fix primal return type methoderror

[oscardssmith]

mostly fixes #2733 (but hits the following error because I'm metaprogramming badly somehow)

ERROR: UndefVarError: `ft` not defined in `Enzyme.Compiler`
Suggestion: check for spelling errors or missing imports.
Stacktrace:
 [1] primal_return_type(mode::ForwardMode{false, FFIABI, false, false, false}, ft::Type{typeof(identity)}, tt::Type{Tuple{}})
   @ Enzyme.Compiler ~/.julia/dev/Enzyme/src/typeutils/inference.jl:139
 [2] autodiff(mode::CMode, f::FA, args::Vararg{Annotation, Nargs}) where {FA<:Annotation, CMode<:Mode, Nargs}
   @ Enzyme ~/.julia/dev/Enzyme/src/Enzyme.jl:556
 [3] autodiff(mode::CMode, f::F, args::Vararg{Annotation, Nargs}) where {F, CMode<:Mode, Nargs}
   @ Enzyme ~/.julia/dev/Enzyme/src/Enzyme.jl:534
 [4] top-level scope
   @ REPL[3]:1

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 50.87%. Comparing base (986bbf4) to head (2e0e7e2).

❗ There is a different number of reports uploaded between BASE (986bbf4) and HEAD (2e0e7e2). Click for more details.

HEAD has 13 uploads less than BASE

Flag	BASE (986bbf4)	HEAD (2e0e7e2)
	18	5

Additional details and impacted files

@@             Coverage Diff             @@
##             main    #2768       +/-   ##
===========================================
- Coverage   68.91%   50.87%   -18.05%     
===========================================
  Files          58       13       -45     
  Lines       19861     1256    -18605     
===========================================
- Hits        13688      639    -13049     
+ Misses       6173      617     -5556     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[vchuravy]

This seems wrong to me, do see #2663 for the change that likely broke the behavior here.

In particular: https://github.com/EnzymeAD/Enzyme.jl/pull/2663/files#diff-a5ddd0c06b3228d600577533c56e26161c6952c6b5c89b4befb5fc6b66ffce45L104-L117

[vchuravy]

No we used stub successfully here before.

[github-actions]

Your PR requires formatting changes to meet the project's style guidelines.
Please consider running Runic (git runic main) to apply these changes.

Click here to view the suggested changes.

diff --git a/src/typeutils/inference.jl b/src/typeutils/inference.jl
index d660fd13..c7259a2f 100644
--- a/src/typeutils/inference.jl
+++ b/src/typeutils/inference.jl
@@ -111,8 +111,10 @@ function primal_return_type_generator(world::UInt, source, self, @nospecialize(m
 
     if mi === nothing
         result = Union{}
-        code = Any[Expr(:call, Base.throw, :(MethodError(ft, tt))),
-        Core.Compiler.ReturnNode(result)]
+        code = Any[
+            Expr(:call, Base.throw, :(MethodError(ft, tt))),
+            Core.Compiler.ReturnNode(result),
+        ]
         # create an empty CodeInfo to return the result
         slotnames = Core.svec(Symbol("#self#"), :ft, :tt)
         ci = create_fresh_codeinfo(primal_return_type, source, world, slotnames, code)

[vchuravy]

We should still use GeneratedFunctionStub

[oscardssmith]

The thing that doesn't make sense to me here with the old code is that you were stubbing primal_return_type when you want to just throw a MethodError? Also, why is GeneratedFunctionStub a thing we want to be using here at all?

[vchuravy]

We are writing a generator for a generated function, so we can't throw a MethodError right here and there. We want to throw a method error from the generated function.

Also, why is GeneratedFunctionStub a thing we want to be using here at all?

Because that's the supported interface?

[vchuravy]

Correct fix in #2771