[Bug c++/53330] New: new() operator can return NULL on a zero-length allocation

[Bug c++/53330] New: new() operator can return NULL on a zero-length allocation

[kilobyte at angband dot pl]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53330

             Bug #: 53330
           Summary: new() operator can return NULL on a zero-length
                    allocation
    Classification: Unclassified
           Product: gcc
           Version: 4.7.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: kilobyte@angband.pl


Created attachment 27386
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=27386
test case

While in general C++ disallows zero-length arrays, they are explicitly allowed
by the new() operator (C++ 3.7.4.1.2), with a guarantee that such an allocation
will always return an unique non-null pointer.

This worked correctly in 4.6 and before (and clang, and MSVC, ...), 4.7.0
(Debian 4.7.0-8) and trunk@187013 return null if elements of the array have a
constructor and have sizeof() > 0 themselves.  For simple types or structs, all
is ok.

Also, if there's a constructor (no regards for sizeof(element)) and the array
length is known at compile time, -Wuninitialized returns incorrect diagnostics
that the returned value is uninitialized.

[Bug c++/53330] new() operator can return NULL on a zero-length allocation

[kilobyte at angband dot pl]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53330

--- Comment #1 from Adam Borowski <kilobyte at angband dot pl> 2012-05-12 11:01:23 UTC ---
Correction: after testing with valgrind, the return value is indeed
uninitialized; the pointer in contructor-but-no-fields case happens to be
non-zero but is junk and will cause a crash when you try to free it.

[Bug c++/53330] new() operator can return NULL on a zero-length allocation

[kilobyte at angband dot pl]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53330

--- Comment #2 from Adam Borowski <kilobyte at angband dot pl> 2013-03-29 13:13:21 UTC ---
Created attachment 29750
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=29750
updated testcase

Updated testcase: it checks for invalid pointers (by freeing them), and returns
an exit code (if it won't crash first).

[Bug c++/53330] new() operator can return NULL on a zero-length allocation

[kilobyte at angband dot pl]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53330

--- Comment #3 from Adam Borowski <kilobyte at angband dot pl> 2013-03-29 13:20:53 UTC ---
Re-tested:
* gcc-4.7.2 works on amd64, armhf, x32, fails on i386
* gcc-4.8.0 works on all of the above
(all Debian)

So it appears to be fixed in 4.8, at least on architectures I tried. 
Regardless of whether you'll fix it in 4.7 or not, it may be worth adding to
the test suite.

[Bug c++/53330] new() operator can return NULL on a zero-length allocation

[paolo at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=53330

--- Comment #4 from paolo at gcc dot gnu.org <paolo at gcc dot gnu.org> ---
Author: paolo
Date: Wed Aug 12 22:38:04 2015
New Revision: 226840

URL: https://gcc.gnu.org/viewcvs?rev=226840&root=gcc&view=rev
Log:
2015-08-12  Paolo Carlini  <paolo.carlini@oracle.com>

        PR c++/53330
        * g++.dg/init/new42.C: New.

Added:
    trunk/gcc/testsuite/g++.dg/init/new42.C
Modified:
    trunk/gcc/testsuite/ChangeLog

[Bug c++/53330] new() operator can return NULL on a zero-length allocation

[paolo.carlini at oracle dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=53330

Paolo Carlini <paolo.carlini at oracle dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
      Known to work|                            |4.9.0, 5.0
         Resolution|---                         |WORKSFORME

--- Comment #5 from Paolo Carlini <paolo.carlini at oracle dot com> ---
I'm adding a testcase and closing the bug.