Macro errors are reported at def, not use

[nikomatsakis]

If you write something like #debug["%u", 10], the type error (unsigned vs signed int) gets reported in core.rs, at the definition of #debug, rather than at the point of use. This is very confusing and makes you think something is fundamentally wrong in the Rust libraries (as in, they don't compile) rather than your own code. I think this ought to be fixed before 0.1 so I associated it with that milestone, but others may disagree.

[brson]

Difficult to solve generally, since each new AST node a syntax extension introduces has to synthesize a reasonable span. All syntax extensions currently suffer from this issue.

See also #1000, #1127, #1387

[brson]

I looked at simplext.rs briefly and couldn't understand what it was doing.

[kevina]

I will try to look into it this week if I can find the time, I want to understand what that code is doing anyway.

[kevina]

Actually using the example given, and the latest Rust compiler (0.10), the problem is closer to #1387.

For an example of this bug try:

fn main() {
    #macro[[#apply[f, [x, ...]], f(x, ...)]];
    fn add(a: int, b: int) -> int { ret a + b; }
    assert (#apply[add, [y, 15]] == 16);
}

Here I know exactly what is going on, the span used is the one for the location of the pattern variable x and not the location of y. Change that too #apply[add, [(1 + y), 15]] and the correct location will be reported. The problem is the AST node for y has two locations associated with it and nether one is necessary the correct choice. What is really needed is a backtrace of sorts so that both locations will be reported, much like is done for C++ templates.

Failing a full backtrace I think the better location is the one for the location of y, I can try to fix that if its important enough, but like I said it will do nothing towards fixing the example given.

[kevina]

Just to clarify, when I try to reproduce the error on the example given using:

fn main() {
    #debug["%u", 10];
}

I get

mac.rs:3:200: 3:201 error: mismatched types: expected `uint` but found `int` (types differ)

which is basically the same as #1387 as there is no column 200 in mac.rs except that I don't get a crash.

[kevina]

This issue should now be fixed in a pull request I will submit shortly.

The actual cause of the bogus location was due to the bug I outlined in an earlier comment. The problem was that a hack to fix Issue #1362 was confusing the manner. I fixed the root cause of #1362 and also fixed this bug.

The actual fix was a fit involved and involved modifying how fold works, but I didn't see any other sensible way.

[brson]

Fixed by kevina