no_mangle causes compilation errors with async-await on armv7-linux-androideabi and aarch64-linux-android targets

[jeswin]

Adding the no_mangle attribute to a function which contains an await results in a compilation error for armv7-linux-androideabi and aarch64-linux-android targets. Removing the attribute fixes the issue (though that's not a solution). There are no problems with x86_64-linux-android and i686-linux-android targets.

#![cfg(target_os = "android")]
#![allow(non_snake_case)]

use tokio::{runtime::Runtime };

#[no_mangle]
pub fn Java_blah_blah_func() -> &'static str {
    Runtime::new().unwrap().block_on(async {
        myfunc().await
    })
}

async fn myfunc() -> &'static str {
    "hello"
}

I expected to see this happen: library should get compiled successfully.

Instead, I get the following error:

Function return type does not match operand type of return inst!
  ret i32 undef
 { i8*, i32 }in function _ZN80_$LT$std..future..GenFuture$LT$T$GT$$u20$as$u20$core..future..future..Future$GT$4poll17h2e6af84f5e832d75E
LLVM ERROR: Broken function found, compilation aborted!

Meta

The error exists in the beta and nightly versions. RUST_BACKTRACE=1 doesn't seem to add additional information.

[jeswin]

Hi @jonas-schievink, Not sure if this is minimal enough, but I've created a repo to demonstrate the issue at https://github.com/jeswin/rust-lang-issue-70098-mcve

Please note that build.sh requires cargo-ndk to build.

[Centril]

@rustbot ping llvm

[rustbot]

Hey LLVM ICE-breakers! This bug has been identified as a good
"LLVM ICE-breaking candidate". In case it's useful, here are some
instructions for tackling these sorts of bugs. Maybe take a look?
Thanks! <3

cc @comex @cuviper @DutchGhost @hanna-kruppe @hdhoang @heyrutvik @JOE1994 @jryans @mmilenko @nagisa @nikic @Noah-Kennedy @SiavoshZarrasvand @spastorino @vertexclique @vgxbj

[Centril]

@rustbot ping cleanup

[rustbot]

Hey Cleanup Crew ICE-breakers! This bug has been identified as a good
"Cleanup ICE-breaking candidate". In case it's useful, here are some
instructions for tackling these sorts of bugs. Maybe take a look?
Thanks! <3

cc @AminArria @chrissimpkins @contrun @DutchGhost @elshize @ethanboxx @h-michael @HallerPatrick @hdhoang @hellow554 @imtsuki @jakevossen5 @kanru @KarlK90 @LeSeulArtichaut @MAdrianMattocks @matheus-consoli @mental32 @nmccarty @Noah-Kennedy @pard68 @PeytonT @pierreN @Redblueflame @RobbieClarken @RobertoSnap @robjtede @SarthakSingh31 @senden9 @shekohex @sinato @spastorino @turboladen @woshilapin @yerke

[jeswin]

If it's of any use, I tested it with rustc 1.39 and got the same error. I'll be happy to help in any way - should you require.

[woshilapin]

Hi @jonas-schievink, Not sure if this is minimal enough, but I've created a repo to demonstrate the issue at jeswin/rust-lang-issue-70098-mcve

Please note that build.sh requires cargo-ndk to build.

Using the repository @jonas-schievink provided, I couldn't reproduce the bug. I'm building on a MacOS Catalina 10.15.3 and I tried Rust stable (rustc 1.42.0 (b8cedc004 2020-03-09)) and Rust beta (rustc 1.43.0-beta.1 (4a71daf1c 2020-03-11)).

[jeswin]

Hi @woshilapin, sorry I forgot to mention - my build environment was Linux 64-bit; Ubuntu 19.10 specifically.

I can try to reproduce the bug on a fresh Linux install.

[jeswin]

@woshilapin Could you also post the llvm and ndk versions you were able to use successfully?

[woshilapin]

@woshilapin Could you also post the llvm and ndk versions you were able to use successfully?

Not sure of what I'll provide but if I remember correctly, version of LLVM would be the same as version of clang, right?

$ clang --version
Apple clang version 11.0.0 (clang-1100.0.33.16)
Target: x86_64-apple-darwin19.3.0
Thread model: posix

For NDK, I downloaded it testing the current issue so it's pretty recent. I have a source.properties file in the NDK folder that seems to give this information.

$ cat source.properties
Pkg.Desc = Android NDK
Pkg.Revision = 21.0.6113669

Please tell me if some other information is needed (and how to get them maybe since I'm not an Android developer so I have no background!).

[jeswin]

@woshilapin I was able to reproduce the issue on a fresh install of Ubuntu 19.10 64-bit. My ndk version is the same as yours - 21.0.6113669.

I think ndk uses (at least on Linux) clang and llvm bundled and prebuilt with the ndk. For me, it gets installed at ~/Android/Sdk/ndk/21.0.6113669/toolchains/llvm/prebuilt/linux-x86_64/bin.

The clang --version (from the directory above) is

Android (5900059 based on r365631c) clang version 9.0.8 (https://android.googlesource.com/toolchain/llvm-project 207d7abc1a2abf3ef8d4301736d6a7ebc224a290) (based on LLVM 9.0.8svn)
Target: x86_64-unknown-linux-gnu
Thread model: posix

So there seems to be a problem with compilation on Linux. I'll proceed with testing it on my ancient Macbook.

[jeswin]

@woshilapin I was able to compile it successfully on OSX. So the problem occurs only on Linux.

[lights0123]

I'm experiencing this same issue with a custom, no_std target:

{
    "arch": "arm",
    "cpu": "arm926ej-s",
    "data-layout": "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64",
    "llvm-target": "armv5te-none-eabi",
    "os": "newlib",
    "relocation-model": "pic",
    "target-endian": "little",
    "target-env": "newlib",
    "target-pointer-width": "32",
    "target-c-int-width": "32",
    "linker": "nspire-gcc",
    "linker-flavor": "gcc",
    "dynamic-linking": false,
    "executables": true,
    "features": "+v5te,+soft-float,+strict-align",
    "singlethread": true
}

Running Arch Linux 5.5.10-arch1-1 and rustc 1.44.0-nightly (7520894 2020-03-27).

Here's a sample source:

#[no_mangle]
fn main() {
	doesnt_work();
	does_work();
}

fn doesnt_work() {
	Executor::new().run(Task::new(do_stuff()));
}

fn does_work() {
	Executor::new().run(Task::new(do_more_stuff()));
}

async fn do_stuff() -> &'static str {
	do_more_stuff().await
}

async fn do_more_stuff() -> &'static str {
	"test"
}

(I'll post the code of the custom types if needed, mostly stolen from this recent blog post) Interestingly enough, the function does_work does work and doesn't produce a compiler error when doesnt_work is commented out. I get the following error with doesnt_work:

Function return type does not match operand type of return inst!
  ret i32 undef
 { i8*, i32 }in function _ZN97_$LT$core..future..from_generator..GenFuture$LT$T$GT$$u20$as$u20$core..future..future..Future$GT$4poll17h8f0e6fe9387a4fc1E
LLVM ERROR: Broken function found, compilation aborted!

---

Edit: this error does not occur with this test case when changing to an i32:

async fn do_more_stuff() -> i32 {
	5
}

[nikomatsakis]

@lights0123 the problem in that case might be applying #[no_mangle] to main. Can you try this example?

fn main() {
    main1();
}

#[no_mangle]
fn main1() {
	doesnt_work();
	does_work();
}

fn doesnt_work() {
	Executor::new().run(Task::new(do_stuff()));
}

fn does_work() {
	Executor::new().run(Task::new(do_more_stuff()));
}

async fn do_stuff() -> &'static str {
	do_more_stuff().await
}

async fn do_more_stuff() -> &'static str {
	"test"
}

[lights0123]

@nikomatsakis Same error:

Function return type does not match operand type of return inst!
  ret i32 undef
 { i8*, i32 }in function _ZN97_$LT$core..future..from_generator..GenFuture$LT$T$GT$$u20$as$u20$core..future..future..Future$GT$4poll17h26160800281fdd7cE
LLVM ERROR: Broken function found, compilation aborted!

Just as a note, I won't be able to test running the code like that as I need to export the symbol main, as that's what my target is expecting to execute. I also recently changed the API of my executor to execute a Future directly instead of one wrapped in a Task. I'm going to create a repo that includes both my new and old executor so you can figure out problems.

[lights0123]

@nikomatsakis @jeswin Minimal example with only 1 dependency: https://github.com/lights0123/ice-70098

This doesn't require installing any toolchains other than cargo-xbuild, which is just one cargo install away, but it does require nightly for no_std stuff.

What I learned:

• Doesn't matter what the name of the function is, as long as one is #[no_mangle] or using #[export_name = "whatever"]
• Occurs when storing a Pin<Box<dyn Future<Output = T>>> in a struct, and then panicing. Before, I was able to get it to return the same error even when it didn't unconditionally panic, but I was unable to replicate it with this.
• Occurs when returning a &'static str, but not when returning a &'static i32 referencing a static i32
• Occurs with an async function calling another async function
• Only occurs in release builds

[nikomatsakis]

Huh! Fascinating. Thanks @lights0123, that's good work.