Restrict notebook communication topology

[Oblynx]

Instigated by #300

In case Pluto wants this, some tests would have to change, because they verify that notebooks can talk to each other.
Maybe workers are intended to be able to talk to each other.

[github-actions]

Try this Pull Request!

Open Julia and type:

julia> import Pkg
julia> Pkg.activate(temp=true)
julia> Pkg.add(url="https://github.com/Oblynx/Pluto.jl", rev="restrict-distrib-comms")
julia> using Pluto

[pankgeorg]

Hey!

I have two concerns

1. this feature is listed as experimental
2. it's not clear what we gain from this; Pluto does not make "security" promises at the moment so it feels limiting without a clear upside (We don't protect you from 'adversary notebooks' - notebooks are yours!)

Also I'm not sure if we actually need this to be open (@Pangoraw_ do you remember?)

Nevertheless, thank you for spending time on this! 💪🏾💪🏾

[Oblynx]

I agree with your concerns. I'd love to be able to use Distributed and would argue that Pluto has to support every part of Julia stdlib if it's to be a Julia IDE. Regarding notebook isolation though, I have no other reason to show this apart from Fons's comment in #300

[pankgeorg]

We are working on broad stdlib support! Latest major addition was the macro support 😄!
Until then we don't even claim to be Julia IDE 🙈🙈🙈 just simple reactive notebooks in Julia 🎈🎈🎈focused on education and teaching!

[Oblynx]

Latest major addition was the macro support

I can't express how important that is! Thank you so much for all your good work! I hope I can also help in a small way as a "power user" :)

[dralletje]

SHUT UP

This is literally all that is necessary to allow Distributed inside Pluto???
Also why can't workers talk to each other now?

• Throws away Distributed clone 🙃 🥲

EDIT
Actually looked at the docs now.. awesome!
Being able to call code on the master process from the worker was a fun feature for tinkering around, but this is definitely what we want (as long as we don't sell it as "secure" because it ain't)

EDIT AGAIN
Okay got too excited, sadly doesn't make Distributed work.
I think this flag isn't too helpful (yet?), as it still allows worker->master eval even..

[Oblynx]

sadly doesn't make Distributed work

No this isn't about making Distributed work, for that check the last discussion in #300

as it still allows worker->master eval even

Really? Then I'm not sure if there's any point in this feature, maybe a bit of performance gain. Why don't you raise this up with Distributed upstream, to see what their use case is, since they mark this flag as "Experimental"