llvm::StructType::getElementType: Assertion `N < NumContainedTys && "Element number out of range!"' failed.

[eddyb]

pub fn maybe_ptr() -> *const u8 {
    static NONE: Option<((), &'static u8)> = None;
    unsafe {
        *(&NONE as *const _ as *const *const u8)
    }
}

Running this in the playpen gives:

rustc: /build/rust-git/src/rust/src/llvm/include/llvm/IR/DerivedTypes.h:285: llvm::Type* llvm::StructType::getElementType(unsigned int) const: Assertion `N < NumContainedTys && "Element number out of range!"' failed.
Aborted (core dumped)

This is actually happening inside LLVM optimizations, in the early-cse pass (-C opt-level=1 still triggers it). Originally found in a memcpy optimized into a load.
Reduced LLVM IR:

target datalayout = "e-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"

@NONE = internal constant { { {}, i8* } } zeroinitializer

define i8* @maybe_ptr() {
entry-block:
  %0 = load i8** bitcast ({ { {}, i8* } }* @NONE to i8**)
  ret i8* %0
}

cc @dotdash

[eddyb]

Current LLVM fork tip: rust-lang/llvm@b820135 on top of llvm-mirror/llvm@0ad623b.

[dotdash]

http://llvm.org/bugs/show_bug.cgi?id=22595

[eddyb]

This needs a test, and there's also two hacks introduced by #21744, that should be reverted now that the LLVM bug is fixed (they both reference this issue).

[dotdash]

@eddyb will you take care of that?

[pczarn]

I'm still getting this failure in #21973 during the compilation of rustc-trans

[dotdash]

I guess you get that in stage0? That will need a new snapshot, so that the
stage0 compiler also uses the new LLVM
Am 26.02.2015 20:40 schrieb "Piotr Czarnecki" [email protected]:

I'm still getting this failure in #21973
#21973

—
Reply to this email directly or view it on GitHub
#21721 (comment).