Heikki Linnakangas wrote:
> ITAGAKI Takahiro wrote:
>> Here is a revised version of the pgbench duration patch.
>
> Looking at the Win32 timer implementation, it's a bit different from the
> one we have in src/backend/port/win32/timer.c. The one in timer.c uses a
> separate thread and WaitForSingleObjectEx() to wait, while your
> implementation uses CreateTimerQueue() and CreateTimerQueueTimer().
> Yours seems simpler, so I wonder why the timer.c is different?
Probably because it was written back when we supported NT4, and the
CreateTimerQueue() stuff requires Windows 2000 to work.
> It's not too bad as it is in the patch, but it would be nice to put the
> setitimer() implementation into src/port, and use the same code in the
> backend as well.
I haven't looked at the patch ;-), but the implementation in
backend/port is tied into the signal emulation layer that's also in
backend/port, so I think doing such a move will require moving a lot
more than just the timer code...
//Magnus