llvm dev - Feb 2018 - Bug in use-list order serialization

Hello,

I want to ask if incorrect serialization of a constant's use-list order
is considered as a bug.

While compiling some programs with

clang -c -emit-llvm -O2 -g

I noticed that for many of the bitcode files the verify-uselistorder
tool terminates with output

*** verify-uselistorder ***
verify bitcode
LLVM ERROR: use-list order changed

where the mismatch is always for the constant i64 0. This happens both
with version 5.0.1 and 6.0 RC3 of LLVM and clang. I managed to trigger
this behavior with the following reduced test-case:

#include <stdio.h>
int main (int argc, char **argv) {
  struct M { char const *a; char const *b; } const map[] = {
    { "a", "b" },
    { NULL, NULL }
  };
  char const *foo = "foo";
  struct M const *m = map;
  while (m->a)
    m++;
  if (m->b)
    foo = m->b;
  printf("%s", foo);
  return 0;
}

Regards,
Daniel

Yes, this is a bug.  Please file a PR, attach a .ll and/or .bc file, and CC me. 
Bonus points if you bugpoint reduce it first.
> On Feb 26, 2018, at 06:37, Daniel Pfister via llvm-dev <llvm-dev at
lists.llvm.org> wrote:
> 
> Hello,
> 
> I want to ask if incorrect serialization of a constant's use-list order
> is considered as a bug.
> 
> While compiling some programs with
> 
> clang -c -emit-llvm -O2 -g
> 
> I noticed that for many of the bitcode files the verify-uselistorder
> tool terminates with output
> 
> *** verify-uselistorder ***
> verify bitcode
> LLVM ERROR: use-list order changed
> 
> where the mismatch is always for the constant i64 0. This happens both
> with version 5.0.1 and 6.0 RC3 of LLVM and clang. I managed to trigger
> this behavior with the following reduced test-case:
> 
> #include <stdio.h>
> int main (int argc, char **argv) {
>  struct M { char const *a; char const *b; } const map[] = {
>    { "a", "b" },
>    { NULL, NULL }
>  };
>  char const *foo = "foo";
>  struct M const *m = map;
>  while (m->a)
>    m++;
>  if (m->b)
>    foo = m->b;
>  printf("%s", foo);
>  return 0;
> }
> 
> Regards,
> Daniel
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev