fix: activate a temp environment when a sysimage is enabled

[pankgeorg]

Starting with Julia 1.8, the sysimage internal manifest (here on Ⅿ₀) is included in every Pkg resolution. So, a new, fresh manifest is equivalent to Ⅿ₀ ∪ ∅. This is strictly better than the "default" manifest that can be inconsistent with the Ⅿ₀, or may include packages that will interfere with the new packages we want to add (e.g. an old version of a package that the package you're adding is depending on).

To address that, the proposal is to instantiate a fresh manifest (temp=true) instead of the default.

[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/fonsp/Pluto.jl", rev="pg/activate-temp")
julia> using Pluto

[fonsp]

Oooh and can we get that internal manifest? is there also a project? in that case we should activate that internal environment, right?

[pankgeorg]

Oooh and can we get that internal manifest?

I don't think we can get the manifest, but the Pkg knows about it.

is there also a project? in that case we should activate that internal environment, right?

Hmm that's a good point. The most important repercussion I think is "What to serialize in cells "0000-...0001" and "0000-...0002", and I don't have a good answer, but I'll ask around

[fonsp]

I think the original assumption was that JH will open Pluto notebooks with a sysimage, and with the Manifest-and-Project-that-was-used-to-generate-the-sysimage stored as the global environment. If people were to use a Julia REPL, we want them to use the global environment, and this way we match that on Pluto.

Is that no longer the case? Happy to merge either way because JH is probably the only user.

[fonsp]

@pankgeorg

[pankgeorg]

I think the original assumption was that JH will open Pluto notebooks with a sysimage, and with the Manifest-and-Project-that-was-used-to-generate-the-sysimage stored as the global environment. If people were to use a Julia REPL, we want them to use the global environment, and this way we match that on Pluto.
Is that no longer the case? Happy to merge either way because JH is probably the only user.

I'm not sure that this is the case currently. Pluto runs with the sysimage and opens the notebooks with the sysimage. But Pkg.activate() doesn't necessarily activate the sysimage environment. And I think we don't need to either. Starting with Julia 1.9, I noticed that the internal pluto package manager does work with sysimages; it's just terribly slow. Maybe the right fix here is to teach it to be faster (e.g. ignoring or fast-pathing the resolution of Base._sysimage_modules). I need to investigate this a bit more.