation
and
has
testing
across
all
three
platforms
when
it
links
mac
windows
and
has
this
you
know,
entry
point
registration
for
the
console
script.
So
if
you
want
a
console,
you
know
when
you
install
you
get
a
console
program
too,
that
you
can
just
run
and
it
has
this
new
style
package.
So
if
you
need
an
example,
it
builds
on.
I
think
why
is
it?
I
think
the
blank
project
will
give
you
one
that
does
the
setup
tools
version
information.
A
So
basically
you
don't
need
to
declare
a
version
if
you
run
this,
you
know
create
me
a
blank
project
thing,
because
it's
just
going
to
yeah,
you
don't
have
to
declare
a
version
it'll.
Just
basically
grab
from
your
git
repo.
What
the
version
of
your
package
is
so
just
know
that
this
is
exists
as
a
reference,
if
you
need
it
or.
A
A
A
A
A
The
bug
is
coming:
where
is
that?
How
do
we
find
okay?
This
bug
is
because
of
indians.
Let
me
look
at
the
bug
and
see
if
it
jogs
my
memory.
This
is
a
weird
bug.
It
did
and
I
think
it's
caused
by
a
workaround
and
that's
why
it's
working
in
3.7,
but
not
in
3.8,
so
maybe
they
fixed
something
in
3.8
and
now
the
workaround
is
no
longer
required,
which
means
that
the
workaround
is
blowing
up.
I
think
that's
what's
happening
here.
So
yes,.
A
A
A
A
Okay,
this
is
this
is
the
thing
so
sometimes,
okay,
so
ipython
notebook,
okay,
so
ipython
notebook
is
supposed
to.
What
are
we
doing
this
for
oh
within
method?
Oh,
this
is
for
the
config
stuff,
okay,
so
this
is
for
the
immutable,
config
stuff,
so
yeah.
This
is.
This
is
dirty
so
the
immutable
config
stuff,
because
we're
using
the
data
classes,
we
went
back
and
forth
when
we
did
that
on.
A
Do
we
actually
change
away
from
using
data
classes
or
do
we
use
just
continue
to
use
data
classes
right,
so
we
went
with
continue
to
use
data
classes
right
because
then
you
get
fields,
you
get
all
the
standard.
It's
a
standard
part
of
the
python
right
and
try
to
stick
to
standards
wherever
possible.
A
A
A
And
this
stuff,
I
think,
is
how
we're
going
to
fix
a
lot
of
the
config
infrastructure
too.
Now
that
we
have
this
set
up.
So
basically
it
goes
through
all
the
fields,
and
it
says
you
know:
here's
a
you
know:
here's
a
getter
that
wraps
the
field
and
here's
the
setter,
the
raster
field
and
then
so
in
in
the
in
the
setter.
A
Basically,
I
we
go
and
this
setter
will
get
called
since
that's
on
creation
of
the
class.
So
this
is
when
I
create
the
class,
I'm
I'm
assigning
a
property
to
the
class.
So
then,
when
this
class
gets
instantiated,
the
setter
gets
called
when
in
the
like,
in
in
the
init
method
of
the
class
that
that
gets
created
when
you
pass
in
the
initial
set
of
arguments
right,
so
we
need
to
say,
are
we
within
the
init
and
if
we're
within
the
init
function
of
the
class,
then
we
set
the
value.
A
Otherwise
we
check
if
we're
enforcing
immutability
and
if
we're
enforcing
immutability,
obviously
we're
going
to
say
you
know
you,
you
may
not
do
that
and
yeah.
So
that's
how
the
immutable
stuff
works,
and
that
would
also
so
this
is
where
I'm
thinking
we
have
a
potential
avenue
here
to
do
correct
type:
conversions,
because
the
config,
the
infrastructure
around
config
in
the
project
is
obviously
vast,
very
important
part
of
the
project
in
configuring.
A
All
these
nested
objects
right
and
the
type
conversion
that
needs
to
happen
is
tricky
because
you
have
these
nested
objects
and
it's
like
well,
what
type
is
it
and
then
how
do
I
instantiate
that
type
and
then
I
have
to
do
that
all
the
way
down
right.
So
this
is
a
potential
place
where
we
could
go
in
and
start
doing.
Those
type
conversions
is
is
like
in
within
the
setter,
so
that
that's
just
just
for
reference.
Okay,
so
then
getting
back
to
this.
A
Basically,
this
within
method-
you
know,
tells
us
if
we're
within
a
method
by
inspecting
the
stack
frame
now
ipython
has.
If
you
look
at
this
issue,
ipython
patches,
the
inspect
module
because
they're
running-
I
don't
know
we
do
they
have
these
kernels
and
stuff.
I
don't
fully
understand
the
eye
of
python
execution
model,
but
somewhere
they
patch
inspect,
so
that
if
somebody
calls
and
specs
yeah
was
this
ever
fixed,
so
okay
yeah
apply
monkey
patch
to
inspect
fine
source.
A
A
A
Okay,
so
yeah,
so
here,
if
you
load
this
when
when
this,
when
this
is
imported,
this
alt
rat
b,
then
you
know
you,
this
is
a
top
level
piece
of
code
that
happens
so
on
import.
It
changes
the
find
source
function
to
be.
This
find
source
function
right,
so
something
in
ipython
basically
says
hey.
If
I
need
to
import
this
thing,
is
this
the
same
bug
yeah?
Then
you
know
you
we
need
to.
We
need
to
use
our
own
version
of
find
source.
A
Now
I
think,
by
the
way
of
how
we're
running
the
tests
with
testbook
testbook
doesn't
set
up
that
fine
source.
It
doesn't
end
up
importing
this,
whereas
you
know
maybe
like
jupyter.
Notebook
does
right,
so
we
have
to
trigger
that
import
and-
and
you
know,
and
and
do
that
patch
right,
so
it
looks
like
from
what
I'm
seeing
in
the
log
here,
it's
mad,
it's
mad
about
it
because
it
says
it
has
no
attribute
find
source.
A
A
We
could,
but
you
know
I
won't
make
you
guys
if
you
guys,
may
not
be
interested
to
go,
find
out
the
entire
history
of
inspect
fine
source.
So
so
we
will
just
call
this
good
and
fix
the
bug
all
right.
So
the
issue
is
showing
up
on
3.8.
You
said
yes,
okay,
so
they
released
a
new
version
where
okay,
so
the
thing
is,
we're
gonna
run
this
and
it
may
it
may
they
may
have
moved
no,
they
may
have
moved
it.
It's
the
thing,
so
we're
gonna
need
to
go.
A
A
A
Yeah
exactly
it's
a
disaster,
and-
and
we
spent
a
bunch
of
put
a
bunch
of
time
into
that
and
then
created
a
massive
disaster
and
then
found
out
that
that
should
never
be
done
so
so.
Instead,
you
know
we're
going
to
add
a
programmatic
workaround
saying
if
3.8,
then
you
know,
we
know
that
we
need
to
do
this
if
3.7,
we
know
that
we
need
to
do
this.
A
A
A
All
right,
I
guess
it
just
disappears
so
we'll
try
to
rerun
this
and
see
what
happens
so,
maybe
it's
no
longer
required.
I
find
that
you
know.
I
find
that
hard
to
believe,
but
we'll
find
out
what
happens.
So
what
test
was
this
that
fails?
So
this
is
test
notebook,
tens,
mb
touring
models
all
right
and
we
need
python
3.8
to
do
this
and
I'm
sure
I'm
on
3.7.
What
do
we
have
here?
Do
we
have
a
throughput?
I
think
I
have
a
version
of
3.8
laying
around.
Where
is
it.
A
A
A
A
B
A
A
C
A
A
B
A
A
You
know
I
would
encourage
you
guys
to
meet
both
of
you
with
yash
and
saksham
and
understand
what
they're
doing
there
and
get
involved
because
you
know
seems
like
you
know,
there's
a
there's,
a
solid
chance
that
we'll
you
know,
move
that
over
there
and
then
we
can
have
permissions
to
make
everybody
maintainers
right
and-
and
I
assume
you
guys-
you
guys
are
still
here.
So
I
assume
you
have
interest
in
you're,
maintaining
things
so
surprise,
you're
a
maintainer
so
yeah
I
would.
I
would
sync
up
with
them.