MethodError with Base.Math.pow_body on v0.8.4

[david-pl]

I don't have an MRE, but running a Newton method I get

MethodError: no method matching pow_body(::UInt64, ::Float64)
  The function `pow_body` exists, but no method is defined for this combination of argument types.
  
  Closest candidates are:
    pow_body(::Float64, ::Float64)
     @ Base math.jl:1162
    pow_body(::Float64, ::Integer)
     @ Base math.jl:1218
    pow_body(::T, ::T) where T<:Union{Float16, Float32}
     @ Base math.jl:1193

It's this line right here:

QuantumPropagators.jl/src/newton.jl

Line 137 in 64dd955

@inline p = p * Base.Math.pow_body(Δ, exponent)

Judging from the types there, pow_body is being called with the types above.

[goerz]

You're passing some number as an integer that should be a float.

It might be possible to insert a conversion somewhere to fix that automatically, but I'd need a MWE.

[goerz]

My guess is your operator matrices (in the Hamiltonian) have a UInt64 eltype.

You don't want to do that, but again, maybe there's a way to make it work or to give a better error. Not sure. In general, QuantumPropagators is "bring your own Hamiltonian", so in some sense it's up to the user to supply something that fulfills the required interface. On that note: what does QuantumPropagators.Interfaces.check_generator give you?

[david-pl]

But you're explicitly converting the Δ to a UInt here:

QuantumPropagators.jl/src/newton.jl

Line 136 in 64dd955

Δ = reinterpret(UInt64, abs(newpoints[i] - leja[j]))

Given you annotate the exponent as ::Float64 here:

QuantumPropagators.jl/src/newton.jl

Line 127 in 64dd955

exponent::Float64 = 1.0 / (n + n_use)

how could this code path ever not call pow_body(::UInt64, ::Float64)?

I'm happy to try and produce an MWE, but this really seems like a bug in the library to me.
As a side note: without changing any code on my end, but going back to v0.8.3 of QuantumPropagators, everything works fine.

[goerz]

Interesting! I'm going to have a deeper look at that.

[goerz]

It turns out Julia 1.12 broke this. That's what I get for using Julia internals as a performance optimization!

[david-pl]

Oh wow, that must have been tricky to debug. Sorry, I should have added my versioninfo().

Thanks for the quick response and the prompt fix!

[goerz]

Oh wow, that must have been tricky to debug.

Actually, not that much, since this also popped up in CI. The only change having been that Julia 1.12 had been released.

In any case, I've just tagged QuantumPropagators 0.8.5 that includes the fix