RFC: #[cfg(…)] syntax for size and alignment of FFI types

[mzabaluev]

Add syntax for the #[cfg(...)] attribute to support conditional compilation dependent on size and alignment of primitive types used in FFI.

Rendered

[dgrunwald]

Conditions in conditional compilation should just be Rust expressions, don't create a new mini-language for this purpose.

Instead of adding tons of special cases to #[cfg], we should seriously consider a D-style static if construct.

[mzabaluev]

@dgrunwald: Thanks, I have added your suggestion to the Alternatives section. I'd like to see more design details on that, as it certainly seems a much more extensive change than the one I'm proposing.

[nikomatsakis]

We discussed this RFC in both the compiler and lang subteam meetings, and found a number of drawbacks:

• Overall, this doesn't seem like the right path for handling enums, as it requires too much careful coding, and we'd prefer to support that as a first-class feature.
• Without enums, the remaining motivation seems relatively thin, and there don't seem to be a lot of examples arising of where this kind of "sizeof-int-oriented-switch" would be useful (vs the current approach of selecting based on O/S etc).
• That said, if we were to add this sort of configuration, we'd prefer to generalize cfg in a universal way (i.e., make cfg understand more than boolean equations), rather than adding special-cased support for the sizeof integers etc.

Thoughts?

[nikomatsakis]

Sorry, I should have added: all that said, I can certainly see why it'd be sometimes be much easier to select based on the size of an integer or something like that, which seems like it might cut across target OS / architectures in a convenient way.

Though I'd be a bit nervous when it comes to struct layout, since I think that can often vary by O/S in small ways, even if the architecture is the same. But perhaps this doesn't matter much in practice.

[DemiMarie]

One feature I would like to see is something like D's std.bitmanip.bitfields!, which produces a bitfield with precisely controlled layout (rather than C layout) and does so with convenient syntax. Obviously the syntax will differ, but a similar syntax extension would be useful for things like JIT compilers and assemblers.

[nikomatsakis]

We discussed this again in the compiler team meeting and decided to close this RFC for now, for the reasons cited previously. Thanks very much for the suggestion!