Re: include/basetsd.h: Fix int64 to int truncation warnings when compiling with a 64-bit PSDK compiler.
Dmitry,
afair the fix is to disable the int truncation warnings in MSVC. For whatever reason they seem to enable those bogus warnings.
bye michael
On 06/28/2012 12:12 PM, Dmitry Timoshkov wrote:
include/basetsd.h | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-)
diff --git a/include/basetsd.h b/include/basetsd.h index 698dd5c..4a59897 100644 --- a/include/basetsd.h +++ b/include/basetsd.h @@ -153,12 +153,12 @@ typedef unsigned int UHALF_PTR, *PUHALF_PTR;
static inline ULONG32 HandleToULong(const void *h) {
- return (ULONG_PTR)h;
- return (ULONG32)(ULONG_PTR)h;
}
static inline LONG32 HandleToLong(const void *h) {
- return (LONG_PTR)h;
- return (LONG32)(LONG_PTR)h;
}
static inline void *ULongToHandle(ULONG32 ul) @@ -173,32 +173,32 @@ static inline void *LongToHandle(LONG32 l)
static inline ULONG32 PtrToUlong(const void *p) {
- return (ULONG_PTR)p;
- return (ULONG32)(ULONG_PTR)p;
}
static inline LONG32 PtrToLong(const void *p) {
- return (LONG_PTR)p;
- return (LONG32)(LONG_PTR)p;
}
static inline UINT32 PtrToUint(const void *p) {
- return (UINT_PTR)p;
- return (UINT32)(UINT_PTR)p;
}
static inline INT32 PtrToInt(const void *p) {
- return (INT_PTR)p;
- return (INT32)(INT_PTR)p;
}
static inline UINT16 PtrToUshort(const void *p) {
- return (ULONG_PTR)p;
- return (UINT16)(ULONG_PTR)p;
}
static inline INT16 PtrToShort(const void *p) {
- return (LONG_PTR)p;
- return (INT16)(LONG_PTR)p;
}
static inline void *IntToPtr(INT32 i)
Michael Stefaniuc mstefani@redhat.com wrote:
afair the fix is to disable the int truncation warnings in MSVC. For whatever reason they seem to enable those bogus warnings.
The warnings are not bogus, the PSDK compiler also emits a warning when truncating from double to float while gcc keeps silence for instance. Anyway, the patch shouldn't hurt, PSDK headers have similar casts. This patch also helps to pay more attention to warnings: when there are lots of existing ones, it's hard to recognize new and possibly real problems in the code.
On 06/28/2012 01:00 PM, Dmitry Timoshkov wrote:
Michael Stefaniuc mstefani@redhat.com wrote:
afair the fix is to disable the int truncation warnings in MSVC. For whatever reason they seem to enable those bogus warnings.
The warnings are not bogus, the PSDK compiler also emits a warning when
They are bogus as the C standard explicitly allows conversion between integer types.
truncating from double to float while gcc keeps silence for instance.
I never looked at that but I assume the same holds true as above.
Anyway, the patch shouldn't hurt, PSDK headers have similar casts. This
In this specific case it doesn't. Start littering the code with casts to get rid of all those warnings and it will majorly hurt. Unneeded casts are *BAD*; worse then ignoring bogus warnings from a compiler.
patch also helps to pay more attention to warnings: when there are lots of existing ones, it's hard to recognize new and possibly real problems in the code.
bye michael
Michael Stefaniuc mstefani@redhat.com wrote:
truncating from double to float while gcc keeps silence for instance.
I never looked at that but I assume the same holds true as above.
The following thread has some interesting details: http://www.daniweb.com/software-development/c/threads/114052/warning-c4305-t...
Anyway, the patch shouldn't hurt, PSDK headers have similar casts. This
In this specific case it doesn't. Start littering the code with casts to get rid of all those warnings and it will majorly hurt. Unneeded casts are *BAD*; worse then ignoring bogus warnings from a compiler.
Obviosuly I'm not planning to add the casts everywhere, but the inline constructs which specifically targeting the pointer to integer conversions that imply truncation should compile without warnings IMHO.
On 06/28/2012 01:50 PM, Dmitry Timoshkov wrote:
Michael Stefaniuc mstefani@redhat.com wrote:
truncating from double to float while gcc keeps silence for instance.
I never looked at that but I assume the same holds true as above.
The following thread has some interesting details: http://www.daniweb.com/software-development/c/threads/114052/warning-c4305-t...
WOW, that's pretty stupid. "0.1" has no type, it is a numeric literal. Type is inferred from the context and only if that isn't possible it will fall back to the intrinsic type. You can even say "int i = 0.1;" and that is correct C.
gcc -Wall -Wextra does not complain about the examples in the thread you quoted as that is the correct C behavior. MSVC is a C++ compiler beaten into submission to do C. And C++ perverted C.
bye michael
Am Donnerstag, 28. Juni 2012, 20:50:15 schrieb Dmitry Timoshkov:
Obviosuly I'm not planning to add the casts everywhere, but the inline constructs which specifically targeting the pointer to integer conversions that imply truncation should compile without warnings IMHO.
Fwiw, I fixed a number of double-to-float conversion warnings in MSVC in wined3d, d3d8 and d3d9 a while ago, as well as some similar integer issues. I fixed most of those warnings by using the proper data types and explicitly using float literals. I added a handful of casts in places where the API differences forced me to(d3d vs GL clipplanes if I remember correctly).
From my experience with wined3d the majority warnings(~90 %) from msvc made
sense and pointed out potential bugs in the code. The rest were cases like the one Dmitry wants to patch where the API definitions force a conversion with potential data loss.
-
Dmitry Timoshkov -
Michael Stefaniuc -
Stefan Dösinger