> On Sat, Mar 30, 2024 at 04:38:11PM +1300, Thomas Munro wrote:
> On Fri, Mar 22, 2024 at 7:15 AM Dmitry Dolgov <9erthalion6(at)gmail(dot)com> wrote:
> > > For verification, I've modified the deform.outblock to call LLVMBuildRet
> > > instead of LLVMBuildRetVoid and this seems to help -- inline and deform
> > > stages are still performed as before, but nothing crashes. But of course
> > > it doesn't sound right that inlining pass cannot process such code.
>
> Thanks for investigating and filing the issue. It doesn't seem to be
> moving yet. Do you want to share the LLVMBuildRet() workaround?
> Maybe we need to consider shipping something like that in the
> meantime?
Yeah, sorry, I'm a bit baffled about this situation myself. Yesterday
I've opened a one-line PR fix that should address the issue, maybe this
would help. In the meantime I've attached what did work for me as a
workaround -- it essentially just makes the deform function to return
some value. It's ugly, but since call site will ignore that, and it's
only one occasion where LLVMBuildRetVoid is used, maybe it's acceptable.
Give me a moment, I'm going to test this change more (waiting on
rebuilding LLVM, it takes quire a while on my machine :( ), then can
confirm that it works as expected on the latest version.