[openib-general] Re: [PATCH] uDAPL/uDAT autotools - Package for udat, udaplcma, udaplscm

[James Lentini]

On Thu, 16 Mar 2006, Arlin Davis wrote:

> > > Index: dapl/udapl/dapl_init.c
> > > ===================================================================
> > > --- dapl/udapl/dapl_init.c	(revision 5854)
> > > +++ dapl/udapl/dapl_init.c	(working copy)
> > > @@ -66,8 +66,7 @@
> > >  *
> > >  * Return Values:
> > >  */
> > > -void -dapl_init ( void )
> > > +static void __attribute__((constructor)) dapl_init ( void )
> > > {
> > >     DAT_RETURN		dat_status;
> > > 
> > > @@ -118,7 +117,6 @@ dapl_init ( void )
> > > 
> > > bail:
> > >     dapl_dbg_log (DAPL_DBG_TYPE_ERR, "ERROR: dapl_init failed\n");
> > > -    dapl_fini ();
> > >    
> > 
> > Why is this call being removed?
> 
> I moved away from the -init dapl_init and -fini dapl_init linker 
> flags and redefined as constructor and destructor.  dapl_fini will 
> always be called. Which brings up a good point. There is no 
> guarantee that the init will always succeed so should we have a 
> check in the open call and the fini call?

The dapl_fini call is supposed to handle a partially successful 
dapl_init call. 

I don't think there needs to be anything in the open function. If 
dapl_init fails, it should not register as a provider and therefore 
open will not be called. If that is not the case, we should fix it.