On Sat, Sep 29, 2012 at 4:05 PM, Bill Meier <wmeier@xxxxxxxxxxx> wrote:
> On 8/21/2012 2:19 AM, Guy Harris wrote:
>>
>>
>> On Aug 20, 2012, at 12:49 PM, Kaul wrote:
>>
>>> If it were git, I'm sure I could easily use 'git bisect' and find the
>>> issue.
>>
>>
>> As Evan Huus indicated, what you probably need to bisect is compiler
>> updates. :-)
>>
>> The offending routine is probably relinquish_special_privs_perm()
>> (which has failed to check the return value of setresuid() and
>> setresgid() since at least 2007), and, if those calls were to fail, it's
>> not clear what Wireshark should do, so it's not clear that
>> relinquish_special_privs_perm() should return a success/failure
>> indication, and it's not clear that there's anything to do if any of the
>> set.*id routines return anything other than 0, so perhaps the right fix
>> is to cast the return values to void, so as to tell the compiler "yes,
>> I'm deliberately ignoring the return value, because there's not much to
>> do if they fail".
>>
>> I suppose we *could* have relinquish_special_privs_perm() return a
>> success/failure indication, and have its *callers* cheerfully ignore
>> the return value, or print a warning message/pop up a warning dialog
>> if they fail (if they fail, then we were started with special
>> privileges, but those privileges weren't sufficient to relinquish
>> them, in which case the developers should be notified so we can
>> figure out whether there's any way to relinquish them in that
>> situation).
>
>
>
> I've started getting this warning also (F17...) (maybe just now since today
> is the first time I've done a 'make clean' etc in a while).
>
> So:
>
> 1. There's a long thread (which started Jul 24,2012) about
> "[PATCH] Declare set*id with warn_unused_result"
>
> at
>
> http://old.nabble.com/-PATCH--Declare-set*id-with-warn_unused_result-td34204904.html
>
> 2. Doing a (void)... as suggested by Guy above doesn't prevent the error.
> Using a hack like 'if(!setresgid() {}' does work but seems ugly;
>
> returning a success/failure indication to the callers of
> relinquish_special_privs_perm() seems like too much work.
>
> My inclination: test the return value of the various set*id calls and if
> fail, do g_error().
>
> Thoughts ?
I'd be tempted to make it a g_warning() since Wireshark will
*probably* keep working despite the problem. The general approach
seems sound though.