Strange warning in channel.cpp

[dirk-zimoch]

When compiling p2pApp/channel.cpp with gcc versions 12 to 14 using option -std=c++20 or -std=c++2a (but not with -std=c++17 or lower or with gcc versions 11 or 10, which is the lowest that supports -std=c++20) I get this confusing warning:

In file included from /opt/rh/gcc-toolset-12/root/usr/include/c++/12/string:50,
                 from /opt/rh/gcc-toolset-12/root/usr/include/c++/12/stdexcept:39,
                 from /home/zimoch/epics/epics-base-7.0/include/epicsTime.h:31,
                 from /home/zimoch/epics/epics-base-7.0/include/epicsTimer.h:20,
                 from ../channel.cpp:4:
In function ‘constexpr decltype (__comp((* __first1), (* __first2))) std::lexicographical_compare_three_way(_InputIter1, _InputIter1, _InputIter2, _InputIter2, _Comp) [with _InputIter1 = __gnu_cxx::__normal_iterator<const unsigned char*, vector<unsigned char> >; _InputIter2 = __gnu_cxx::__normal_iterator<const unsigned char*, vector<unsigned char> >; _Comp = __detail::_Synth3way]’,
    inlined from ‘constexpr std::__detail::__synth3way_t<_Iterator> std::operator<=>(const vector<_Tp, _Alloc>&, const vector<_Tp, _Alloc>&) [with _Tp = unsigned char; _Alloc = allocator<unsigned char>]’ at /opt/rh/gcc-toolset-12/root/usr/include/c++/12/bits/stl_vector.h:2056:52,
    inlined from ‘constexpr bool std::less<_Tp>::operator()(const _Tp&, const _Tp&) const [with _Tp = std::vector<unsigned char>]’ at /opt/rh/gcc-toolset-12/root/usr/include/c++/12/bits/stl_function.h:408:20,
    inlined from ‘std::pair<std::_Rb_tree_node_base*, std::_Rb_tree_node_base*> std::_Rb_tree<_Key, _Val, _KeyOfValue, _Compare, _Alloc>::_M_get_insert_unique_pos(const key_type&) [with _Key = std::vector<unsigned char>; _Val = std::pair<const std::vector<unsigned char>, std::weak_ptr<MonitorCacheEntry> >; _KeyOfValue = std::_Select1st<std::pair<const std::vector<unsigned char>, std::weak_ptr<MonitorCacheEntry> > >; _Compare = std::less<std::vector<unsigned char> >; _Alloc = std::allocator<std::pair<const std::vector<unsigned char>, std::weak_ptr<MonitorCacheEntry> > >]’ at /opt/rh/gcc-toolset-12/root/usr/include/c++/12/bits/stl_tree.h:2117:35:
/opt/rh/gcc-toolset-12/root/usr/include/c++/12/bits/stl_algobase.h:1846:39: warning: ‘int __builtin_memcmp(const void*, const void*, long unsigned int)’ specified bound [9223372036854775808, 18446744073709551615] exceeds maximum object size 9223372036854775807 [-Wstringop-overread]
 1846 |                     = __builtin_memcmp(&*__first1, &*__first2, __blen) <=> 0;
      |                       ~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Strangely, the warning does not reference the line in channel.cpp that causes the warning. But I pinned it down to:

pva2pva/p2pApp/channel.cpp

Line 182 in 8fa2313

ment->weakref = ment;

(Pinned down meaning: The warning goes away when I comment out that line.)

May actually be a compiler problem (false positive).

Any ideas?

[mdavidsaver]

From the mentions of std::less, _Rb_tree, and _M_get_insert_unique_pos. I think the warning comes from the previous line.

pva2pva/p2pApp/channel.cpp

Line 181 in 8fa2313

entry->mon_entries[ser] = ment; // ref. wrapped

Specifically, the complaint seems to be when comparing the key during tree insertion. ser is a std::vector. mon_entires de-duplicates by matching the pvRequest in serialized form.

... specified bound [0x8000000000000000, 0xffffffffffffffff] exceeds maximum object size 0x7fffffffffffffff ...

Substituting the hex values makes this look like int64_t vs. uint64_t. I do not know where this would come into play. __builtin_memcmp does return int, but that should be int32_t here. It might be worth looking at /opt/rh/gcc-toolset-12/root/usr/include/c++/12/bits/stl_algobase.h:1846 to see if there is any #if __cplusplus nearby.

Regardless of whether this is real or false positive, no real pvRequest blob will be that big. I'm not even sure if PVA would transport one. So I do not think this issue would have practical effect.

[dirk-zimoch]

There is #if __cpp_lib_three_way_comparison but no #if __cplusplus check.
See https://en.cppreference.com/w/cpp/feature_test.html

[dirk-zimoch]

So I do not think this issue would have practical effect.

I am aiming for -Werror without exceptions.

[mdavidsaver]

__cpp_lib_three_way_comparison

This is one of the c++20 feature test macros.

I am aiming for -Werror without exceptions.

It looks like you will need to mask this one somehow, it is an unresolved compiler/stdlib bug.

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114622

[dirk-zimoch]

That bug report is about compiling with -O0. This compilation was with -O3.

[mdavidsaver]

That bug report is about compiling with -O0. This compilation was with -O3.

By my reading of 114622, -O0 is only about reproducibility. That issue could happen at high optimization levels.

[dirk-zimoch]

It looks like you will need to mask this one somehow

I tried this

#if __GNUC__ > 11 && __cpp_lib_three_way_comparison
// silence strange false positive (?) gcc warning in c++20 mode
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wstringop-overread"
#endif

but it did not work. I still get the warning. The compiler branch is taken, proven by an added #warning.