Inference issue in Julia 1.12

[jakobnissen]

This is a duplicate of JuliaLang/JuliaC.jl#88, here reported upstream, as it is likely an inference failure in Julia that has stumped me. It's possible I did something dumb, but I can't figure it out.

To reproduce:

1. Clone https://github.com/jakobnissen/AoC2025
2. Check out commit 206a08a253bf7898d0cda32726ce0a658c7fcac0
3. Outcomment line 181 in AoC2025.jl - the line containing println(Core.stdout, Day6.solve(ImmutableMemoryView(b"abc")))
4. Attempt to compile with JuliaC 0.2.2, Julia 1.12.3. I used command juliac --output-exe aoc2025 --bundle build --trim=safe --experimental .

This will fail, with an unresolved call in AoC2025.solve. Some interesting observations:

• The print statement at AoC2025.jl:181 somehow makes inference work
• Changing the map at AoC2025.jl:184 to a for-loop that pushes to a vector also makes it infer, so maybe the issue is with map
• Could it be a captured variable? Seems not; JET reports no such variables, and renaming solution in the closure makes no difference.
• @code_warntype and JET both claim the relevant functions AoC2025.solve and AoC2025.Day6.solve are type stable and well-inferred. JET claims AoC2025.main "failed to optimize due to recursion", but none of the functions are recursive.

[oscardssmith]

Can you confirm whether this is a regression in 1.12 or a pre-existing issue?

[jakobnissen]

It's difficult to tell, because the issue is lack of JuliaC-compatibility, which can only be checked on Julia 1.12. Both on 1.11 and 1.12, JET's @report_opt reports some issues for AoC2025.main, but they are different, and the "failed to optimize due to recursion" one, which I suspect is the culprit, it 1.12 only