Attributes used with derive have incorrect span information

[sgrif]

Given the following struct:

#[derive(AsChangeset)]
#[table_name = "users"]
struct UserForm {
    id: i32,
    name: String,
}

the input to the custom derive will have incorrect span information for every token in the table_name attribute except for the pound. Every other token will have the span Span { lo: BytePos(0), hi: BytePos(0), ctxt: #0 }.

[dtolnay]

I was able to reproduce this with the following macro on rustc 1.25.0-nightly (56733bc 2018-02-01). The four tokens in the attribute, [...], table_name, =, "users", all have span Span { lo: BytePos(0), hi: BytePos(0), ctxt: #0 }.

src/main.rs

#![feature(proc_macro)]

extern crate repro;
use repro::AsChangeset;

#[derive(AsChangeset)]
#[table_name = "users"]
struct UserForm {
    id: i32,
    name: String,
}

fn main() {}

src/lib.rs

#![feature(proc_macro)]

extern crate proc_macro;
use proc_macro::TokenStream;

#[proc_macro_derive(AsChangeset, attributes(table_name))]
pub fn derive(input: TokenStream) -> TokenStream {
    for tt in input {
        println!("{:#?}", tt);
    }
    TokenStream::empty()
}

[sgrif]

The span of the pound token is wrong as well. An error on that span shows the right source code, but has the wrong file/line numbers displayed.

error: `treat_none_as_null` must be in the form `treat_none_as_null = "true"`
 --> <macro expansion>:1:1
  |
1 | #[changeset_options(treat_none_as_null("true"))]
  | ^

[sgrif]

This has regressed significantly on recent nightlies. All tokens except for the first token have the useless dummy span, and the first token still has the incorrect file and line number.

[alexcrichton]

This is caused by #43081, specifically this block of code

[peterhuene]

I suspect #52162 is a dupe of this issue? I'll close.

[sgrif]

This has continued to regress. In the most recent nightly, not even the pound token (first token) of the attribute has a meaningful span.

[peterhuene]

Indeed, that was the behavior I was observing in my repro (https://github.com/peterhuene/inner-attr-span-repro).

[peterhuene]

I've verified the fix in the latest nightly. Thanks very much!