Support pod extern_cpp_types

[adetaylor]

Test for #1192

[RReverser]

Thinking about this a bit more: maybe this test should actually be failing, unless the target Rust type implements Copy? IIUC, that's basically what pod! expects?

[RReverser]

(as in, my example from the original issue should probably also fail because I forgot #[derive(Clopy, Copy)] which is the only guarantee that the type is, in fact, trivial)

[RReverser]

Thoughts?

[adetaylor]

Sorry, no time for autocxx at the moment - I will reply next time I get back to the project. (However your mental model of how pod! works is wrong: I need to spend some time documenting exactly what are the expectations are for extern_cpp_type! users in the POD and non-POD case.)

[adetaylor]

I improved the documentation here in #1221.

TL;DR: deriving things in Rust is not material to whether things can be POD or non-POD; it's all about the nature of the pre-existing C++ constructors/destructors, and by the time we get as far as #[derive] there's no way we can add extra C++ assertions.

Overall, this extern_cpp_type facility was intended primarily for one set of autocxx bindings to refer to a type generated by some other autocxx/cxx bindings, i.e. cross-mod links, rather than for folks such as your good self to generate whole new fresh shiny Rust types. I've clarified that too.

[RReverser]

For now added this to my project via { version = "0.23.1", git = "https://github.com/google/autocxx", rev = "bd0efed" } and can confirm this fixes my first immediate issue, thanks.

[RReverser]

Two questions. For example, I have a struct like this:

struct RansacStats {
    size_t refinements = 0;
    size_t iterations = 0;
    size_t num_inliers = 0;
    double inlier_ratio = 0;
    double model_score = std::numeric_limits<double>::max();
};

I generate it with generate! + pod!, and it works great. But:

1. Why does the generated binding have Drop? Probably related to my misunderstanding of what pod! means, but it seems clearly unnecessary in this case?
2. Is it possible to have Default generated for such struct, where all fields actually have default initializers on the C++ side?

[RReverser]

Wait... now I see at least generated fn new for this struct. Not sure what changed between reruns.

[RReverser]

On some classes it still doesn't exist, but that sounds like #1155.