[Bug tree-optimization/94335] False positive -Wstringop-overflow warning with -O2

[msebor at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94335

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |diagnostic
                 CC|                            |msebor at gcc dot gnu.org

--- Comment #1 from Martin Sebor <msebor at gcc dot gnu.org> ---
This type of warning is new GCC 10; it was added in the commit below.  It works
as designed here.  It sees the following IL (the memset calls don't do
anything).  The MEM[] = 65; statement is what triggers it.

  <bb 2> [local count: 1073741824]:
  aDummyBuffer_4 = malloc (10);
  it ={v} {CLOBBER};
  if (aDummyBuffer_4 != 0B)
    goto <bb 3>; [70.00%]
  else
    goto <bb 18>; [30.00%]

  <bb 3> [local count: 751619281]:
  _24 = aDummyBuffer_4 - ⁢
  it.d = _24;
  itCopy ={v} {CLOBBER};
  if (_24 != -9223372036854775808)
    goto <bb 4>; [94.29%]
  else
    goto <bb 5>; [5.71%]

  <bb 4> [local count: 708669601]:
  _23 = aDummyBuffer_4 - &itCopy;
  itCopy.d = _23;
  *aDummyBuffer_4 = 65;
  aDummySource_97 = malloc (10);
  D.40357 ={v} {CLOBBER};
  _17 = aDummyBuffer_4 - &D.40357;
  D.40357.d = _17;
  goto <bb 6>; [100.00%]

  <bb 5> [local count: 365072224]:
  itCopy.d = -424242;
  MEM[(char *)&itCopy + -424242B] = 65;   <<< warning here
  aDummySource_105 = malloc (10);
  D.40357 ={v} {CLOBBER};
  D.40357.d = -424242;
  ...
  <bb 18> [local count: 322122544]:
  it.d = -9223372036854775808;
  itCopy ={v} {CLOBBER};
  goto <bb 5>; [100.00%]


It doesn't matter (much) whether the initial address is or can be null (the
warning persists even with operator new that doesn't return null or when the
ctor never does set d to  kEmptyPointer).  The branch of the code that sets d
to -424242 isn't eliminated because the pointer subtraction in either ctor
could, as far as GCC can tell, result in the same value as kEmptyPointer.

Asserting that the subtraction doesn't result in such a value, for instance
like so:
            if (d == kEmptyPointer) __builtin_unreachable ();
and also guaranteeing that the initial address isn't null (e.g., by using
operator new) eliminates the warning.

Short of teaching GCC that the magnitude of the difference between any two
pointers must be less than PTRDIFF_MAX I don't think there's anything that can
be done do improve things (either codegen, or avoid the warning in this case). 
 I'll leave this report unresolved in case someone feels otherwise.

commit b631bdb3c16e85f35d38e39b3d315c35e4a5747c
Author: Martin Sebor <msebor@redhat.com>
Date:   Thu Jul 25 00:29:17 2019 +0000

    PR tree-optimization/91183 - strlen of a strcpy result with a conditional
source not folded

    PR tree-optimization/91183 - strlen of a strcpy result with a conditional
source not folded
    PR tree-optimization/86688 - missing -Wstringop-overflow using a non-string
local array in strnlen with excessive bound