[Bug 25945] New: C_ASSERT doesn't fail if given non-constant expression

http://bugs.winehq.org/show_bug.cgi?id=25945

Summary: C_ASSERT doesn't fail if given non-constant expression
           Product: Wine
           Version: 1.1.22
          Platform: x86
        OS/Version: Linux
            Status: UNCONFIRMED
          Severity: minor
          Priority: P2
         Component: -unknown
        AssignedTo: [e-mail]
        ReportedBy: [e-mail]

In commit wine-1.1.21-63-g5d31eb9 AJ changed the definition of C_ASSERT (in
include/winnt.h) to make it compatible with newer versions of GCC.

Roughly, the change was:

-#define C_ASSERT(e) extern char __C_ASSERT__[(e)?1:-1] __attribute__((unused))
+#define C_ASSERT(e) extern void __C_ASSERT__(int [(e)?1:-1])

Unfortunately, in GCC 4.4.3, the new definition of C_ASSERT doesn't fail when
given a non-constant expression. E.g.:

C_ASSERT(rand() == -1);      /* ignored */

I think this is because GCC (as per C99) is treating the parameter of
__C_ASSERT__ as a VLA (variable length array), and so its type is equivalent to
int[*].

[Bug 25945] C_ASSERT doesn't fail if given non-constant expression

http://bugs.winehq.org/show_bug.cgi?id=25945

Alexander Scott-Johns  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |download, source
                 CC|                            |[e-mail]

--- Comment #1 from Alexander Scott-Johns  2011-01-31 19:07:50 CST ---
We could try something like (based on linux/kernel.h's BUILD_BUG_*):

#define C_ASSERT(e)  (sizeof(struct { signed __static_assert : (e)?1:-1; }))

This will evaluate to sizeof(int), and can be used in constant expressions.
I'm not sure if this version of C_ASSERT can be made into a declaration, as
using it inside a function prototype makes GCC complain: "anonymous struct
declared inside parameter list".

If C99 features are available, this should work (it uses array literals and
array designators):

#define C_ASSERT(e)  \
     extern void __C_ASSERT__(int [sizeof((char[]){ [-!(e)] = 0 })])

The _Static_assert(e,m) thing from C1x is implemented in GCC 4.6. If it's
available, we should use it, instead of using a homebrew solution (which
can then be a hacky, not-accepted-by-modern-GCC thing).

#ifdef _Static_assert  /* There is a way to check for this, right? */
  # define C_ASSERT(e)  _Static_assert((e), #e " was false")
  #elsif defined(_MSC_VER)
  # define C_ASSERT(e)  typedef char __C_ASSERT__[(e)?1:-1]
  #elsif __STDC_VERSION__ >= 199901L
  # define C_ASSERT(e)  \
              extern void __C_ASSERT__(int [sizeof((char[]){ [-!(e)] = 0 })])
  #else  /* C99 not supported: no VLAs */
  # define C_ASSERT(e)  extern void __C_ASSERT__(int [(e)?1:-1])
  #endif

[Bug 25945] C_ASSERT doesn't fail if given non-constant expression

http://bugs.winehq.org/show_bug.cgi?id=25945

--- Comment #2 from Alexander Scott-Johns  2011-02-02 10:34:34 CST ---
According to the GCC documentation, GCC 2.95 supports array designators and
(rvalue) array literals. And according to the c1x draft, <assert.h> should
define a static_assert macro. So this should work:

#if defined(static_assert)
  # define C_ASSERT(e) static_assert((e), #e " should be true")
  #elif defined(_MSC_VER)
  # define C_ASSERT(e) typedef char __C_ASSERT__[(e)?1:-1]
  #else
  # define C_ASSERT(e)  \
               extern void __C_ASSERT__(int [sizeof((char[]){ [-!(e)] = 0 })])
  #endif

[Bug 25945] C_ASSERT doesn't fail if given non-constant expression

http://bugs.winehq.org/show_bug.cgi?id=25945

Alexandre Julliard  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|[e-mail]    |

[Bug 25945] C_ASSERT doesn't fail if given non-constant expression

http://bugs.winehq.org/show_bug.cgi?id=25945

Jerome Leclanche  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

[Bug 25945] C_ASSERT doesn't fail if given non-constant expression

http://bugs.winehq.org/show_bug.cgi?id=25945

André H.  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

--- Comment #3 from André H.  2013-07-01 16:06:54 CDT ---
code wasn't touched since 2009, still true, though i didn't test

[Bug 25945] C_ASSERT doesn't fail if given non-constant expression

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=25945

IvanDSM  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

--- Comment #4 from IvanDSM  ---
The code for C_ASSERT is still the same in Wine 5.0. Is this an actual issue or
can we ignore this?