Re: getrusage for MinGW

[David Bateman]

John W. Eaton wrote:

>On 21-Mar-2006, David Bateman wrote:
>
>| The getrusage and therefore the cputime function don't work under MinGW
>| as MinGW lacks both the getrusage and times functions. It however has
>| the GetProcessTimes windows API function. Could the following patch be
>| considered to allow MinGW to use at least the user and system times of
>| the getrusage function and therefore have the cputime function work
>| correctly?
>
>| *** ./src/DLD-FUNCTIONS/getrusage.cc.orig2   2006-03-21 16:38:49.486021589 
>+0100
>| --- ./src/DLD-FUNCTIONS/getrusage.cc 2006-03-21 16:39:16.063621581 +0100
>| ***************
>| *** 191,196 ****
>| --- 191,210 ----
>|     tv_tmp.assign ("usec", static_cast<double> (fraction * 1e6 / HZ));
>|     m.assign ("stime", octave_value (tv_tmp));
>|   
>| + #elif defined (__MINGW32__)
>| +   HANDLE hProcess = GetCurrentProcess ();
>| +   FILETIME ftCreation, ftExit, ftUser, ftKernel;
>| +   GetProcessTimes (hProcess, &ftCreation, &ftExit, &ftKernel, &ftUser);
>| + 
>| +   EIGHT_BYTE_INT itmp = *(reinterpret_cast<EIGHT_BYTE_INT *> (&ftUser));
>| +   tv_tmp.assign ("sec", static_cast<double> (itmp / 10000000U));
>| +   tv_tmp.assign ("usec", static_cast<double> (itmp % 10000000U) / 10.);
>| +   m.assign ("utime", octave_value (tv_tmp));
>| + 
>| +   itmp = *(reinterpret_cast<EIGHT_BYTE_INT *> (&ftKernel));
>| +   tv_tmp.assign ("sec", static_cast<double> (itmp / 10000000U));
>| +   tv_tmp.assign ("usec", static_cast<double> (itmp % 10000000U) / 10.);
>| +   m.assign ("stime", octave_value (tv_tmp));
>|   #else
>|   
>|     tv_tmp.assign ("sec", 0);
>
>I would prefer that system-dependent code be place in a function
>somewhere else, like sysdep.cc (so perhaps the other code that is
>already there needs to be hidden away as well).
>  
>
Yes I know. The other alternative is to replace the "#if defined
(__MINGW32__)" with "#if defined (HAVE_GETPROCESSTIMES)" and add a test
for GetProcessTimes to configure.in. I'd prefer this to moving the code
to sysdep.cc as I believe that would make the code less readable. The
reason I used __MINGW32__ was to avoid having to reconfigure for a rapid
patch. If you accept the patch, I'll try and update configure.in tonight..

>Also, is FILETIME always required to be an integer data type?  If so,
>why do you need to cast the pointer value?  Why can't you simply write
>
>  tv_tmp.assign ("sec", static_cast<double> (ftKernel / 10000000U));
>
>?  If FILETIME is not required to be an integer data type (maybe it
>could be defined as a pointer to some array or structure) then casting
>the pointer is likely to cause trouble anyway, isn't it?
>  
>
No this cast is needed, and is suggested on the Microsoft website.
Though they use C style casts. With the code as you suggest I get

getrusage2.cc:204: error: no match for 'operator/' in 'ftUser / 10000000u'

As for the cast to a pointer, this is also suggested by Microsoft and
without it I get an illegal cast.

>Finally, what file that is included by getrusage.cc gives you the
>definitions of HANDLE and FILETIME?
>  
>
Oh, damn!!! Yes I cut and paste between a linux emacs and vmware/windows
emacs, and forgot to copy the #include.. Its in windows.h.

>jwe
>
>  
>


-- 
David Bateman                                address@hidden
Motorola Labs - Paris                        +33 1 69 35 48 04 (Ph) 
Parc Les Algorithmes, Commune de St Aubin    +33 6 72 01 06 33 (Mob) 
91193 Gif-Sur-Yvette FRANCE                  +33 1 69 35 77 01 (Fax) 

The information contained in this communication has been classified as: 

[x] General Business Information 
[ ] Motorola Internal Use Only 
[ ] Motorola Confidential Proprietary