Improve type inference

[topolarity]

This makes the following example compile with --trim:

using FFTW

Base.@ccallable function main() :: Cint
    v = fft([0; 1; 2; 1])
    for elem in v
        println(Core.stdout, elem)
    end
    return 0
end

$ ./fftw
4.0 + 0.0im
-2.0 + 0.0im
0.0 + 0.0im
-2.0 + 0.0im

[codecov]

Codecov Report

Attention: Patch coverage is 40.00000% with 3 lines in your changes missing coverage. Please review.

Project coverage is 70.99%. Comparing base (f888022) to head (55b8143).
Report is 1 commits behind head on master.

Files with missing lines	Patch %	Lines
src/fft.jl	40.00%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #308      +/-   ##
==========================================
- Coverage   73.45%   70.99%   -2.46%     
==========================================
  Files           5        5              
  Lines         535      531       -4     
==========================================
- Hits          393      377      -16     
- Misses        142      154      +12     

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

[stevengj]

Why is inlining needed here?

[topolarity]

The core issue is that foreach in Base does not specialize on its function argument: https://github.com/JuliaLang/julia/blob/a007e807623f0bbb820315b8ce3340bd3d41262b/base/abstractarray.jl#L3213-L3214

This is missing something like foreach(f::F, ...) where F which would cause this to specialize better - otherwise as-is it contains a dynamic dispatch. Arguably that should be fixed upstream, but this improves things for released versions of Julia too