-Wfree-nonheap-object compiler warning at link time

[wrrrzr]

Luanti version

Luanti 5.11.0-dev-e5542e5b0 (Linux)
Using LuaJIT 2.1.1731601260
Built by GCC 14.2
Running on Linux/6.12.8 x86_64
BUILD_TYPE=Release
RUN_IN_PLACE=1
USE_CURL=1
USE_GETTEXT=1
USE_SOUND=1
STATIC_SHAREDIR="."
STATIC_LOCALEDIR="locale"

Operating system and version

Arch linux

Summary

[ 50%] Linking CXX executable ../bin/luanti
In member function ‘deallocate’,
    inlined from ‘deallocate’ at /usr/include/c++/14.2.1/bits/alloc_traits.h:513:23,
    inlined from ‘_M_deallocate’ at /usr/include/c++/14.2.1/bits/stl_vector.h:389:19,
    inlined from ‘__dt_base ’ at /usr/include/c++/14.2.1/bits/stl_vector.h:368:15,
    inlined from ‘__dt_base ’ at /usr/include/c++/14.2.1/bits/stl_vector.h:738:7,
    inlined from ‘runTimeouts’ at /home/me/src/opensource/minetest/src/network/mtp/threads.cpp:270:1:
/usr/include/c++/14.2.1/bits/new_allocator.h:172:33: warning: ‘operator delete’ called on pointer ‘_594’ with nonzero offset [3, 9223372036854775806] [-Wfree-nonheap-object]
  172 |         _GLIBCXX_OPERATOR_DELETE(_GLIBCXX_SIZED_DEALLOC(__p, __n));
      |                                 ^
In member function ‘allocate’,
    inlined from ‘allocate’ at /usr/include/c++/14.2.1/bits/alloc_traits.h:478:28,
    inlined from ‘_M_allocate’ at /usr/include/c++/14.2.1/bits/stl_vector.h:380:33,
    inlined from ‘_M_create_storage’ at /usr/include/c++/14.2.1/bits/stl_vector.h:398:44,
    inlined from ‘__ct ’ at /usr/include/c++/14.2.1/bits/stl_vector.h:334:26,
    inlined from ‘__ct ’ at /usr/include/c++/14.2.1/bits/stl_vector.h:603:61,
    inlined from ‘getPeerIDs’ at /home/me/src/opensource/minetest/src/network/mtp/impl.h:268:10,
    inlined from ‘runTimeouts’ at /home/me/src/opensource/minetest/src/network/mtp/threads.cpp:168:60:
/usr/include/c++/14.2.1/bits/new_allocator.h:151: note: returned from ‘operator new’
  151 |         return static_cast<_Tp*>(_GLIBCXX_OPERATOR_NEW(__n * sizeof(_Tp)));
[100%] Built target luanti
[100%] Built target minetest_alias

Steps to reproduce

build luanti with -DRUN_IN_PLACE=1

[appgurueu]

I recall looking into this warning a while ago. If I'm not mistaken, it's a false positive. I think our code is similar to the example shown in https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115048.

Maybe we should disable this warning entirely by default: It has a history of false positives. I'm yet to see it being useful.

The reason this (now) happens at link time is LTO.