►
From YouTube: Node.js Loaders team
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
B
All
right
looks
like
we
are
live
on
youtube.
This
is
the
january
11th
2022
meeting
of
the
node.js
loaders
team.
C
Sure
alex
yeah,
I
wanted
to
talk
about
the
get
5
system
book,
so
I
implemented
at
first
sight
and
then
I
like
to
get
an
idea
of.
What's
the
next
step
from
what
I
understood,
the
the
main
issue
right
now
is
to
explain
how
it's
different
from
the
reserve
book,
coupled
with
a
load
hook.
C
So
basically
the
main
the
main
issue
is
that
all
the
not
resolution
algorithms
are
made
to
work
on
the
file
directly
on
the
file
system.
They
made
various
file
system
codes.
So
if
we
don't
have
a
get
file
system
that
exposes
a
via
trial
file
system,
then
everything
that
we
do
on
resolve
will
not
be.
We
will
not
follow
the
same
code
path
for
the
rest
of
the
node
resolution
algorithm,
and
so
they
will
have
different
results.
C
C
So
the
get
file
system
has
a
different
use
case
than
resolve
plus
load
and
even
the
result
plus
load
can
be
used
for
some
simple
usage,
where
you're
building
some
virtual
file
for
one
specific
file,
it's
not
viable.
Once
you
start
to
get
to
the
use
case
where
you
want
to
basically
emulate
a
folder
on
the
disk.
B
Now
it
sounds
to
me
like
the
you're,
almost
trying
to
do
like
mocking
like
right,
like
you
want
to
like
imagine
like
we're
in
like
jest
or
some
other
type
of
like
unit
testing
environment,
and
you
wanted
to
you
know
like
say:
there's
a
say:
you
have
a
command
line
utility
that,
like
I
don't
know,
opens
a
text
file
and
writes
another
line
to
it
and
saves
it.
Okay,
something
trivial
and
you
want
to
mock.
B
C
And
I
accept
that
in
my
case,
so
you
mentioned
writing
to
the
file
system.
In
my
case,
I
need
to
basically
mock
what's
inside
the
node
itself,
so
it's
it's
not
exactly
the
same
thing.
I
would
not
be
able
to
leverage
the
same
api
for
mocking,
as
I
would
for
the
loaders
that
make
sense.
B
C
C
C
If
we
don't
have
a
way
to
specify
the
logic
that
nod
needs
to
use.
When
making
those
calls,
then
it
will
always
query
the
file
system.
So
if
what
we
are
so
if,
if
the
virtual
files
are
not
on
the
actual
file
system,
all
those
calls
will
fail.
B
C
No,
it
cannot
because
that's
the
other
problem
with
composition.
You
need
different
loaders
to
interact
with
each
other
without
having
to
be
aware
of
the
others.
So
if
jan
was
to
just
tell
this
package
is
at
this
location,
then
it
cannot
just
override
the
whole
resolution.
It
just
needs
to
replace
some
very
specific
part
of
the
of
the
resolution
and
then
let
the
rest
of
the
resolution
go
as
you.
C
D
What
I'm
thinking
is,
I
think,
like
20
minutes
ago,
I
linked
to
a
test
in
the
repository
in
node,
currently
that
deals
with
virtual
files
so
like
none
of
those
things
exist.
So
what
I'm
thinking
is,
if
you
were
able
to
do
that,
and
just
have
a
yarn
say
in
front,
then
it
can
just
tell
you:
don't
have
to
engage
node's
default
stuff.
It
can
just
say
like
forget
about
that.
Here.
Use
this.
B
B
C
C
B
Okay,
so
like
this
should
be
like
I
like,
before
we
build
out
a
whole
other
ui
or
a
whole
other,
you
know
set
of
hooks
like
we
have
to
see
why
it's
not
possible
in
the
current
one,
because
if
there's
two
ways
to
override
loading
files,
then
that's
a
problem
because
like
then,
we
can't
make
that
composable.
B
So,
like
the
the
existing
design
that
we
have
now
like.
If
you,
if
you
look
at
the
that
example,
I
sent
you
of
like
if
you
go
to
the
loader's
repo
and
you
click
on
design
and
you
click
on
the
middleware
design
or
even
if
you
don't
go
into
the
chaining
part,
you
know
like
you
could
just
assume
that
yarn's,
the
only
loader
for
for
now
at
least
it
like
all
the
examples
show
like
a
specifier
of
like
one
of
the
examples,
is
like
https,
where
node
would
just
choke.
B
There's
no
reason
that
that
couldn't
apply
to
just
like
everything
like
you
wanted
to
fly
to
all
specifiers,
because
you
want
to
load
everything
essentially
from
memory
instead
of
from
disk,
but
that
should
be
possible.
Today.
You
just
have
to
like
write
a
hook
that
can
handle
everything.
C
C
B
Like
you
want
an
alternate
like
you
like,
like,
let
me
this
is
another
way.
If
you
were
to
build
this
by
forking
node,
you
would
fork
node
and
edit
the
esm,
loader
and
replace
all
these
like
file
system
calls
with
other
code
that
you
know
calls
you
know
read
something
from
memory
or
whatever
right
like.
That
could
be
one
way
to
build,
plug
and
play
the
the
intent
of
the
loader
hooks.
Is
this
like?
C
If
we
are
replacing
everything,
then
we
are
replacing
everything.
It
means
that
the
other
loaders
don't
have
a
chance
to
do
anything
else.
So
if
you're
using
plug-and-play
with
a
ts
node,
it
will
not
work
because
ts9
will
not
have
a
chance
to
do
anything
because
already
replacing
the
resolution,
so
it
needs
to
have
something
where
jan
can
do
some
some
processing
then
pass
that
to
the
other
resolver,
which
can
then
do
their
own
processing
as
they
want
with
that
around.
D
Can,
oh,
I
think
it
can
do
that
because,
as
long
as
you're
not
hitting
any
of
the
default
things,
then
node
is
not
going
to
do
anything.
All
it's
going
to
do
is
orchestrate
these
the
loaders
and
the
hooks
interacting
with
each
other.
So
let's
say
I
don't
know,
you've
got
yarn
and
ts
node
or
whatever
yarn
does
something,
and
it
provides
virtual
information
and
passes
that,
along
to
whoever
else,
ts,
node
or
whatever
ts
node
says
cool.
D
I
believe
you
and
it
uses
it
and
it
itself
doesn't
make
any
fs
calls
or
anything
like
that.
Then
it.
C
In
this
case,
then,
you
will
have
yarn
that
will
take
the
variant
here
will
resolve
it
so
that
you
have
the
path
virtual
path
and
then
it
will
send
that
to
ts
node.
How
does
ts
node
check
whether
this
file
exists
with
the
ts
extension
in
order
to
return
it,
it
will
need
to
query
the
file
system,
so
either
it
will
go
through
node,
which
is
probably
the
best
option.
C
That's
what
we
discussed
with
with
the
file
system
oops
or
it
will
need
to
use
the
fs
module,
but
the
fs
model
is
not
sorry.
It's
not
he's
not
aware
of
those
special
logic.
D
C
Yes,
that's
what
it
would
implement
inside
the
get
high
system
hook.
So
all
the
other
loaders,
which
would
use
those
books
as
well
in
order
to
make
their
own
festivals
would
be
able
to
just
get
the
right
result
from
yarn
without
having
to
be
aware
that
they
are
operating
with
that
within
a
plug-and-play
context.
D
So
what
I'm
thinking
is
if
the
the
resolve
hook
says
this
is
like
where
the
thing
is
and
what
its
format
is
and
then
the
load
hook
says:
here's
the
stuff
like
here's,
the
source
it
hears
authoritatively.
D
C
C
Jan
just
resolved
the
verizon
tv
here
to
a
folder
but
leaves
the
rest
of
the
path
untransformed
because
he
doesn't
know
what
extension
to
add.
So,
let's
say
that
you
have
a
full
slash
index,
it
will
return,
slash
path,
slash
two
slash
foo
slash
index
and
it
will
pass
that
forward
to
ts
node.
C
C
A
A
D
A
To
check
if
something
exists,
it
would
just
call
like
yarn,
loaders
dot
load
hook
with
the
proper
url,
and
it
would
see
an
error
if
it
doesn't
exist
and
it
could
try
multiple
at
once.
If
it
wanted.
B
Next,
in
the
middleware
version
yeah,
I
think
that
what
what
male
is
saying
is-
I
was
like
it's
part
of
resolution
because,
like
ts,
node
is
trying
to
implement
like
common.js
style
resolution,
where
it's
like
dot
slash
file,
so
ts
node
is
like
well.
Is
there
a
file.ts?
Okay?
No,
is
there
a
file.js?
No,
you
know,
and
so,
as
it's
like
running
through
the
file
system,
to
try
to
figure
out
what
the
file
name
is
to
then
like
it
needs
to
do
that
first
and
then
figure
out
what
the
file
is
to
try
to
load.
A
C
Problem,
which
is
that,
in
a
get
file
system,
we
need
to
have
some
primitives
that
are
faster
than
just
load.
For
instance,
if
you
want
to
check
whether
a
file
exists
or
not,
you
don't
need
to
load
it.
That's
something
that
has
been
raised
before,
which
is
that
some
primitives,
like
I
think
it's
stat
and
red
jason,
are
currently
using
a
specific
native
function
in
order
to
be
fast.
B
My
concern
is
that,
like
you're
now
for
like,
if
we,
if
we
supply
file
system
hooks,
then
we
we
kind
of
break
composability
in
a
different
way
like
yes,
it
makes
composability.
It
adds
the
composability
that
ts
node,
like
you,
have
a
way
to
like
mock
how
ts
node
works
and
like
supply
your
own
answers
for
stat
calls,
but
because
you're
also
mocking
like
read,
file
or
replacing
read
file.
A
Servers
so
load
actually
has
transform
involved
these
days.
So
it's
no
longer
like
a
thing
when
you've
complete
load
you're
expected
to
try
to
coerce
it
properly,
to
something
you
so
transpilation
steps
and
stuff
do
occur.
So,
like
ts
node,
if
you
called
it,
you
wouldn't
actually
see
the
contents
on
disk.
B
Like
right
now,
it's
a
clear
story
for
like
setting
up
a
chain
of
loaders
that
you
know
do
transformations
like
you
could
have
ts,
node
and
then
babble
after
that,
or
something
like
that
right
and
they
would
just
both
be
load
hooks
and
you
would.
It
would
be
clear
how
to
chain
them
you
could,
but
you
could
also
like
implement
ts
node
by
just
using
the
file
system
hooks
and
like
when
a
request
for
file.cs
happens,
ts,
node,
intercepts
it
and
returns
it
as
javascript.
And
now
it's
like
you
know
like
the
load.
E
B
A
Yeah,
so
one
thing
that's
pretty
important
to
note
here
is
even
if
you
tried
to
do
that,
like
various
parts
would
break
so
I
tried
to
do
data
urls
http,
like
you
literally
just
can't
support
him
with
that,
and
you
could
be
like
if
you
try
to
do
this
with
files.
It's
not
going
to
work
because
there
are
modules
that
are
not
file
based
and
so
like,
even
if
somebody
were
to
write
a
transform
of
your
file
system.
A
A
A
That's
it.
They
do
overlap
in
some
ways.
It's
inevitable,
because
loaders
do
use
files
like
that's.
We
call
out
to
file
system
right
now,
so
there's
some
overlap.
There's
delegation
going
on
and
like
trying
to
prevent
any
overlap
is
difficult.
Preventing
problematic
overlap,
I'm
all
for,
but
I
don't
see
like
overlap
where
it's
like.
You
should
never
ever
have
virtual
files,
because
this.
A
Like
we're
talking
about
allowing
virtual
files
via
the
load
hook
and
like
that's
complicated
because
the
load
hook
is
actually
like
trying
to
return
executable
stuff,
not
the
source
text
of
a
file.
A
So
if
you
have
a
tool,
that's
like
trying
to
get,
I
don't
know
line
information
to
show
where
an
error
is
it's
going
to
be
lost
in
your
load
hook.
You
got
to
start
putting
source
maps
in
there
just
to
act
like
that,
and
we
don't
have
a
hook
for
when
we
read
package
json
like
resolve,
doesn't
expose
it
right
now
we
could
have
some
utility
functions
that
you
recreate
all
of
node's
loader
with,
but
then
you're
having
composition,
problems
like
miles
pointing
out.
A
B
B
B
C
B
B
Some
examples
of
how
to
use
it
and
then-
and
one
of
the
examples
should
be
yarn,
plug
and
play,
but
also
like
something
simple
like
a
what's
whatever
the
simplest
hello
world
you
can
think
of
is
ideally
something
that's
not
also
already
achievable,
using
resolve
and
load,
and
then
we
can
start
iterating
on
that
and
I'd
say
like
let's
get
that
nailed
down
before
we
try
to
implement
any
pr's.
B
I'd
also
like
to
get
chaining
finished
before
we
add
this,
because
I
feel
like
that's
important,
to
get
that
working
and
and
out
there,
and
then
we
layer
this
on
and
make
sure
that
we
don't
break
chaining
load
because
that'll
be
the
challenge
of
like
that's
the
that's.
The
ultimate
challenge
with
this
is:
how
do
we
deal
with
the
intermix
composability
of
load
and
like
read,
file
right?
So
if
we
can
get
change
load
first,
then
that'll
already
be
like
there
as
a
basis
from
from
when
we
add
this.
On
top
of
it.
C
B
Okay,
and
maybe
the
design
is
you
know
exactly
what
you
have
in
the
pr
and
I'm
assuming
that's,
probably
what
your
first
pass.
The
design
doc
will
be
is
just
describing
what's
in
the
pr,
but
I
also
think
that,
like
having
some
examples
like
as
intended
for
users
to
use
like
look
at
the
middleware
design
dock
in
in
the
loader's
repo,
it's
basically
a
series
of
examples
that
loader
authors
might
write
to
use
this
api.
C
That's
something
that
I'd
like
to
propose
later,
but
for
now
I
right.
A
My
child
was
sick
for
a
few
weeks,
but
I
was
planning
to
finish
it
over
christmas
break.
We
have
everything
we
need
to
move
them
off
thread.
Then
they
can
intercept
common
js,
fine.
A
A
A
Yeah
it'll,
be
their
performance,
is
the
same
astro
just
built
a
thing
that
could
make
performance
better
on
a
worker,
even
if
we,
even
if
we
just
block
the
main
thread,
because
they
have
a
shared
hash
map
structure,
but
it
would
mean
that
loaders
couldn't
use
global
state
reliably,
because
any
given
loader,
at
least
in
astros
design,
can
have
two
instances
running
at
the
same
time
with
two
different
globals.
A
A
Okay,
but
astro
saw
60
percent
perf
gain
when
they
did
it
right.
So
it's
all
about
what
you're
willing
to
do.
I
don't
think
we'll
do
what
astra
did,
because
you
would
need
to
write
very
custom
code
to
do
that.
F
B
All
right,
so
I'm
just
writing
in
the
in
the
agenda
and
the
notes
document
the
google
doc
so
next
steps
mail,
writes
the
design
dock
for
file
system
hooks.
If
you
could
submit
that
as
a
pr
to
the
loaders
repo.
Just
look
at
the
other
design.
C
B
B
B
So
that
was
added
as
a
pr
against
the
loaders
repo,
so
just
add
a
new
file.
You
know
somewhere
appropriate
another
markdown
file
in
this
repo
and
submit
it
as
a
pr
and
we
and
make
it
similar
to
the
ones
you
see
there
where
it's
like.
This
is
the
api
signature
of
the
hooks,
and
this
is
like
some
examples
of
how
to
use
them,
and
then
we
can
like
nail
down
what
this
api
should
be
by
iterating
on
that
pr
and
then
once
that's
landed
in
here,
then
that's
like
okay.
B
This
is
like.
We
have
consensus
that
this
is
what
we
want
to
build
and
then
it's
like,
then
your
pr
implementing
it
can
just
be
about
like
the
implementation
details
and
not
about
like
the
design
of
it.
B
You
know
what
I
mean
because,
like
that's
bitten
us
in
the
past,
where,
like
someone
tries
to
do
them,
both
together
in
a
pr
against
node
like
the
like
with
code
and
it's
just
becomes
unmanageable,
because
you
have
like
hundreds
of
comments
on
the
pr
thread
and
you
can't
get
any
work
done
so
so
yeah.
If
that
works
for
you
so
that'll
be
the
next
step
for
the
file
system,
one
and
then
the
other.
So
but
I'm
I'm
putting
down
here
this
deadly.
The
next.
D
A
Plus
they're
already
off
thread
javascript
wise.
If
people
are
mucking
around
with
fs
and
we're
putting
this
as
generic
fs
hooks
rather
than
scope
to
loading,
I
know
the
old
mpm
crew
still
is
mucking
around
with
fs
for
their
run
time,
it's
probably
gonna
break
in
tropic.
B
A
B
C
Though
it
would
be
not
to
break
those
existing
use
cases
before
there
is
a
replacement
that
is
ready
to
be
to
be
used.
Okay,.
B
A
A
So
if
a
loader
doesn't
call
the
default
resolve
when
a
common
js
request
comes
in,
that's
when
it
wouldn't
go
back
to
the
main
thread
and
you
break
compatibility
but
we'll
have,
there
is
a
path
that
was
designed
for
allowing
it.
B
All
right,
but
getting
back
to
the
to
what
I
was
writing.
Okay,
maybe
the
off
thread
doesn't
need
to
happen
before
the
file
system
hooks,
but
it
sounds
like
it
will
anyway,
just
because
it's
closely
being
finished
and
we
have
to
go
through
a
design
phase
for
file
system
so
either
way
the
same
same
result
in
the
end
sure
how
much
time
do
you
have
in
the
next
few
weeks?
Now?
C
B
Okay,
because
the
other
thing
I
was
going
to
say,
is
like
here
I'll
put
this
just
like.
B
That
it
could
possibly
go
after
file
system
folks,
if
you,
if
we
finish
the
whole
design
process
quickly
but
then
chaining
still
isn't
landed.
Yet
jacob
is
as
deep
in
https
imports,
so
you
too
might
want
to
talk
and
see
if,
like
you
know,
if
you're
in
a
hurry
to
get
file
system
done,
maybe
you
could
help
with
chaining
and
get
that
out
the
door
so
that
the
prerequisite
for
file
system
is
done.
D
I
think
the
https
imports
thing
is
nearly
done.
I'm
fixing
up
the
last
few
test
cases
or
fixing
up
the
things
that
are
not
working
and
the
test
cases
are
proving
that
they're
not
working,
but
I
think
that
should
be
ready
soon.
A
B
B
A
long
time
to
review
this
this
pr,
so
that
could
that
could
occupy
you
for
a
while,
but
hopefully
that,
like
a
lot
of
that,
time,
will
just
be
like
waiting
on
people
to
get
back
to
you.
So
maybe,
if
you
have
time
you
can
get
started
on
chaining,
while
you're
like
waiting
on
crap
from
from
the
https
vr.
D
B
Okay,
well,
anyway,
we
can
play
this
by
ear
and
talk.
B
A
D
D
F
B
A
B
B
A
C
B
A
B
Do
that
later,
that's
that
was
we.
B
Right
now
and
it
not
being
a
break
change
you
could
follow.
I
would
get
to
do
the
offset
thing
first,
because
that
seems
more
important,
but
then
I
think
once
chaining
is
done.
It
makes
more
sense
to
add
this,
like
other
references
to
other
loader
stuff,
once
chaining
is
implemented,
so
you
have
other
loaders
to
reference
so
begin
land
chaining
and
then
you
can
land
this
like
api,
refactor,
sure,
okay,
so
loaders
loading,
while
loading
loaders.
C
Is
that's
really
doable,
but
I
think
it's
bound
to
happen
that
someone
will
want
to
make
dash
loaders
ts,
not
dash
dash
loaders
or
something
else
that
is
written
with
that
all
right.
If
yarn,
it
will
happen
that
you
will
want
to
reference
the
loader
he
wants
really
entrenched.
Yes,.
D
A
A
A
I
think
this
is
an
extremely
rare
use
case.
Personally,
is
it
why,
like
like
it,
does
all
sorts
of
weird
stuff
to
have
reentrancy
like
this?
Is
why
the
specs
for
javascript
html,
like
avoid
re-interncy
like
it's,
the
devil
like
it,
is
really
hard
to
reason
about
your
data
model.
If
your
data
model
becomes
cyclic,
but
it's
not
secret,
though
it
would
become
cyclic
unless
you
have
like,
because
now
you
have
like
bi-directional,
so
data
comes
in
from
the
import
site
goes
through.
A
C
A
It
only
knows
that
it
can
call
the
functions
of
the
next
loader.
It
doesn't
have
like
a
file
name,
it
doesn't
have
yeah,
it
doesn't
know
what
it
is.
C
B
The
last
thing
hold
that
hold
on
for
a
second
loaders
are
urls
right,
so
they.
B
A
A
B
B
You
this
mail,
if
you
wanted
to
implement
this
say
you
had
plug
and
play
as
a
global
like
plug
and
play
itself,
has
to
be
on
disk
right.
So
it's
like
npm
install
dash
g
plug
and
play
or
something
along
those
lines
right.
So
then
you
could
have
like
tmp
or
whatever
the
you
know.
Binary
is
pnp
stress,
loader,
ts,
node
dash
just
loader,
or
something
else
you
know
entry.js
or
whatever
plug
and
play
could
parse
the
command
line.
C
It
seems
far
too
complicated
and
maybe
not
complicated.
It's
not
that
complicated
to
parse
the
command
line.
It's
mostly
that
there
are
edge
cases
that
we
are
not
going
to
think
about,
for
example,
what
if
you're
go
sending
dash
dash
loader?
I
don't
know
through
to
another
process
that
you're
forwarding
the
arguments
proxying
and
making
mutation
at
this
level,
I
think,
is
bound
to
fail
at
some
point.
A
F
A
C
First,
the
pnp
loader
is
evaluated.
Then
it's
called
to
resolve
ts
node.
Then
it
returns
and
then
node
keeps
going
by
then
calling
the
pmp
with
a
different
next
for
this
time,
because
for
the
first
time
the
next
is
the
default
loader,
the
second
time
the
next
is
ts
another.
C
C
A
A
So
one
of
the
things
that
already
exists
is
loader
chains
are
not
the
same
across
worker
threads,
one
worker
thread
could
have
ts
node
installed,
one
could
have
coffeescript
installed,
I
mean-
and
so
in
this
case,
if
they
instrument
each
other,
you
can't
reuse
a
loader
chain
across
stuff
right
now,
they're
they
don't
instrument
each
other,
so
your
loader
chain
isn't
dependent
on
the
other
loaders
in
your
chain.
This
would
make
if
you
have
three
loaders,
a
b
c
in
order
to
reuse
it.
A
A
Sure,
right
now
I
could
spin
up
a
loader
as
a
you
know,
service
process
and
just
leave
it
there
effectively
for
node,
because
it's
not
instrumented
anywhere.
You
can't
like
muck
around
with
it
really
from
another
loader.
A
So
if
ts
node
has
a
language
server,
which
it
spins
up
as
a
service
process
which
it
does
currently
in
tsc,
the
command
line,
it'll
spin
up
a
background
process
to
handle
stuff
and
keep
it
warm
and
cached,
and
everything
we
can
do
that
right
now
and
it's
not
really
a
problem.
A
C
A
A
C
A
A
C
A
C
Yes,
but
in
that
case
it
would
be
the
same
thing
except
that
one
in
one
instance,
you
would
have
pnp
coffee
street
and
what's
the
other
one,
you
said
a
coffee
script
on
typescript
and
in
another
thread
you
could
have
just
pnp
and
coffeescript,
but
not,
but.
B
Guys,
I'm
gonna
make
bradley
the
host
because
I
need
to
get
going.
I
assume
that
some
other
team
is
coming
any
minute
because
there's
another
meeting
on
the
schedule
at
noon,
but
until
they
show
up.
A
A
C
So
I
think
I
understand
what
you
mean,
but
we
already
do
that,
for
example,
in
yarn
we
already
have
the
required
pnp
dutch
require
ts9,
a
term
that
has
to
be
followed,
and
it's
it's
already
been
used
with
a
fair
amount
of
success.
Of
course
we
have
to
monkey
bash
things
because
there
isn't
the
primitives
that
we
need,
but
overall
it
seems
to
work.
A
C
C
A
In
usually,
there
are
different
scenarios
where
that's
not
always
the
case.
Maybe
you
want
to
spin
up
two
instances
of
babel
because
it's
cpu
heavy
and
not
differentiate
them
and
stuff
like
that,
but
the
moment
that
they
can
instrument
each
other,
I'm
unclear,
I'm
not
saying
it's
impossible.
A
So
another
example
is
startup.
Time
is
a
big
problem
if
you're
requiring
every
loader
to
finish
loading
before
you
start
loading
the
next
loader
I
mean
you
really
want
really
fast
to
load
loaders
at
that
point,
so
you're
going
to
want
to
avoid
static
imports.
A
Any
loader
is
determined
entirely.
It
is.
There
is
no
such
thing
as
any
of
the
other
loaders,
basically
only
the
first
loader
matters,
because
it
can
intercept
and
destroy
or
replace
or
mock
or
whatever
any
of
the
subsequent
loaders.
Your
cli
doesn't
matter
what
you
passed
in.
Only
the
last
loader
is
your
security
model.
C
D
Like
you're
saying,
if
you
have
babble
and
typescript
and
something
something
you're
expecting
babel
to
be
able
to
manipulate
the
typescript
loader,
I
wouldn't
really
expect
that
I
would
expect
them
to
be
independent.
C
D
It
like
opts
into
doing
so,
like
the
https
loader.
Let's
say
that
wasn't
in
node
like
that,
loader
would
be
responsible
for
deciding
whether
it
does
something.
A
When
you
can't
make
any
static
guarantees
about
what
a
loader
does
so
like,
I
write
a
top
level
loader
and
I
just
prevent
all
other
loaders
from
functioning
like
that's
a
valid
use
case.
Under
this
model,
I
can
basically
be
like
I
want
to
like
ignore.
What's
on
the
command
line,
arguments
and
stuff
like
that.
A
Always
loaded,
it's
always
loaded
right
now,
so
you
can
short
circuit,
which
is
yeah.
It's
a
little
bit
different
here,
you're,
basically
saying:
there's
nothing
on
the
command
line.
No
loader
can
protect
itself
because
they're
all
under
control
of
the
other
loaders.
C
D
A
Loaders,
the
question
is:
do
you
need
all
the
hooks,
or
do
you
just
need
the
file
hook
or
like?
What
do
you
need
to
enable
your
use
case
with
pmp?
Is
it
do
you
need
all
loaders
to
be
able
to
do
this?
Does
a
loader
have
to
manually
say
they
need
this
behavior
like
there's,
all
sorts
of
things
that
you
can
work
around
like
pnp
could
set
some
variable
somewhere
and
it'd,
be
like
I'm
a
loader,
but
I'm
about
to
hijack
all
the
other
loaders.
A
A
D
C
C
C
So
we
could
have
something
like
dash
dash
ambient
loaders
sure
that
are
loaded
before
everything
else
in
secondary
and
then
everything
else,
good.
That's
fine.
D
A
Okay.
So
like
every
time
you
do
it
you're
just
adding
another
ambient
on
the
end.
Basically,
okay,
I
mean
that's
fine,
like
node,
won't
be
able
to
do
any
tricks
to
save
memory
or
cpu.
With
these
things,
which
is
fine.
C
A
Yeah,
that's
fine!
Okay,
but
like
there
have
been
several
companies
that
started
moving
to
worker
threads
babbles,
moving
to
worker
threads,
all
these
things
and
just
the
cpu
and
memory
the
memory
usage
goes
up
a
lot.
If
you
don't
reuse
your
worker,
so
you
can
easily
break
a
gig
on
accident,
doing
it
with
memory
usage,
so
the
last
the
most
ambient
loader
we
could
probably
reuse,
but
all
the
other
ones
you
couldn't.
C
A
D
Right
are
we
possibly
going
to
run
into
some
kind
of
stateful
situation
within
the
ambient
loaders
where,
like
let's
say,
you've
got
five
threads
and
your
ambient
loader
over
here
is
behaving
differently
than
the
other
ambient
loaders.
D
A
The
ambient
loader
it's
whatever
the
ambient
loader
said
it
could
do
like
you
could
load
with
in
electrons
case,
like
you
could
load
where
one
worker
is
using
one
sr
file
and
the
other
is
using
a
different
asar
file.
However,
they're
pronounced
and
so
they'd
inherently
have
different
file
system
operations,
but
then
you
just
have
insanity.
A
A
C
So
I
can
make
a
design
dock
as
well.
For
that,
yes,.
A
So
basically,
we
have
the
separate
module
map
right
now.
Instead
of
having
a
separate
module
map,
that's
shared,
we
wouldn't
do
any
of
that.
We
would
basically
spin
up
a
user
land
application
for
a
loader.