GCC 12.1 and LTO yields buffer overflow warnings

[mattgodbolt]

With LTO and optimization, and with some datetime functions referenced across TUs, GCC 12.1 errors about std::move() moving things into temporary buffers:

In function 'swap',
    inlined from 'iter_swap' at /opt/compiler-explorer/gcc-12.1.0/include/c++/12.1.0/bits/stl_algobase.h:182:11,
    inlined from '__reverse' at /opt/compiler-explorer/gcc-12.1.0/include/c++/12.1.0/bits/stl_algo.h:1107:18,
    inlined from 'reverse' at /opt/compiler-explorer/gcc-12.1.0/include/c++/12.1.0/bits/stl_algo.h:1134:21,
    inlined from 'read' at /opt/compiler-explorer/libs/date/v3.0.1/include/date/date.h:6513:21,
    inlined from 'read.constprop' at /opt/compiler-explorer/libs/date/v3.0.1/include/date/date.h:6474:9:
/opt/compiler-explorer/gcc-12.1.0/include/c++/12.1.0/bits/move.h:205:11: warning: writing 16 bytes into a region of size 11 [-Wstringop-overflow=]
  205 |       __a = _GLIBCXX_MOVE(__b);
      |           ^
/opt/compiler-explorer/libs/date/v3.0.1/include/date/date.h: In function 'read.constprop':
/opt/compiler-explorer/libs/date/v3.0.1/include/date/date.h:6506:15: note: destination object 'buf' of size 11
 6506 |         CharT buf[std::numeric_limits<unsigned>::digits10+2u] = {};
      |               ^

This took a while to repro, but:https://godbolt.org/z/e75YG6ano and could easily be a compiler bug, but I figured I'd post here first of all. This code has been working fine for years on earlier compilers (not that that's a guarantee of correctness!).

[HowardHinnant]

Thanks for the heads up.

If you are able to edit date.h, can you try changing:

        CharT buf[std::numeric_limits<unsigned>::digits10+2u] = {};
        auto e = buf;

to:

        CharT buf[std::numeric_limits<unsigned>::digits10+2u] = {};
        CharT* e = buf;

?

(just the second line changes above).

I'd try this experiment myself, but I couldn't figure out how to edit date.h on compiler explorer, and I'm so far unable to duplicate on wandbox. Though I did note that on an older clang, with auto e = , the type of e comes out to CharT*.

[mattgodbolt]

Thanks! Ironically I have as much difficulty editing date.h as you, both on CE and on the one place I use it at work (which is where I discovered the issue). Will give it a go!

[mattgodbolt]

This heinous link: https://godbolt.org/z/hPaqz7TY1 shows on line 6514 (!) of the date.h inlined I have modified to CharT *e = buf; but the error persists, unfortunately.

[HowardHinnant]

Thanks for your investigation. At this point I'm tempted to blame it on compiler error.

[mattgodbolt]

Sounds good to me! I'll file it, closing this!

[mattgodbolt]

The gcc bug I filed: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106020