Perhaps implement a "corrode-ld" and "corrode-ar" for inter-translation-unit conversion? about corrode

[jameysharp]

Yes, I want this. 😄

I've been wondering: can the intermediate form you mention just be Rust source code? It's hardly efficient to have to re-parse the text, but I hope nobody will replace their regular compiler with Corrode, so I don't think we should worry much about runtime efficiency.

And to what extent do we need any cross-module analysis to get the translation to work? We'll definitely need cross-module analysis to make the generated code properly follow Rust idioms, but I'm deferring that problem to a later tool. I'm guessing that with everything marked #[no_mangle], cross-module references to functions and global variables should work fine; and although type definitions will be duplicated in every module that uses them, that should still lead to correct code being generated.

Regarding inline in particular: it's just a performance hint, not a requirement for correctness. We can do some conservative translation that ensures the function is callable everywhere it's supposed to be, without guaranteeing that it will be inlined everywhere GCC would have inlined it, if that makes the translation feasible.

[eternaleye]

The difficulty with inline isn't making sure it's callable; it's making sure that the symbols don't collide when you have multiple definitions, not merely multiple declarations.

Similarly, that implies that Rust would be a poor intermediate form, because it would lose some of those distinctions.

Or in other words, the difference between Rust's #[inline] and C's inline is that C's inline is a requirement for correctness at link time.