Detection of `inplace` for fixed point
jlperla opened this issue · comments
I am finding that the code using the fixedpoint and solve functions look uglier than they need to be when using inplace = false all over the place.
I think this could have an almost trivial change:
For the following:
and if you want to do it for the nlsolve interface:
- https://github.com/JuliaNLSolvers/NLsolve.jl/blob/master/src/nlsolve/nlsolve.jl#L49
- https://github.com/JuliaNLSolvers/NLsolve.jl/blob/master/src/nlsolve/nlsolve.jl#L84
CHange inplace = true
to something like
inplace = !applicable(f, initial_x)
that is, check if it can be called with just the initial condition... if so, then we can use inplace = false
Of course, there is a small chance that there is a method for both f(out, x) and f(x) for the same function (at which point this would choose the f(x) one by default)... but I doubt that would happen in practice. People are very careful to use the ! for modifying functions, so it is pretty rare to have both.
My guess is that checking that the typeof(f) <: Union{InplaceObjective, NotInplaceObjective} might also be important for the nlsolve case.. But if we only do this for fixedpoint that would be fine with me.
Is applicable a Base function?
Yeah
TIL
Opened #192 to incorporate this change (only for fixedpoint() for now). Also, it turns out it's actually just Base.applicable
Closing because the PR #192 has been merged.