On Mon, 2006-05-29 at 09:34 +1000, James Peach wrote: > On Mon, 2006-05-29 at 08:47 +1000, James Peach wrote: > > On 5/29/06, Volker Lendecke <[EMAIL PROTECTED]> wrote: > > > On Sun, May 28, 2006 at 08:34:57PM +0200, Thomas Bork wrote: > > > > Please accept this patch for configure.in, removing the function > > > > introduced in 15508 (Use clock_gettime for profiling timestamps if it is > > > > available. Use the fastest clock available on uniprocessors.): > > > > > > Hmmm. Fully removing it might not be really the right thing. > > > It might however be good to only link in -lrt (and thus the > > > depending pthreads) if --with-profiling-data=yes. > > > > > > James, what do you think? > > > > Yeh, that seems like the right thing to do. > > OK, now I've *actually* thought about this .... we can't just detect > clock_gettime if profiling is enabled, because on Linux, that would let > the user fall into a situation where they had profiling but broken > oplocks. We should never allow oplocks to be broken. > > Unfortunately, I think that this means that we need a new configure > macro to test whether linking with librt implicitly links libpthread. If > it does, we can disable clock_gettime without disabling > --with-profiling-data.
This was way more horrible than I expected, but the attached patch implements the checks I describe above. -- James Peach | [EMAIL PROTECTED]
-- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba