"Found raw symbol in code returned from lowering. Expected all symbols to have been resolved to GlobalRef or slots."

[KristofferC]

These two packages encounter this:

https://s3.amazonaws.com/julialang-reports/nanosoldier/pkgeval/by_hash/c78016e_vs_d63aded/DocSeeker.primary.log
https://s3.amazonaws.com/julialang-reports/nanosoldier/pkgeval/by_hash/c78016e_vs_d63aded/DRIPs.primary.log

I haven't checked if these do anything "crazy" but putting it up here for tracking.

[Keno]

Let me assign @mlechu to take an initial look. This is likely in lowering and @mlechu has been looking at the .scm code recently.

[mlechu]

Looks like two separate issues.

DRIPs

Reduces to:

r = 
    try
        function ge_drip()
            iter = 0
            while (iter < 10)
                global ge = 1
                iter += 1
            end
            return ge
        end
        ge = ge_drip()
    catch x
    end

DocSeeker

Notice anything wrong here? (I'm not sure if/how this worked in the first place). The error we get is still a regression.

1.11:

julia> function foo()__
       1
       end
foo (generic function with 1 method)

julia> foo()
ERROR: UndefVarError: `__` not defined in `Main`
Suggestion: check for spelling errors or missing imports.
Stacktrace:
 [1] foo()
   @ Main ./REPL[1]:1
 [2] top-level scope
   @ REPL[2]:1

nightly:

julia> function foo()__
       1
       end
ERROR: Found raw symbol in code returned from lowering. Expected all symbols to have been resolved to GlobalRef or slots.
Stacktrace:
 [1] top-level scope
   @ REPL[1]:1

[Keno]

function foo()__

I believe the intended behavior here is a lowering-time error for use of an underscore symbol in value position

julia> ___
ERROR: syntax: all-underscore identifiers are write-only and their values cannot be used in expressions
Stacktrace:
 [1] top-level scope
   @ REPL[1]:1

[Keno]

julia> x = @Meta.lower r =
           try
               function ge_drip()
                   iter = 0
                   while (iter < 10)
                       global ge = 1
                       iter += 1
                   end
                   return ge
               end
               ge = ge_drip()
           catch x
           end;

julia> typeof(x.args[1].code[8].args[3].code[end - 1])
Symbol

should have been a globalref. That said, we seem to have a larger problem with scoping here:

julia> try; f() = (global ge = 2; ge)
       ge = 2
       catch x
       end
2

julia> ge
ERROR: UndefVarError: `ge` not defined in `Main`
Suggestion: add an appropriate import or assignment. This global was declared but not assigned.

julia> try; function f(); (global ge = 2; ge); end
       ge = 2
       catch x
       end
2

julia> ge
2

I believe the second case is a bug. But perhaps @JeffBezanson could also chime in.

[mlechu]

Ah, __ isn't in value position according to compile-body (we get the desired error when it is: function foo(); 1; __; end).

I believe the second case is a bug.

I thought the different syntax caused it! The IR is identical, but putting the same thing into the REPL twice assigns to the global, I guess by thinking the first run contains a declaration.

[Keno]

Ah, __ isn't in value position according to compile-body (we get the desired error when it is: function foo(); 1; __; end).

If a symbol falls in the forest, but nobody is around to hear it, does it make a sound?

I understand why lowering thinks its not in value position, but I think it should probably still be a lowering error, since otherwise it would be a valid symbol if and only if nobody uses the result, which is weird.