incr.comp.: "change nothing" test fails.

[lseugene]

I've been slightly modified incremental/struct_change_nothing.rs test with impl block added. And now it fails.

// Copyright 2014 The Rust Project Developers. See the COPYRIGHT
// file at the top-level directory of this distribution and at
// http://rust-lang.org/COPYRIGHT.
//
// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
// option. This file may not be copied, modified, or distributed
// except according to those terms.

// Test incremental compilation tracking where we change nothing
// in between revisions (hashing should be stable).

// revisions:rpass1 rpass2
// compile-flags: -Z query-dep-graph

#![feature(rustc_attrs)]
#![allow(dead_code)]
#![allow(non_snake_case)]
#![allow(unused_variables)]
#![rustc_partition_reused(module="struct_change_nothing", cfg="rpass2")]

#[cfg(rpass1)]
pub struct X {
    pub x: u32,
}

#[cfg(rpass2)]
pub struct X {
    pub x: u32,
}

#[cfg(rpass1)]
impl X {
    fn get(&self) -> u32 {
        self.x
    }
    pub fn dbl(&self) -> u32 {
        self.get() + self.get()
    }
}

#[cfg(rpass2)]
impl X {
    fn get(&self) -> u32 {
        self.x
    }
    pub fn dbl(&self) -> u32 {
        self.get() + self.get()
    }
}

pub struct EmbedX {
    x: X,
}

pub struct Y {
    pub y: char,
}

#[rustc_clean(label="TypeckTables", cfg="rpass2")]
pub fn use_X() -> u32 {
    let x: X = X { x: 42 };
    x.x as u32
}

#[rustc_clean(label="TypeckTables", cfg="rpass2")]
pub fn use_EmbedX(x: EmbedX) -> u32 {
    let x: X = X { x: 42 };
    x.x as u32
}

#[rustc_clean(label="TypeckTables", cfg="rpass2")]
pub fn use_Y() {
    let x: Y = Y { y: 'c' };
}

pub fn main() {}

pass2 output:

$ rustc -Zquery-dep-graph -Zincremental=inc -Zincremental-info src/struct_change_nothing.rs --cfg=rpass2
incremental: session directory: 4 files hard-linked
incremental: session directory: 0 files copied
incremental: module WorkProductId("struct_change_nothing") is dirty because HirBody("struct_change_nothing/8cd878b::{{impl}}[0]::dbl[0]") changed or was removed
error: expected module named `struct_change_nothing` to be Reused but is Translated
  --> src/struct_change_nothing.rs:21:1
   |
21 | #![rustc_partition_reused(module="struct_change_nothing", cfg="rpass2")]
   | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

incremental: 23693 nodes in dep-graph
incremental: 39644 edges in dep-graph
incremental: 1080 nodes in reduced dep-graph
incremental: 3646 edges in serialized dep-graph
incremental: 686 hashes in serialized dep-graph
incremental: re-using 0 out of 1 modules
error: aborting due to previous error(s)

$ rustc --version
rustc 1.19.0-nightly (76242aebb 2017-06-06)
$ cargo --version
cargo 0.20.0-nightly (38ca9b702 2017-05-14)

[MageSlayer]

@nikomatsakis
Can it be related somehow to #37121?

[nikomatsakis]

It might be because the spans changed -- and the + operation may panic (on overflow). In other words, something did change in between the two copies of the impl.

[MageSlayer]

@nikomatsakis
As the original author does not respond, may I ask - what do mean by saying "something did change"?

Personally, I fail to see any differences between two "impl X".
Or maybe you mean something internal has changed? Due to changed line numbers, invisible formatting, etc?

[nikomatsakis]

@MageSlayer

Due to changed line numbers, invisible formatting, etc?

Yes, precisely. The code we have to generate has changed, since the line numbers are different. This isn't always true, but it is true for things that may panic (e.g., +).

[steveklabnik]

Triage: i was going to try to reproduce this, but ran into #57479

The file is still living in the same place, at least. That said, it seems from the above discussion that maybe this is intentional?

[cjgillot]

Closing as intended behaviour: the generated code must include the span of the panic, which indeed changes.