We now prefer to use in, out and ref modifiers on parameters instead of pointers, even on the original interface method definition. The reason is the CLR and Core CLR cannot marshal pointers across processes, but it can marshal these modifiers without a problem. Some interfaces are used across these boundaries, although CsWin32 doesn't know which. So we just err on generating interfaces that can be used in this way.
This does unfortunately mean that some methods that took pointers and had overloads that took Span<T> no longer will, which means more marshaling and more array allocations. This only impacts the less perf-critical (and default) scenario where marshaling is allowed though, so I don't think it will be a big problem.
We now prefer to use
in
,out
andref
modifiers on parameters instead of pointers, even on the original interface method definition. The reason is the CLR and Core CLR cannot marshal pointers across processes, but it can marshal these modifiers without a problem. Some interfaces are used across these boundaries, although CsWin32 doesn't know which. So we just err on generating interfaces that can be used in this way.This does unfortunately mean that some methods that took pointers and had overloads that took
Span<T>
no longer will, which means more marshaling and more array allocations. This only impacts the less perf-critical (and default) scenario where marshaling is allowed though, so I don't think it will be a big problem.Fixes #860