sl (step into last call in line) is unreliable

I'm not sure how to make a "MWE" but here is a run-through of the problem: If I run sl on the following it works fine:

[1/1] \(A, B) at /Users/solver/.julia/packages/QuasiArrays/B65Uq/src/matmul.jl:34
>34  @inline \(A::AbstractQuasiArray, B::AbstractQuasiArray) = ldiv(A,B)

  A::Fourier{Float64} = Fourier{Float64}()  (arg)
  B::QuasiArrays.BroadcastQuasiVector{…} = <cos.(Inclusion(DomainSets.RealNumb...>  (arg)

→ ldiv(Fourier{Float64}(), cos.(Inclusion(DomainSets.RealNumbers())))

1|debug> sl

But the following is equivalent to so instead of stepping into the materialize call:

[1/3] ldiv(A, B; kwds) at /Users/solver/.julia/packages/ArrayLayouts/RCHQ8/src/ldiv.jl:98
>98  @inline ldiv(A, B; kwds...) = materialize(Ldiv(A,B); kwds...)

  A::Fourier{Float64} = Fourier{Float64}()  (arg)
  B::QuasiArrays.BroadcastQuasiVector{…} = <cos.(Inclusion(DomainSets.RealNumb...>  (arg)
  kwds::Base.Pairs{…} = Base.Pairs{Symbol, Union{}, Nothing, @NamedTuple{}}()  (arg)

→ (ArrayLayouts.Ldiv)(Fourier{Float64}(), cos.(Inclusion(DomainSets.RealNumbers())))
1|debug> sl

Thus it steps out and produces the following:

[1/1] \(A, B) at /Users/solver/.julia/packages/QuasiArrays/B65Uq/src/matmul.jl:34
>34  @inline \(A::AbstractQuasiArray, B::AbstractQuasiArray) = ldiv(A,B)

  A::Fourier{Float64} = Fourier{Float64}()  (arg)
  B::QuasiArrays.BroadcastQuasiVector{…} = <cos.(Inclusion(DomainSets.RealNumb...>  (arg)

→ <return [-7.16273e-18, 1.08148e-18, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, ...>

Reproduce the report with the shown Julia calls and the debugger commands sl and so, starting from the calls at matmul.jl:34 and ldiv.jl:98. Trace how sl handles the materialize call and compare it with stepping out; done means sl reliably enters the last call in the line instead of behaving like so.