regression: error[E0277]: the trait bound Entity: AdapterExt is not satisfied

[BoxyUwU]

• https://crater-reports.s3.amazonaws.com/beta-1.88-1/beta-2025-05-26/gh/Buggaboo.flat_sqlite_rs/log.txt

[INFO] [stdout] error[E0277]: the trait bound `Entity: AdapterExt` is not satisfied
[INFO] [stdout]   --> src/lib.rs:42:22
[INFO] [stdout]    |
[INFO] [stdout] 42 |         let entity = t.clone() as Rc<dyn AdapterExt>;
[INFO] [stdout]    |                      ^^^^^^^^^ the trait `AdapterExt` is not implemented for `Entity`
[INFO] [stdout]    |
[INFO] [stdout]    = note: required for the cast from `Rc<Entity>` to `Rc<dyn AdapterExt>`

[theemathas]

I am unable to reproduce this error on my computer.

The code in this crate is doing something extremely weird with macros, and I suspect that the error is nondeterministic.

The relevant code is:

#[derive(LeanBufferWrite)]
pub struct Entity {
    x: String,
    y: i64,
}

include!(concat!(env!("OUT_DIR"), "/Entity_lb_gen.rs"));

The LeanBufferWrite derive macro outputs an empty TokenStream. However, as part of its execution, the code writes to the Entity_lb_gen.rs file in OUT_DIR.

My guess for the cause of the error is: for some reason, the include!() in the above code executes concurrently with the LeanBufferWrite macro. And include!() happened to read the Entity_lb_gen.rs file after it's created, but before it's populated, so the include!() reads an empty file. As a result, the Entity type does not have code that implements the traits that it is supposed to implement.

I am unfortunately unable to verify if this guess is correct.

@rustbot labels +A-macros +A-proc-macros

[theemathas]

I think I have verified the above guess.

I created a workspace containing the following:

build.rs

// Do nothing. This exists so OUT_DIR is set.
fn main() {}

src/lib.rs:

#![allow(dead_code)]

#[derive(bar::MyDerive)]
struct Thing;

include!(concat!(env!("OUT_DIR"), "/data"));

#[test]
fn wut() {
    println!("{}", X);
}

bar/src/lib.rs

use std::{env, fs, path::PathBuf};

use proc_macro::TokenStream;

#[proc_macro_derive(MyDerive)]
pub fn my_derive(_: TokenStream) -> TokenStream {
    let out_dir = env::var_os("OUT_DIR").unwrap();
    let out_file = PathBuf::from(out_dir).join("data");
    let contents = "\n".repeat(1000000) + "const X: i32 = 1;";
    fs::write(out_file, contents).unwrap();
    TokenStream::new()
}

Zipped code of all the files for your convenience: repro.zip

I then repeatedly compiled the code with the following command:

while true; do cargo clean && cargo +beta test --no-run || break; done

If you leave that running for a while, it will eventually produce an error with "cannot find value X in this scope".

The same behavior also occurs when compiling with stable rust instead of beta.

I conclude that the error found by crater isn't a regression. It is actually a nondeterministic error that rarely occurs.

[BoxyUwU]

incredible

[theemathas]

It still kinda bugs me that I can't reliably reproduce this with a sleep() call. And also, the proc macro appears to be called twice with the same OUT_DIR. I'm not going to investigate further though.