Re: [dpdk-dev] [PATCH v6 07/12] app/test: clean LTO build warnings (maybe-uninitialized)

[Andrzej Ostruszka <amo@semihalf.com>]

Yipeng

Thank you for your comment.  See my reply below.

On 11/1/19 6:15 PM, Wang, Yipeng1 wrote:
>> -----Original Message-----
[...]
>> diff --git a/app/test/test_hash_readwrite.c b/app/test/test_hash_readwrite.c
>> index 4376b099b..615767fb6 100644
>> --- a/app/test/test_hash_readwrite.c
>> +++ b/app/test/test_hash_readwrite.c
>> @@ -298,7 +298,7 @@ test_rw_reader(void *arg)
>>
>> 	begin = rte_rdtsc_precise();
>> 	for (i = 0; i < read_cnt; i++) {
>> -		void *data;
>> +		void *data = arg;
> 
> [Wang, Yipeng] 
> Hi Andrzej, thanks for the fix! Maybe you could initialize the data to be "NULL"? 
> I think it makes more sense to be NULL rather than arg.

That actually is not a good idea.  The conditional test below does not
check return value from lookup and expects the returned pointer to data
to be equal to the iteration count (these pointers do not point at
anything).  For the first loop iteration "data = NULL" could miss the
problem of the key not being in the hash (key not in the hash, return
value not checked but the pointer "data" is NULL as we expected since i
= 0).

What I used is a bit hackish - I assumed that arg pointer (which is
pointing to the "read_cnt") will not be within [0-read_cnt] address
range.  If that is too hackish then it would be better to rework this
function to actually test for return value of hash lookup.  Let me know
what you think.

Regards
Andrzej