[Bug middle-end/103483] context-sensitive ranges change triggers stringop-overread

[msebor at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103483

Martin Sebor <msebor at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|c++                         |middle-end
           Keywords|                            |missed-optimization
                 CC|                            |msebor at gcc dot gnu.org

--- Comment #3 from Martin Sebor <msebor at gcc dot gnu.org> ---
The warning is only issued at -O1.  It's based on the statement in the IL and
the values or ranges of its arguments.  In this case the IL and the argument
values are below:

=========== BB 6 ============
Imports: _1  d_6(D)  
Exports: _1  d_6(D)  
_1      sizetype [0, 0][2, 15]
    <bb 6> [local count: 362271902]:
    if (_1 == 0)
      goto <bb 8>; [100.00%]
    else
      goto <bb 7>; [0.00%]

6->8  (T) _1 :  sizetype [0, 0]
6->7  (F) _1 :  sizetype [2, 15]          >>> _1 > 1

=========== BB 7 ============
    <bb 7> [local count: 346397698]:
    # _2 = PHI <&s.D.26133._M_local_buf(6), _19(3)>
    __builtin_memcpy (_2, &buffer, _1);   <<< -Wstringop-overread

The memcpy() call reads between 2 and 15 bytes from the one-byte buffer.  So
the warning code is working as designed.  The problem is that at -O1 the code
isn't optimized sufficiently to discover that the memcpy call only reads 1 byte
(the function call isn't inlined and so the constant argument isn't propagated
into the call). 

GCC 11 doesn't warn because it's unable to determine the range of the last
memcpy() argument.  Thanks to the Ranger improvements GCC 12 is able to do
better.  In some cases this should improve the S/N ratio of the middle end
diagnostics that depend on ranges.  Unfortunately, in others like this one
where other optimizations are disabled it can make things worse.