Use u32 to represent variant tags.

[bitonic]

This makes it possible to exchange binary serialized structures between
different architectures, and reduces bandwidth. I doubt it's even
possible to have more than 4 billions variants anyway :).

[TyOverby]

Right now my code works between architectures because self.read_uint() and self.emit_uint() are producing u64 on both x86 and x86-64. However, I do buy your argument that 32-bits are more than enough. I'll merge this when I figure out what is happening with the Travis-CI build.

[TyOverby]

Actually, what about forcing emit_uint() to fit in a u32 instead of a u64? I'll need to bump the version number because it is a breaking change, but it will make it so that once a message is encoded somewhere, it can be decoded anywhere.

[bitonic]

I hadn't noticed that you use write_be_u64 for uint. I think that's a bad idea, I would use write_uint and leave the responsibility to the user -- code that works with machine dependent types are always going to have different semantics on different machines.

On the other hand, I think that forcing uint to u32 is an even worse idea, because it's unexpected and without any good reason.

The build might be failing because you're depending on an outdated rustc-serialize.

[TyOverby]

The thing is that uint shows up a lot in rust data structures like Vec and HashMap, so it's not really their fault. I could do the checking on the deserialization side as an alternative. If it tries to deserialize a value that is to big to fit in a 32-bit uint, it could fail.

[TyOverby]

Also, yeah, the build was failing because I was using an old version of rustc-serialize and because rustc broke derive(RustcSerialize) unless you have a flag enabled.

[bitonic]

Mhm, right, I guess serializing uint in a machine dependent way could be problematic if a lot of established data structures use it. In any case, I think falling back to u64 is a better choice, if we have to fall back to something. I just checked and the cereal and binary Haskell libraries do the same.

What's certain is that encoding variant tags to u32 is a win :).

[TyOverby]

Thanks for the pull request, I've thought about it for a bit and decided to merge!