ICE in debug generation with match guard.

[meh]

This is the full project that causes that failure, make sure to use that commit.

The culprit line is here.

Inside a match:

e if display.randr.as_ref().map_or(false, |rr| e == rr.event(xrandr::RRScreenChangeNotify))

Results in:

!dbg attachment points at wrong subprogram for function
!21995 = distinct !DISubprogram(name: "{{closure}}", linkageName: "_ZN12screenruster6locker8{{impl}}5spawn11{{closure}}11{{closure}}E", scope: !21991, file: !352, line: 214, type: !21992, isLocal: true, isDefinition: true, scopeLine: 214, flags: DIFlagPrototyped, isOptimized: false, templateParams: !457, variables: !457)
i1 (i8*, %"locker::display::Extension"*)* @"_ZN12screenruster6locker6Locker5spawn28_$u7b$$u7b$closure$u7d$$u7d$28_$u7b$$u7b$closure$u7d$$u7d$17hb41874d5593ec665E"
  call void @llvm.dbg.declare(metadata %closure.114** %__debuginfo_env_ptr, metadata !59153, metadata !59154), !dbg !59155
!59155 = !DILocation(line: 214, scope: !21990)
!21990 = distinct !DISubprogram(name: "{{closure}}", linkageName: "_ZN12screenruster6locker8{{impl}}5spawn11{{closure}}11{{closure}}E", scope: !21991, file: !352, line: 214, type: !21992, isLocal: true, isDefinition: true, scopeLine: 214, flags: DIFlagPrototyped, isOptimized: false, templateParams: !457, variables: !457)
!21990 = distinct !DISubprogram(name: "{{closure}}", linkageName: "_ZN12screenruster6locker8{{impl}}5spawn11{{closure}}11{{closure}}E", scope: !21991, file: !352, line: 214, type: !21992, isLocal: true, isDefinition: true, scopeLine: 214, flags: DIFlagPrototyped, isOptimized: false, templateParams: !457, variables: !457)
LLVM ERROR: Broken function found, compilation aborted!

If I change that to:

e if e == display.randr.as_ref().map_or(0, |rr| rr.event(xrandr::RRScreenChangeNotify))

Then it compiles.

I tried to come up with a reduced test case but failed, so sorry for the full project as test case.

[nagisa]

Building the project at the specified commit fails with

error: Unable to update https://github.com/meh/x11-rs?branch=dpms#cafb1f92

The branch dpms seems to be deleted for the repository in question and referenced commit does not exist anymore.

I tried using master branch but alas.

[meh]

@nagisa my bad, forgot to cargo update before committing, I made a rustc-bug branch on screenruster, try using that, also updated the OP.

[nagisa]

Minimised:

fn main() {
    match 0 {
        e if (|| { e == 0 })() => unimplemented!(),
        1 => unimplemented!(),
        _ => unimplemented!()
    }
}

On nightly, when compiled with debug info rustc test.rs -g fails with

!dbg attachment points at wrong subprogram for function
!64 = distinct !DISubprogram(name: "{{closure}}", linkageName: "_ZN4test4main11{{closure}}E", scope: !61, file: !47, line: 4, type: !62, isLocal: true, isDefinition: true, scopeLine: 4, flags: DIFlagPrototyped, isOptimized: false, templateParams: !51, variables: !51)
i1 (i8*, i1)* @"_ZN4test4main28_$u7b$$u7b$closure$u7d$$u7d$17h6ca029a2196c88a1E"
  call void @llvm.dbg.declare(metadata %closure** %__debuginfo_env_ptr, metadata !144, metadata !145), !dbg !146
!146 = !DILocation(line: 4, scope: !60)
!60 = distinct !DISubprogram(name: "{{closure}}", linkageName: "_ZN4test4main11{{closure}}E", scope: !61, file: !47, line: 4, type: !62, isLocal: true, isDefinition: true, scopeLine: 4, flags: DIFlagPrototyped, isOptimized: false, templateParams: !51, variables: !51)
!60 = distinct !DISubprogram(name: "{{closure}}", linkageName: "_ZN4test4main11{{closure}}E", scope: !61, file: !47, line: 4, type: !62, isLocal: true, isDefinition: true, scopeLine: 4, flags: DIFlagPrototyped, isOptimized: false, templateParams: !51, variables: !51)
LLVM ERROR: Broken function found, compilation aborted!

The problem seems to be the use of variable bound in match arm (the e) inside a closure that appears in match arm’s if portion.

Works with MIR (-Z orbit).

[nagisa]

cc @michaelwoerister

[meh]

Ah it's a regression, that's why I couldn't come up with a reduced test-case, didn't realize I was using stable to compile the test-case 🐼

[michaelwoerister]

Thanks for the report and the reduced test case. I'll look into it.

[arielb1]

1.11 is now beta

[michaelwoerister]

I was able to reproduce this and it is very strange that there are two DISubprogram debuginfo entries describing the same closure. This should not happen.

[michaelwoerister]

For some reason (which I haven't looked into yet) closure::trans_closure_expr is called twice for the closure in the match guard. I would guess that the error would go away if we prevented closures from being translated more than once. I'll look into it some more tomorrow.