[REPL] Autocompletion of `ModuleName.@macro_name` is broken

[giordano]

On 5230d27

julia> Base.@time nothi<TAB>

autocompletes nothi to nothing_sentinel:

julia> Base.@time nothing_sentinel
ERROR: UndefVarError: `nothing_sentinel` not defined in `Main`
Suggestion: check for spelling errors or missing imports.
Stacktrace:
 [1] macro expansion
   @ ./timing.jl:579 [inlined]
 [2] top-level scope
   @ ./REPL[17]:1

If I understand it correctly, ModuleName.@macro symbol<TAB> autocompletes symbol inside ModuleName, not Main:

julia> module Issue55518
       macro name(expr) :($(esc(expr))) end
       const not_really_what_I_wanted = something
       end
Main.Issue55518

julia> using .Issue55518

julia> Issue55518.@name not<TAB>

This works correctly in Julia v1.10 and v1.11.0-rc1

[giordano]

I just ran again into this by trying to autocomplete in the REPL a Threads.@threads for loop.

[giordano]

We have an upgrade: also autocompletion of keyword arguments is now broken:

julia> module Issue55518
       macro name(expr) :($(esc(expr))) end
       const not_really_what_I_wanted = something
       f(; x) = x
       end
Main.Issue55518

julia> using .Issue55518

julia> Issue55518.f(; x=<TAB>
@name
eval
f
include
not_really_what_I_wanted

Trying to autocomplete the value of the keyword arguments complete only symbols within the module, also those that aren't exported and so you shouldn't even be able to reference them unless you imported them explicitly, which I didn't here.

[giordano]

More on this:

julia> module Issue55518
           macro name(expr) :($(esc(expr))) end
           const not_really_what_I_wanted = something
           f(; x) = x
       end
Main.Issue55518

julia> using .Issue55518

julia> Issue55518.f; nothi<TAB>

doesn't complete the symbol after the semicolon, same for macros:

julia> Issue55518.f; @ti<TAB>

Basically after you type the module name you're damned to complete things within the module in the entire line, including after a semicolon to separate expressions.

[IanButterworth]

@aviatesk any idea on this one? I think Mose's last summary above is key, I just can't figure out where the restriction to the module is happening.

[aviatesk]

It looks like the prefix given to complete_symbol! is being something like :Issue55518 in those cases

julia/stdlib/REPL/src/REPLCompletions.jl

Line 1219 in 5b677a1

complete_symbol!(suggestions, prefix, name, context_module; complete_modules_only, shift)

, which causes the method to use the prefix as the module context:

julia/stdlib/REPL/src/REPLCompletions.jl

Lines 161 to 177 in 5b677a1

	res = repl_eval_ex(prefix, context_module)
	res === nothing && return Completion[]
	if res isa Const
	val = res.val
	if isa(val, Module)
	mod = val
	if !shift
	# when module is explicitly accessed, show internal bindings that are
	# defined by the module, unless shift key is pressed
	complete_internal_only = true
	end
	else
	t = typeof(val)
	end
	else
	t = CC.widenconst(res)
	end

So I think we need to fix the prefix calculation that happens within complete_identifiers!.

[IanButterworth]

I assume this is related to #54858 ? I see that's not on 1.11 and Base.@time nothi<TAB> works on 1.11

[giordano]

Other related breakage:

x = sin.([1]); y = ex<TAB>

completes to export, instead of exp. The dot again confuses everything. @aviatesk can you please take a look at this? TAB completetion in presence of dots horribly broken, making the REPL experience very frustrating.

[IanButterworth]

We may want to revert the PR that introduced module specific completion. I think the experience may be worse than before that.