| //@ aux-build:amputate-span.rs |
| //@ run-rustfix |
| //@ edition:2018 |
| //@ compile-flags: --extern amputate_span |
| |
| // This test has been crafted to ensure the following things: |
| // |
| // 1. There's a resolution error that prompts the compiler to suggest |
| // adding a `use` item. |
| // |
| // 2. There are no `use` or `extern crate` items in the source |
| // code. In fact, there is only one item, the `fn main` |
| // declaration. |
| // |
| // 3. The single `fn main` declaration has an attribute attached to it |
| // that just deletes the first token from the given item. |
| // |
| // You need all of these conditions to hold in order to replicate the |
| // scenario that yielded issue 87613, where the compiler's suggestion |
| // looks like: |
| // |
| // ``` |
| // help: consider importing this struct |
| // | |
| // 47 | hey */ async use std::process::Command; |
| // | ++++++++++++++++++++++++++ |
| // ``` |
| // |
| // The first condition is necessary to force the compiler issue a |
| // suggestion. The second condition is necessary to force the |
| // suggestion to be issued at a span associated with the sole |
| // `fn`-item of this crate. The third condition is necessary in order |
| // to yield the weird state where the associated span of the `fn`-item |
| // does not actually cover all of the original source code of the |
| // `fn`-item (which is why we are calling it an "amputated" span |
| // here). |
| // |
| // Note that satisfying conditions 2 and 3 requires the use of the |
| // `--extern` compile flag. |
| // |
| // You might ask yourself: What code would do such a thing? The |
| // answer is: the #[tokio::main] attribute does *exactly* this (as |
| // well as injecting some other code into the `fn main` that it |
| // constructs). |
| |
| #[amputate_span::drop_first_token] |
| /* what the |
| hey */ async fn main() { |
| Command::new("git"); //~ ERROR [E0433] |
| } |
| |
| // (The /* ... */ comment in the above is not part of the original |
| // bug. It is just meant to illustrate one particular facet of the |
| // original non-ideal behavior, where we were transcribing the |
| // trailing comment as part of the emitted suggestion, for better or |
| // for worse.) |
| |
| #[allow(dead_code)] |
| mod inner { |
| #[amputate_span::drop_first_token] |
| /* another interesting |
| case */ async fn foo() { |
| Command::new("git"); //~ ERROR [E0433] |
| } |
| } |