[AGENT++] Consts in SNMP++

[Frank Fock]

Hi Marek,

Since friendly_name() seems not be called by SNMP++/AGENT++ internally,
it would be sufficient if the user of an address object synchronize 
access to this
method when the address object is used in multiple threads. On the other 
hand,
it would also be no problem to make the method synchronized in SNMP++.

Let's see what Jochen decides ;-)

Thanks again for sorting this out,
Frank

Marek Malowidzki wrote:

>>Unfortunately, the original v2.x of SNMP++ did not use const at all
>>(or only accidentially). The reason for it were some internal needs at
>>HP. Peter Mellquist wrote me, that they otherwise had some compilation
>>problems with existing applications.
>>
>>Since SNMP++v2.x seems to be a dead end these days, I do not think
>>that we have to worry about those SNMP++v2.x users. That's why
>>Jochen has already made most of the const methods const. Obviously,
>>we missed some of those. So thanks for your hints, we will try to fix
>>    
>>
>them.
>
>There is no problem and I realize that the library is being continuously
>improved. I hit the problem after I had added some 'const's to SNMP++.NET
>and got multiple compilation errors.
>
>By the way, are you going to add locking to friendly_name()? It uses
>strcpy() to write the friendly name to a buffer. In fact, the danger is
>quite slight (short strings, most likely the same names got in multiple
>calling threads) but still exists. In many places, you use:
>
>iv_friendly_name[0]=0;
>
>and friendly_name() checks the first byte. The problem is that when one
>thread checks this first byte while the other one is writing the firiendly
>name, a crash may cause (if the terminating '\0' is not present
>accidentally). The solution is either:
>1.) locking
>2.) when we assume that the resolver returns the same friendly name each
>time it is called, or that we can tollerate "messed" names, it would be
>sufficient to either fill the buffer with zeros (completely) or write the
>first byte as the last one:
>
>// strcpy(iv_friendly_name, s);
>strcpy(iv_friendly_name + 1, s + 1);
>iv_friendly_name[0] = s[0];    // now the buffer is valid
>
>Note that setting the validity flag is prone to race conditions, so it will
>not help.
>
>Marek
>
>_______________________________________________
>AGENTPP mailing list
>AGENTPP at agentpp.org
>http://agentpp.org/mailman/listinfo/agentpp
>
>  
>