Group import hints

[ajwheeler]

This is my attempt to resolve #53000.

julia> solve
ERROR: UndefVarError: `solve` not defined in `Main`
Suggestion: check for spelling errors or missing imports.
Hint: a global variable of this name also exists in CommonSolve.
    - Also exported by SciMLBase.
    - Also exported by DiffEqBase.
    - Also exported by JumpProcesses.
    - Also exported by LinearSolve.
    - Also exported by BracketingNonlinearSolve (loaded but not imported in Main).
    - Also exported by SimpleNonlinearSolve.
    - Also exported by NonlinearSolve.
    - Also exported by OrdinaryDiffEqLowStorageRK (loaded but not imported in Main).
    - Also exported by OrdinaryDiffEqSSPRK (loaded but not imported in Main).
    - Also exported by OrdinaryDiffEqVerner (loaded but not imported in Main).
    - Also exported by OrdinaryDiffEqBDF (loaded but not imported in Main).
    - Also exported by OrdinaryDiffEqTsit5 (loaded but not imported in Main).
    - Also exported by OrdinaryDiffEqRosenbrock (loaded but not imported in Main).
    - Also exported by OrdinaryDiffEqDefault (loaded but not imported in Main).
    - Also exported by Sundials.

I would have beefed up the test case, but can't follow how it works.

[Keno]

Seems sensible overall, modulo comment.

[Keno]

Does need a test though.

[LilithHafner]

There's also some trailing whitespace that needs to be removed. You can see where on the "files changd" tab of this PR. (Some editors have an option to strip trailing whitespace automatically)

[ajwheeler]

In thinking about how to test this, I've come across what may be a problem. If I do

module A
export f
f() = 0.0
end

module B
import ..A: f
export f
end

module C
import ..B: f
public f
end

module D
public f
f() = 1.0
end

append!(Base.loaded_modules_order, [A, B, C, D]) # is there a less gross way?
import .B

Then the hints for julia> f look like this

ERROR: UndefVarError: `f` not defined in `Main`
Suggestion: check for spelling errors or missing imports.
Hint: a global variable of this name also exists in Main.A.
    - Also exported by Main.B (loaded but not imported in Main).
    - Also made available as public by Main.C (loaded but not imported in Main).
Hint: a global variable of this name also exists in Main.D.

The problem here is that (unlike C), B is imported. edit: I now believe that neither B nor C should get this parenthetical, since they are defined in Main

The same thing happens with 1.11.

hints on 1.11

julia> f
ERROR: UndefVarError: `f` not defined in `Main`
Suggestion: check for spelling errors or missing imports.
Hint: a global variable of this name may be made accessible by importing Main.A in the current active module Main
Hint: a global variable of this name may be made accessible by importing Main.B in the current active module Main
Hint: a global variable of this name may be made accessible by importing Main.C in the current active module Main
Hint: a global variable of this name may be made accessible by importing Main.D in the current active module Main

I think the problem arises from calling Symbol(m), where m is a Module, but I might be confused.

I would welcome input here about whether or not this is out of scope for this PR, and whether there is a better way to test this.

[ararslan]

I think the problem arises from calling Symbol(m), where m is a Module, but I might be confused.

I haven't looked closely enough to determine whether Symbol(m) is the problem but if you suspect it is then you could try nameof(m) instead. (However, note that nameof here isn't fully equivalent; consider e.g. Symbol(Base.Broadcast) vs. nameof(Base.Broadcast). To get the same behavior, you'd need Symbol(join(fullname(m), '.')).)

[ararslan]

By the way, GitHub says this is your first contribution to the Julia repo, so welcome, and excellent first PR!

[ajwheeler]

I've added @test fail in the catch block here, which seems like the right thing to do? Otoh, the other REPL tests don't do it.

[ararslan]

I think this query might be inaccurate if active_mod has a binding of some kind with the same name as m but referring to something different.

That said, this probably won't be worse than the current behavior, which is rather silly (note the difference in hints here):

julia> using StatsBase

julia> StatsBase.StatsAPI
ERROR: UndefVarError: `StatsAPI` not defined in `StatsBase`
Suggestion: check for spelling errors or missing imports.
Hint: StatsAPI is loaded but not imported in the active module Main.
Stacktrace:
 [1] getproperty(x::Module, f::Symbol)
   @ Base ./Base_compiler.jl:47
 [2] top-level scope
   @ REPL[2]:1

julia> using StatsBase

julia> StatsAPI = "🅰🅿ℹ";

julia> StatsBase.StatsAPI
ERROR: UndefVarError: `StatsAPI` not defined in `StatsBase`
Suggestion: check for spelling errors or missing imports.
Hint: a global variable of this name also exists in StatsAPI.
Stacktrace:
 [1] getproperty(x::Module, f::Symbol)
   @ Base ./Base_compiler.jl:47
 [2] top-level scope
   @ REPL[5]:1

[ajwheeler]

I think I'm now handing this case correctly. (See updated test.)

[ararslan]

An additional important case to test is that of available but unresolved bindings, e.g.

julia> module AAAA
           export f
           f() = 0.0
       end
Main.AAAA

julia> module B
           using ..AAAA
           export g, f
           g(x) = 69x + 420
       end
Main.B

julia> Base.isbindingresolved(B, :f)
false

julia> isdefined(B, :f)  # note that `isdefined` forces binding resolution
true

[ajwheeler]

Interesting, thank you. What is the best place to find docs on this? Is getting the desired behavior just a matter of changing Base.isbindingresolved to isdefined in line 81?

        if !Base.isbindingresolved(m, var) || (!Base.isexported(m, var) && !Base.ispublic(m, var))

[ajwheeler]

I've now done that.

[ajwheeler]

Friendly post-holiday bump :)

I think the failures here were also on master and that this is ready modulo additional review.

[ajwheeler]

Ping :)

Are the failing tests here my fault or just flaky?

[vtjnash]

there was a merge conflict, so CI needs to re-run anyways, and I fixed an incorrect comment (this PR appears to suppress messages for modules where the binding exists but doesn't currently have a value assigned to them) which didn't get updated in f2885c6 after the code change there