[Bug c++/106974] New: g++ bug caused by r260621

[Bug c++/106974] New: g++ bug caused by r260621

[duan.db at linux dot alibaba.com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106974

            Bug ID: 106974
           Summary: g++ bug caused by r260621
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: duan.db at linux dot alibaba.com
  Target Milestone: ---

hi, I found a g++ bug.

code is shown as below：
https://godbolt.org/z/5Ex564eGE

It works fine on GCC 8, but will Segmentation fault on GCC9, GCC10...
By using bisect, I found the commit which causing this problem.

https://gcc.gnu.org/git/?p=gcc.git&a=commit;h=955da5e5443724cb59f8fbd854c13e78c68bf000

But I'm not familiar with this part of GCC, hope to get your help, thanks.

[Bug c++/106974] g++ bug caused by r260621

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106974

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |marxin at gcc dot gnu.org

--- Comment #1 from Martin Liška <marxin at gcc dot gnu.org> ---
Thanks for the repo, next time, please attach source code directly here w/o any
external service.

[Bug c++/106974] g++ bug caused by r9-595-g955da5e5443724cb

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106974

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
            Summary|g++ bug caused by r260621   |g++ bug caused by
                   |                            |r9-595-g955da5e5443724cb
   Last reconfirmed|                            |2022-09-20
     Ever confirmed|0                           |1
                 CC|                            |jason at gcc dot gnu.org

--- Comment #2 from Martin Liška <marxin at gcc dot gnu.org> ---
$ g++ pr106974.C -fsanitize=undefined && ./a.out
pr106974.C: In instantiation of ‘const NumType&
TestVector<NumType>::operator[](int) const [with NumType = unsigned int]’:
pr106974.C:24:24:   required from here
pr106974.C:16:34: warning: returning reference to temporary
[-Wreturn-local-addr]
   16 |       return shard_index_[0].data[0];
      |              ~~~~~~~~~~~~~~~~~~~~^
pr106974.C:16:36: runtime error: reference binding to null pointer of type
'const unsigned int'
pr106974.C:24:11: runtime error: load of null pointer of type 'const unsigned
int'
Segmentation fault (core dumped)

while clang is fine:

$ clang++ pr106974.C -fsanitize=undefined && ./a.out
0

Really started with r9-595-g955da5e5443724cb.

[Bug c++/106974] g++ bug caused by r9-595-g955da5e5443724cb

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106974

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
I think gcc is correct here. You cannot take the address of an element of a
packed structure.

[Bug c++/106974] g++ bug caused by r9-595-g955da5e5443724cb

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106974

--- Comment #4 from Martin Liška <marxin at gcc dot gnu.org> ---
Code:

#include <vector>
#include <cstdio>

template <typename NumType>
struct __attribute__ ((__packed__)) Shard {
  NumType data[128]; 
};

template<typename NumType>
class TestVector {
 public:
  TestVector() {
      shard_index_.push_back(Shard<NumType>());
  }
  const NumType& operator[](int i) const {
      return shard_index_[0].data[0];
  }
private:
  std::vector<Shard<NumType>> shard_index_;
};

int main(){
    TestVector<unsigned int> vec;
    printf("%u\n",vec[0]);
    return 0;
}

[Bug c++/106974] g++ bug caused by r9-595-g955da5e5443724cb

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106974

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|---                         |INVALID

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
https://gcc.gnu.org/legacy-ml/gcc-patches/2003-07/msg01664.html

This was an expected change really with that patch.