On Tue, Apr 20, 2010 at 2:03 PM, Kai Tietz <ktietz70@...> wrote:
> 2010/4/20 Ozkan Sezer <sezeroz@...>:
>> On Tue, Apr 20, 2010 at 12:41 PM, Ozkan Sezer <sezeroz@...> wrote:
>>> On Tue, Apr 20, 2010 at 7:37 AM, Kai Tietz <ktietz70@...> wrote:
>>>> Hi Patrick,
>>>>
>>>> 2010/4/20 Patrick Galbraith <patg@...>:
>>>>> Hi all,
>>>>>
>>>>> I've stumbled across an issue with gettimeofday() returning an incorrect
>>>>> value with mingw-w64. A simple code sample:
>>>>>
>>>>> #include <stdio.h>
>>>>> #include <sys/time.h>
>>>>>
>>>>> int main (void)
>>>>> {
>>>>> struct timeval timer;
>>>>> int retval;
>>>>> retval= gettimeofday(&timer,NULL);
>>>>> fprintf(stderr, "retval: %d\n", retval);
>>>>> fprintf(stderr, "now: %ld\n", time(0));
>>>>> fprintf(stderr, "now: %ld\n", timer.tv_sec);
>>>>> return(0);
>>>>>
>>>>> }
>>>>>
>>>>> When compiled using -
>>>>>
>>>>> regular mingw (correct output)
>>>>>
>>>>> Patrick Galbraith@... ~/dev/stuff
>>>>> $ gcc -o mtime mtime.c
>>>>>
>>>>> Patrick Galbraith@... ~/dev/stuff
>>>>> $ ./mtime.exe
>>>>> retval: 0
>>>>> now: 1271722468
>>>>> now: 1271722468
>>>>>
>>>>> mingw-w64:
>>>>>
>>>>>
>>>>> Patrick Galbraith@... ~/dev/stuff
>>>>> $ /c/mingw/bin/x86_64-w64-mingw32-gcc -o mtime mtime.c
>>>>>
>>>>> Patrick Galbraith@... ~/dev/stuff
>>>>> $ ./mtime.exe
>>>>> retval: 0
>>>>> now: 1271722480
>>>>> now: 31293027
>>>>>
>>>>> Are there any known issues with gettimeofday() with mingw-w64?
>>>>>
>>>>> Thanks!
>>>>>
>>>>> Patrick
>>>>
>>>> what you are seeing here is that time_t type has changed for x64. x64
>>>> doesn't use any longer a 32-bit time_t type. I use 64-bit wide scalar
>>>> instead.
>>>> If you want to have 32-bit time_t the use please the <timefunction>32
>>>> function set and use instead tim32_t type.
>>>>
>>>> Regards,
>>>> Kai
>>>
>>> I can't see the use of a time_t type anywhere in the report?
>>> Is there a misunderstanding somewhere?
>>>
>>> The problem may be due to strict aliasing violations.
>>> A warning doesn't appear with default CFLAGS, but
>>> after adding -Wstrict-aliasing=2, you get
>>>
>>> gettimeofday.c: In function 'getntptimeofday':
>>> gettimeofday.c:42: warning: dereferencing type-punned pointer will
>>> break strict-aliasing rules
>>>
>>> The mingw.org version of gettimeofday.c doesn't exhibit
>>> such an issue. Proper use of a union as it the mingw.org
>>> version might be a solution.
>>>
>>> --
>>> Ozkan
>>>
>>
>> Hmm, changing the FILETIME_1970 macro from 11644473600ull
>> to 116444736000000000ULL in gettimeofday.c as in the mingw.org
>> version seems to make the testcase output the same value.
>>
>> Kai, is this correct? (see attached g3.diff.) Please verify.
>>
>> --
>> Ozkan
>>
>
> yeah, it is ok for apply and looks right.
>
> Thanks,
> Kai
Applied to trunk as rev. 2259, and to the v1.0 branch as rev. 2260.
Patrick, please verify that all your uses work as expected.
--
Ozkan
|