very unsound assumptions about REPL state injection

[vtjnash]

This code runs on a thread, which makes this an unsafe data race to call this function:

Pkg.jl/ext/REPLExt/REPLExt.jl

Lines 311 to 321 in 8c99679

	if isdefined(Base, :active_repl)
	repl_init(Base.active_repl)
	else
	atreplinit() do repl
	if isinteractive() && repl isa REPL.LineEditREPL
	isdefined(repl, :interface) || (repl.interface = REPL.setup_interface(repl))
	repl_init(repl)
	end
	end
	end
	push!(empty!(REPL.install_packages_hooks), try_prompt_pkg_add)

It is VERY strongly inadvisable to have an __init__ function defined in this code, due to the unreliability of the environment it runs in, and the corruption this will cause to the REPL module.

[KristofferC]

I'm missing a solution here. This is how external REPL modes are installed AFAIU.

[vtjnash]

You install REPL hooks by calling empty!? that doesn't seem very scalable 😭

[IanButterworth]

empty! explained and fixed here #3934

[KristofferC]

You install REPL hooks by calling empty!

You linked a much larger piece of code than that single one.

[vtjnash]

The other major mistake here is calling Base.active_repl and assuming that returns REPL.active_repl