Milter-greylist

On Wed, Sep 16, 2009 at 04:36:10AM +0200, manu@... wrote:
> Petar Bogdanovic <petar@...> wrote:
> 
> >    Until -r1.217 of milter-greylist.c, dump_sleepflag was initialized
> >    through pthread_cond_init(), called by dump_init().  That's why I
> >    didn't notice earlier.  dump_init() is in store.c now.
> 
> But with your patch you now initiliaze ir twice. Could'nt that cause
> problems?

From what I've seen, no.  But OTOH I don't know much about libpthread.

But even if you want to do it right (through mg_init() which contains
dump_init() and is in store.c):  The problem is that you have a
chicken/egg situation here since conf_load() needs mg_init() (because of
dump_init()) and mg_init() needs conf_load() (because of the conf
struct).

I would suggest dropping pthread_cond_init() which is the only thing
dump_init() actually does.  man pthread_cond_init() says:

	``In cases where default condition variable attributes are
	  appropriate, the macro PTHREAD_COND_INITIALIZER can be used to
	  initialise condition variables that are statically allocated.
	  The effect is equivalent to dynamic initialisation by a call
	  to pthread_cond_init() with parameter attr specified as NULL,
	  except that no error checks are performed.''

Since dump.c already uses PTHREAD_MUTEX_INITIALIZER instead of
pthread_mutex_init(), this would be consistent.



   Petar Bogdanovic



P.S.
	Why am I the only person with this problem?