Hi Richard,
On 2026-01-30 15:04, Richard Biener wrote:
On Fri, 30 Jan 2026, Torbjorn SVENSSON wrote:
Hi Richard,
Thanks for the quick reply.
On 2026-01-30 14:28, Richard Biener wrote:
On Fri, 30 Jan 2026, Torbjörn SVENSSON wrote:
Ok for trunk and releases/gcc-15?
--
When building GCC with host=mingw, the following warning can be seen
multiple times in the build log if the MinGW GCC version is older than
r13-4881-g9149a5b7e0a:
In file included from /build/gcc/c/c-typeck.cc:27:0:
/build/gcc/system.h:1172:0: warning: "CONST_CAST2" redefined
#define CONST_CAST2(TOTYPE,FROMTYPE,X) (const_cast<TOTYPE> (X))
In file included from
/usr/lib/gcc/x86_64-w64-mingw32/7.3-win32/include/c++/x86_64-w64-mingw32/bits/gthr.h:148:0,
from
/usr/lib/gcc/x86_64-w64-mingw32/7.3-win32/include/c++/ext/atomicity.h:35,
from
/usr/lib/gcc/x86_64-w64-mingw32/7.3-win32/include/c++/memory:73,
from /build/gcc/gcc/system.h:231,
from /build/gcc/gcc/c/c-typeck.cc:27:
/usr/lib/gcc/x86_64-w64-mingw32/7.3-win32/include/c++/x86_64-w64-mingw32/bits/gthr-default.h:33:0:
note: this is the location of the previous definition
#define CONST_CAST2(TOTYPE,FROMTYPE,X) ((__extension__(union {FROMTYPE
_q; TOTYPE _nq;})(X))._nq)
To ensure that the recent definition is used, always undefine any
potential define before defining the new macro.
I think it's better if we'd remove CONST_CAST2 the few remaining
uses are all from within C++ code?
A simple grep shows the following references in C++ code:
gcc/tree-nested.cc: tree *ref_p = CONST_CAST2 (tree *, const tree *, (const
tree *)e);
gcc/collect2.cc: const char **object_file = CONST_CAST2 (const char **, char
**, object_lst);
gcc/collect2.cc: lto_c_ptr = CONST_CAST2 (const char **, char **,
lto_c_argv);
gcc/collect2.cc: ld1 = CONST_CAST2 (const char **, char **, ld1_argv);
gcc/collect2.cc: ld2 = CONST_CAST2 (const char **, char **, ld2_argv);
gcc/collect2.cc: object = CONST_CAST2 (const char **, char **, object_lst);
gcc/collect2.cc: c_ptr = CONST_CAST2 (const char **, char **, c_argv);
gcc/collect2.cc: object - CONST_CAST2 (const char
**, char **,
gcc/collect2.cc: object = CONST_CAST2 (const char **,
char **,
gcc/collect2.cc: = CONST_CAST2 (const char **, char **, object_lst);
gcc/collect2.cc: const char ** strip_argv = CONST_CAST2 (const char
**, char **,
gcc/collect2.cc: const char **nm_argv = CONST_CAST2 (const char **, char**,
real_nm_argv);
gcc/collect2.cc: const char **ldd_argv = CONST_CAST2 (const char **, char **,
real_ldd_argv);
gcc/collect2.cc: const char ** argv = CONST_CAST2 (const char **, char **,
gcc/config/sparc/driver-sparc.cc: ksp = kstat_lookup (kc, CONST_CAST2
(char *, const char *, "cpu_info"),
gcc/config/sparc/driver-sparc.cc: kstat_data_lookup (ksp, CONST_CAST2
(char *, const char *, "brand"));
gcc/toplev.cc: CONST_CAST2 (const
char **,
For the above files, it should be safe to replace CONST_CAST2() with
CONST_CAST() without any actual code change.
gcc/gcc-ar.cc: CONST_CAST2 (char * const *, const char **, nargv) + 1, f);
gcc/gcc-ar.cc: CONST_CAST2 (char * const *, const char **,
nargv),
gcc/gcc.cc: CONST_CAST2 (char *const *, const char **,
&new_argv[1]),
I'm not sure what would be the outcome if changing these to CONST_CAST()
instead of CONST_CAST2(). Is it safe?
I see
#ifdef __cplusplus
#define CONST_CAST2(TOTYPE,FROMTYPE,X) (const_cast<TOTYPE> (X))
#else
I'd argue no CONST_CAST2 users with !__cplusplus should be around,
so to validate that I'd enable this define unconditionally (and
kill the #else part).
That can be done, but it does not have any impact on the redefinition of the
CONST_CAST2() macro.
The actual cleanup would of course replace CONST_CAST users
with const_cast<TOTYPE> (X), but indeed I only grepped for
CONST_CAST2, not realizing CONST_CAST is defined in terms of
CONST_CAST2.
I cannot assess whether #undef is better than the diagnostic since
if the define is ever different the previous define might be the
correct one? Thus I wonder whether an #undef after the include of
the problematical header would be better? Is there such a convenient
place?
The problem that I see, in my environment, is that I build with a MinGW GCC
v7.3 and that contains:
/* Make sure CONST_CAST2 (origin in system.h) is declared. */
#ifndef CONST_CAST2
#define CONST_CAST2(TOTYPE,FROMTYPE,X) ((__extension__(union {FROMTYPE _q;
TOTYPE _nq;})(X))._nq)
#endif
....
{
if (TlsSetValue (__key, CONST_CAST2(void *, const void *, __ptr)) != 0)
return 0;
As there is no way to do #undef in that header file (already released
toolchain), the only way to avoid the redefinition warning is to do #undef just
before the #define.
It will never be known what version of the CONST_CAST2 macro that is defined in
the toolcahin used to build GCC when the block in system.h is to be evaluated,
so keeping any external definition could cause unclear build/runtime errors.
For example, if I leave the definition from the GCC v7.3 toolchain and do not
replace the definition in my system.h, then I get really complex errors due to
invalid cast usage.
If the warning is kept, I get approx 710 warnings about this when building
r16-6992-gd7e5113e592.
Let me know how you would like me to proceed.
Kind regards,
Torbjörn
Richard.
libgcc/config/i386/gthr-win32.h:/* Make sure CONST_CAST2 (origin in system.h)
is declared. */
libgcc/config/i386/gthr-win32.h:#ifndef CONST_CAST2
libgcc/config/i386/gthr-win32.h:#define CONST_CAST2(TOTYPE,FROMTYPE,X)
(const_cast<TOTYPE> (X))
libgcc/config/i386/gthr-win32.h:#define CONST_CAST2(TOTYPE,FROMTYPE,X)
((__extension__(union {FROMTYPE _q; TOTYPE _nq;})(X))._nq)
libgcc/config/i386/gthr-win32.h: if (TlsSetValue (__key, CONST_CAST2(void *,
const void *, __ptr)))
libgcc/config/pa/gthr-dce.h:/* Make sure CONST_CAST2 (original in system.h) is
defined. */
libgcc/config/pa/gthr-dce.h:#ifndef CONST_CAST2
libgcc/config/pa/gthr-dce.h:#define CONST_CAST2(TOTYPE,FROMTYPE,X)
(const_cast<TOTYPE> (X))
libgcc/config/pa/gthr-dce.h:#define CONST_CAST2(TOTYPE,FROMTYPE,X)
((__extension__(union {FROMTYPE _q; TOTYPE _nq;})(X))._nq)
libgcc/config/pa/gthr-dce.h: (__key, CONST_CAST2(void *, const void *,
__ptr));
I'm not sure if we can remove these references as the gthr file is included
indirectly from system.h (circle if we try to include system.h to get our
definition).
gcc/system.h:#define CONST_CAST2(TOTYPE,FROMTYPE,X) (const_cast<TOTYPE> (X))
gcc/system.h:#define CONST_CAST2(TOTYPE,FROMTYPE,X) ((__extension__(union
{FROMTYPE _q; TOTYPE _nq;})(X))._nq)
gcc/system.h:#define CONST_CAST2(TOTYPE,FROMTYPE,X) \
gcc/system.h:#define CONST_CAST2(TOTYPE,FROMTYPE,X) ((TOTYPE)(FROMTYPE)(X))
gcc/system.h:#define CONST_CAST(TYPE,X) CONST_CAST2 (TYPE, const TYPE, (X))
gcc/tsystem.h:#define CONST_CAST2(TOTYPE,FROMTYPE,X) ((__extension__(union
{FROMTYPE _q; TOTYPE _nq;})(X))._nq)
gcc/tsystem.h:#define CONST_CAST(TYPE,X) CONST_CAST2 (TYPE, const TYPE, (X))
These references could be cleaned up as you said.
gcc/tree.h:(*(CONST_CAST2 (tree *, typeof (T)*, \
gcc/tree.h:(*(CONST_CAST2 (tree*, typeof (T)*, \
For these 2, is it guaranteed that these are only referenced in C++? If it is,
then I suppose they could be replaced with CONST_CAST() instead.
The overall feeling is that there is more risk associated with this change
than the one that I proposed.
I can still send another patch to do the cleanup of the safe transition from
CONST_CAST2() to CONST_CAST().
Kind regards,
Torbjörn
Richard.
gcc/ChangeLog:
* system.h: Avoid redefinition of CONST_CAST2 macro.
Signed-off-by: Torbjörn SVENSSON <[email protected]>
---
gcc/system.h | 1 +
1 file changed, 1 insertion(+)
diff --git a/gcc/system.h b/gcc/system.h
index 588e65453f5..c6f76ae9d65 100644
--- a/gcc/system.h
+++ b/gcc/system.h
@@ -1168,6 +1168,7 @@ extern void fancy_abort (const char *, int, const
char *)
change after the fact). Beyond these uses, most other cases of
using this macro should be viewed with extreme caution. */
+#undef CONST_CAST2
#ifdef __cplusplus
#define CONST_CAST2(TOTYPE,FROMTYPE,X) (const_cast<TOTYPE> (X))
#else
