[Pvfs2-developers] Re: [Pvfs2-cvs] commit by slang in pvfs2/src/io/bmi: bmi-method-callback.h bmi-method-support.c bmi-method-support.h bmi.c op-list.c op-list.h reference-list.c reference-list.h

[Pete Wyckoff]

slang at mcs.anl.gov wrote on Wed, 07 Nov 2007 10:19 -0600:
> On Nov 7, 2007, at 10:06 AM, Sam Lang wrote:
> >On Nov 7, 2007, at 7:17 AM, Pete Wyckoff wrote:
> >>cvs at parl.clemson.edu wrote on Tue, 06 Nov 2007 18:08 -0500:
> >>>diff -p -u -r1.88 -r1.89
> >>>--- bmi.c	20 Oct 2007 15:40:13 -0000	1.88
> >>>+++ bmi.c	6 Nov 2007 23:08:33 -0000	1.89
> >>>@@ -214,7 +214,7 @@ int BMI_initialize(const char *method_li
> >>> 		if(!strncmp(listen_addrs[j], proto, strlen(proto)))
> >>> 		{
> >>> 		    /* found the right addr */
> >>>-		    this_addr = listen_addrs[j];
> >>>+		    this_addr = strstr(listen_addrs[j], "://") + 3;
> >>> 		    break;
> >>> 		}
> >>
> >>Instead of passing down tcp://foo:343/bar, you now pass
> >>foo:343/bar.  Looks like you fixed up the TCP method to
> >>expect that, maybe, but not the others.  Is this a bug?
> >>Maybe it's fixed by your big patch, and just an artifact
> >>of the separation.
> >
> >I didn't change the tcp method to start at the hostname.  it was  
> >doing that already, so I'm not sure what changed.  It looks like  
> >the other methods expect the string to start at the hostname as  
> >well.  IB has some magic to tack on ib:// component to the string,  
> >but then you wrote that code, so maybe I'm missing something.
> 
> Hmm...ok I was reading that string_key code wrong.  I still don't see  
> how the server and BMI_initialize could have been passing in the full  
> id string (tcp://hosta:3334) and bmi-tcp could have been expecting  
> just the host:port.  This obviously worked previously, just not sure  
> what I'm missing at this point.

I think bmi-tcp was (before your patch) expecting the full string too.
It did work in the past, which is why I noticed that as a
potentially "interesting" change.

		-- Pete