pp omits parens needed to disambiguate call-with-block sugar

[jruderman]

This is a passing Rust program:

fn plus_one(f: block() -> int) -> int {
  ret f() + 1;
}

fn ret_plus_one() -> fn(block() -> int) -> int {
  ret plus_one;
}

fn main() {
  let z = (ret_plus_one()) {|| 2};
  assert z == 3;
}

The pretty-printer omits the parens around the call in main, so the result of pretty-printing does not compile.

[graydon]

Goodness!

I wonder if it might make sense to just hold on to any parens in the input, as AST nodes. expr_paren(@expr) is likely a harmless no-op at almost every level of the compiler pipeline and it would relieve the pretty printer from having to work out the (sometimes quite complicated) rules regarding the "proper" amount of parenthesization required to reproduce the input unambiguously.

[brson]

Still a problem with do

fn plus_one(f: fn() -> int) -> int {
  ret f() + 1;
}

fn ret_plus_one() -> fn(fn() -> int) -> int {
  ret plus_one;
}

fn main() {
    let z = do (ret_plus_one()) || { 2 };
    assert z == 3;
}

[catamorphism]

@brson 's example doesn't parse at all now. I filed #3861