Convert the runtime to a crate

[brson]

Converting the runtime to a Rust crate appears to be an interesting and useful thing to do.

Benefits:

• It would have the same linkage rules as most of our libraries
• Lets us exercise bindgen
• It gives us the opportunity to use rustdoc and other tools with it
• It would either simplify the makefiles or not make them worse

Cons:

• Possible porting complications (I haven't considered it)

Guide to success:

• Create a bindgen pass (Add a bindgen pass #2124)
• Add src/rt/rt.rc with a directive to auto-generate runtime bindings
• Statically link the rt crate to the native runtime library
• (Optional) Have rustc build the runtime (rustc should support compiling C code #1862)
• Change the snapshot rules

[brson]

This would end up meaning that you can't use rust without linking to another rust crate. That's a pretty large negative, sadly.

[graydon]

Why is it a negative? Seems ok to me.

[brson]

I don't remember. I think because it makes porting harder since there are no standalone test cases.

[brson]

After attempting to convert the runtime to a crate independent of core I now think it is appropriate (significantly easier) to just merge rt and core. The primary reason is that the runtime crate will have to export the entire runtime interface as Rust functions, and that will depend on Rust libc definitions. As more Rust runtime code is written, more of core will need to migrate to rt.

This will introduce the new limitation that it is not possible to compile rust code without linking to core.

[brson]

This will also mean that no crate except for core can use native runtime functions directly (they will be internal to core). We currently have a number of test cases that use the runtime for testing native functions.

[brson]

We're not doing this. Instead merging rt/core.