show() methods aren't dependencies

[cmcaine]

show() methods aren't considered when determining which cells to re-evaluate, so you can end up with a situation like this where the rendered notebook is inconsistent with how it would look if you re-evaluated cell 2:

Suggested fix: maintain a dependency DAG for outputs as well as cells; outputs depend on the cell that calculated them and on the definitions of show or showerror.

So the definition of a more specific showerror method in cell 3 will cause the output of cell 2 to be re-rendered (but won't cause cell 2 to be re-evaluated).

I suggest this partly because the DAG looks like it works on symbol names rather than methods and it would be a pain to invalidate every cell any time a method is added to show or showerror or whatever.

Reproducible example:

### A Pluto.jl notebook ###
# v0.11.9

using Markdown
using InteractiveUtils

# ╔═╡ 4f2dc5b0-ea04-11ea-3f4c-073ac0ed718b
struct MyException <: Exception
	msg::String
end

# ╔═╡ a46df5e0-ea04-11ea-0700-df5e78befa62
throw(MyException("foo"))

# ╔═╡ 35271030-ea05-11ea-00e8-2328e8688cb7
Base.showerror(io::IO, e::MyException) = print(io, "MyException: ", e.msg)

# ╔═╡ 26b7b090-ea05-11ea-2edd-09565dee8c58
throw(MyException("foo"))

# ╔═╡ Cell order:
# ╠═4f2dc5b0-ea04-11ea-3f4c-073ac0ed718b
# ╠═a46df5e0-ea04-11ea-0700-df5e78befa62
# ╠═35271030-ea05-11ea-00e8-2328e8688cb7
# ╠═26b7b090-ea05-11ea-2edd-09565dee8c58

[fonsp]

Thanks! The suggested solution is too complex, but we can also just re-show every cell when a new show method is defined?

[cmcaine]

That sounds like it would have the same effect 👍

[cmcaine]

I thought I might take a look at this. Are you open to a PR on it or have you already started on it or something?

Would you like to give some pointers on where to start or how you'd like it done? (optional)

[fonsp]

That would be great!

I think that you should check if any function defines a method for show or Base.show here:

Pluto.jl/src/evaluation/Run.jl

Line 82 in cc7248d

and if so, call this on every cell:

Pluto.jl/src/evaluation/Run.jl

Lines 89 to 98 in cc7248d

	function run_single!(notebook::Union{Notebook,WorkspaceManager.Workspace}, cell::Cell)
	run = WorkspaceManager.eval_format_fetch_in_workspace(notebook, cell.parsedcode, cell.cell_id, ends_with_semicolon(cell.code))
	cell.runtime = run.runtime
	cell.output_repr = run.output_formatted[1]
	cell.repr_mime = run.output_formatted[2]
	cell.errored = run.errored
	return run
	end

except it should not run again, it should only call the output fetcher.

[cmcaine]

I'm looking at adding a test in test/React.jl, which I think is the right place?

And while there I see this line:

Pluto.jl/test/React.jl

Line 84 in cc7248d

begin

I think you probably want let, rather than begin. begin does not introduce a new scope.

I can open a PR for that, too.

[fonsp]

Haha thanks but that's too small to spend time on