[ECOS] Bug in snmp2c in current build tree

[ECOS] Bug in snmp2c in current build tree

[Ron Kundla]

I am not sure how many people have been messing with the MIB compiler
included in the latest build tree but I have been having some problems using
mib2c to compile the RFC-3083 (DOCSIS Baseline Privacy MIB).

My system setup is the same as described in the eCos SNMP documentation,
using Perl 5.005_03 and the SNMP-3.1.0 Perl package. I am running on a
Windows 2000 box using Cygwin 1.3.2.

The compiler would run on the MIB and generate output. What I noticed was
that the header file only contained one definition of for a table when there
are seven of them defined in the MIB. 

Now, if you haven't looked at mib2c, it is pretty darn confusing! After some
tinkering, I found a line of code that didn't quite jive with my test debug
print statements. So I made the following change:

Old Code:

#============================================
#
#       Table-handling routines.
#
#============================================
foreach $vtable (@table_list) {
    foreach $ptable (@processtable) {
        $variables{$ptable}{'processed'} =
            (eval "\"$variables{$ptable}{'code'}\"") . "\n\n";
    }
    $var_routines .=
        (eval "\"$variables{'code-var_table'}{'code'}\"") . "\n\n";
}

New Code:

#============================================
#
#       Table-handling routines.
#
#============================================
foreach $vtable (@table_list) {
    foreach $ptable (@processtable) {
        $variables{$ptable}{'processed'} .=  <--- *** NEW CHANGE ***
            (eval "\"$variables{$ptable}{'code'}\"") . "\n"; <--- here too,
cosmetic
    }
    $var_routines .=
        (eval "\"$variables{'code-var_table'}{'code'}\"") . "\n\n";
}

and voila! My header file had the proper tables entries and the
corresponding c file also had the proper functions to access the tables.

Has anyone else reported this as a problem? In discussing this with a
co-worker, we made the observation that the DOCSIS MIBs could be compliant,
but the fact that in mib2c they concatenate the one variable but not the
other proved the reason why I was only seeing the *last* table in the header
file.

Ron

==============================================
Ron Kundla          nBand Communications, Inc.
Systems Engineer       Mountain View, CA 94041
rkundla@nband.com           Main: 650.716.0700

Re: [ECOS] Bug in snmp2c in current build tree

[Hugo Tyson]

Ron Kundla <rkundla@nband.com> writes:

> The compiler would run on the MIB and generate output. What I noticed was
> that the header file only contained one definition of for a table when
> there are seven of them defined in the MIB.

I noticed that same issue when initially getting the SNMP package
introduced to eCos' network stack, but time pressure and...
 
> Now, if you haven't looked at mib2c, it is pretty darn confusing!

...is why I just lived with it and fixed up the header by hand; IIRC
only having to tinker with a couple of files.

> some tinkering, I found a line of code that didn't quite jive with my
> test debug print statements. So I made the following change:
> <snip>
> and voila! My header file had the proper tables entries and the
> corresponding c file also had the proper functions to access the tables.

That's excellent, so I have committed the change (below) to our source
repository - your name in lights! ;-)

> Has anyone else reported this as a problem? In discussing this with a
> co-worker, we made the observation that the DOCSIS MIBs could be compliant,
> but the fact that in mib2c they concatenate the one variable but not the
> other proved the reason why I was only seeing the *last* table in the header
> file.

Yes, I agree with that analysis.  I'll copy the guys at UCD with this to
let them adopt the change, if they haven't got it already.

Thanks again,

	- Huge



Index: net/snmp/agent/current/ChangeLog
===================================================================
RCS file: /home/cvs/ecc/ecc/net/snmp/agent/current/ChangeLog,v
retrieving revision 1.30
diff -u -5 -p -r1.30 ChangeLog
--- net/snmp/agent/current/ChangeLog    2001/06/07 12:21:28     1.30
+++ net/snmp/agent/current/ChangeLog    2001/07/31 14:54:07
@@ -1,5 +1,13 @@
+2001-07-31  Hugo Tyson <hmt@redhat.com>
+2001-07-31  Ron Kundla <rkundla@nband.com>
+
+       * utils/mib2c/mib2c (Table-handling): Concatenate rather than
+       replace subsequent processed variables.  The symptom was that in
+       the generated header file you only get the last table in the MIB.
+       This fixes it.  Thanks Ron!
+
 2001-06-07  Jesper Skov  <jskov@redhat.com>
 
        * src/mibgroup/mibII/ip.c (var_ipNetToMediaTable): Only handle
        IPNETTOMEDIAPHYSADDRESS for devices with associated hardware.
 
Index: net/snmp/agent/current/utils/mib2c/mib2c
===================================================================
RCS file: /home/cvs/ecc/ecc/net/snmp/agent/current/utils/mib2c/mib2c,v
retrieving revision 1.1
diff -u -5 -p -r1.1 mib2c
--- net/snmp/agent/current/utils/mib2c/mib2c    2000/06/02 13:05:26     1.1
+++ net/snmp/agent/current/utils/mib2c/mib2c    2001/07/31 14:54:07
@@ -130,11 +130,11 @@ $varlen = $varLengths[$varlen];
 #      Table-handling routines.
 #
 #============================================
 foreach $vtable (@table_list) {
     foreach $ptable (@processtable) {
-       $variables{$ptable}{'processed'} = 
+       $variables{$ptable}{'processed'} .= 
            (eval "\"$variables{$ptable}{'code'}\"") . "\n\n";
     }
     $var_routines .= 
        (eval "\"$variables{'code-var_table'}{'code'}\"") . "\n\n";
 }