Bug 470428

Created attachment 159329 [details]
New crash information added by DrKonqi

DrKonqi auto-attaching complete backtrace.

Cannot reproduce on Manjaro GNU/Linux. My GCC version is: gcc (GCC) 12.2.1 20230201

In my stl_algobase.h I see this code:
```
  template<typename _ForwardIterator, typename _Tp>
    _GLIBCXX20_CONSTEXPR
    inline typename
    __gnu_cxx::__enable_if<__is_scalar<_Tp>::__value, void>::__type
    __fill_a1(_ForwardIterator __first, _ForwardIterator __last,
	      const _Tp& __value)
    {
      const _Tp __tmp = __value;
      for (; __first != __last; ++__first)
	*__first = __tmp;
    }

  // Specialization: for char types we can use memset.
  template<typename _Tp>
    _GLIBCXX20_CONSTEXPR
    inline typename
    __gnu_cxx::__enable_if<__is_byte<_Tp>::__value, void>::__type
    __fill_a1(_Tp* __first, _Tp* __last, const _Tp& __c)
    {
      const _Tp __tmp = __c;
#if __cpp_lib_is_constant_evaluated
      if (std::is_constant_evaluated())
	{
	  for (; __first != __last; ++__first)
	    *__first = __tmp;
	  return;
	}
#endif
      if (const size_t __len = __last - __first)
	__builtin_memset(__first, static_cast<unsigned char>(__tmp), __len);
    }
```

The line at kdevplatform/serialization/itemrepository.h:734:
            std::fill_n(m_nextBucketHash, NextBucketHashSize, 0);
Declaration:
    unsigned short* m_nextBucketHash
Since `unsigned short` is not a byte type (the size is 2 bytes), the loop overload of __fill_a1 should be called. But your backtrace ends with `__memset_avx2_unaligned_erms`, which suggests that the memset overload is called on your system or the compiler optimizes the loop into a memset instruction. What is your GCC version and the function at stl_algobase.h:930 on your system? Which optimization flags have you built KDevelop with? Is memset-ing a memory-mapped region supported by GCC? If so, maybe there is some memset+memmap bug in your libstdc++ version?

Hi, first of all, after clearing the cache it no longer crashes. But if it helps to prevent this in the future:

The entire system is built with CFLAGS="-march=native -ggdb3 -O2 -pipe -fdiagnostics-color". See attached compiler.info for what -march=native translates to on my system.

This is the content of stl_algobase.h around that line.

      template<typename _ForwardIterator, typename _Tp>
        _GLIBCXX20_CONSTEXPR
        inline typename
        __gnu_cxx::__enable_if<__is_scalar<_Tp>::__value, void>::__type
        __fill_a1(_ForwardIterator __first, _ForwardIterator __last,
                  const _Tp& __value)
        {
          const _Tp __tmp = __value;
930:      for (; __first != __last; ++__first)
            *__first = __tmp;
        }

Created attachment 159346 [details]
compiler.info

(In reply to Andrei Slavoiu from comment #3)
> Hi, first of all, after clearing the cache it no longer crashes.
Did you recently update from a custom-built unreleased KDevelop master version? If yes, perhaps the cache wasn't cleared for some reason. On the other hand, the merge request that introduced the crashing code https://invent.kde.org/kdevelop/kdevelop/-/merge_requests/445 increments KDEV_ITEMREPOSITORY_INCREMENT, which forces clearing the cache. But let us close this bug for now in case this is a cache issue. Reopen if it occurs again.

*** This bug has been marked as a duplicate of bug 488043 ***