Awesome, you're an implementastic whirlwind! I'm not a reviewer, but I guess some one will ask how this change deals with nested tuples and the syntax t.2.1 (it was supposed to be rejected if it looks like a float).

I would like to see many more tests, e.g.:

• moving out: check that let x = (box 1,); let y = x.0; let z = x.0; correctly fails to compile
• fat types: check let x: &[int] = &[1, 2, 3]; let z = (x,); z.0 == x; just generally does the right thing/doesn't crash
• borrows:
  • check that let x = (box 1, 2); let r = &x.0; let y = x; fails to compile
  • check that let x = (1, 2); let a = &x.0; let b = &x.0; works
  • check that let mut x = (1, 2); let a = &x.0; let b = &mut x.0; and ... let a = &mut x.0; ... fails
  • check that let mut x = (1, 2); let a = &x.0; let b = &mut x.1; works

I’ve added the tests you suggested, and they all pass. I’ve also removed the code duplication in librustc/middle/trans/expr.rs. I’m not sure about renaming ExprTupField—I suppose that ExprNumField is slightly better at suggesting that it works with tuple structs, but the word ‘tuple’ is in the name ‘tuple struct’ anyway. I’m not really sure it matters, anyway—it’s only inside the compiler.

Rebased. r?