trunk broken on Windows

[Vladislav Vaintroub]

Current sysbench does not compile on Windows, due to the use of off_t without appropriate header, C99 round() function, and using "inline" in plain C code. Additionally, if the above is fixed, db_reset_stats() would overwrite already initialized critical sections with all zeros, and this would result in crashes at runtime

[Vladislav Vaintroub (wlad-montyprogram)]

please find fixes in the attachment

[Alexey Kopytov (akopytov)]

Wlad,

Thank you for the patch. One thing is not quite clear to me though:

so sb_timer_update() is defined with 'inline' which appears to be a problem on Windows, where '__inline' must be used instead. But sb_win.h redefines 'inline' to address this problem:

#ifndef __cplusplus
#ifndef inline
#define inline __inline
#endif
#endif

And sb_win.h is included from sb_timer.h which in turn is included from sb_timer.c. So it looks like that should _not_ be a problem after all. Can you clarify please?

[Vladislav Vaintroub (wlad-montyprogram)]

>But sb_win.h redefines 'inline' to address this problem:
Does not seem to be here:
http://bazaar.launchpad.net/~sysbench-developers/sysbench/0.5/view/head:/sysbench/sb_win.h

If I look for "inline" in this file, the only thing I find is a usage of VC++-compatible __inline in

static __inline int usleep(int micros)

[Alexey Kopytov (akopytov)]

On 26.09.11 14:32, Vladislav Vaintroub wrote:
>> But sb_win.h redefines 'inline' to address this problem:
> Does not seem to be here:
> http://bazaar.launchpad.net/~sysbench-developers/sysbench/0.5/view/head:/sysbench/sb_win.h
>
> If I look for "inline" in this file, the only thing I find is a usage
> of VC++-compatible __inline in
>
> static __inline int usleep(int micros)
>

Doh, I was looking at sysbench 0.4:
http://bazaar.launchpad.net/~sysbench-developers/sysbench/0.4/view/head:/sysbench/sb_win.h#L36

So I guess it is sufficient to just port that part to 0.5 to fix the
'inline' problem?

[Vladislav Vaintroub (wlad-montyprogram)]

Yep, this would do the trick