[ltt-dev] [PATCH 2/9] keep the last position for failed lookup

[laijs@cn.fujitsu.com (Lai Jiangshan)]

On 11/15/2011 07:10 PM, Mathieu Desnoyers wrote:
> * Lai Jiangshan (laijs at cn.fujitsu.com) wrote:
>> Signed-off-by: Lai Jiangshan <laijs at cn.fujitsu.com>
>> ---
>>  rculfhash.c            |   19 +++++++++++--------
>>  tests/test_urcu_hash.c |    5 +++--
>>  urcu/rculfhash.h       |   17 +++++++++++------
>>  3 files changed, 25 insertions(+), 16 deletions(-)
>>
>> diff --git a/rculfhash.c b/rculfhash.c
>> index bda3bd6..cbfc79e 100644
>> --- a/rculfhash.c
>> +++ b/rculfhash.c
>> @@ -1381,27 +1381,28 @@ struct cds_lfht *_cds_lfht_new(unsigned long init_size,
>>  	return ht;
>>  }
>>  
>> -void cds_lfht_lookup(struct cds_lfht *ht, unsigned long hash,
>> +bool cds_lfht_lookup(struct cds_lfht *ht, unsigned long hash,
>>  		cds_lfht_match_fct match, void *key,
>>  		struct cds_lfht_iter *iter)
>>  {
> 
> [...]
> 
>>  #define cds_lfht_for_each_duplicate(ht, hash, match, key, iter, node)	\
>> -	for (cds_lfht_lookup(ht, hash, match, key, iter),		\
>> -			node = cds_lfht_iter_get_node(iter);		\
>> +	for (node = cds_lfht_lookup(ht, hash, match, key, iter) ?	\
>> +			cds_lfht_iter_get_node(iter) : NULL;		\
> 
> Hrm, is it me or this just complexifies use of this API ? Now, instead
> of a 
> 
>   cds_lfht_lookup -> cds_lfht_iter_get_node -> check for null
> 
> chain, we need to do:
> 
>   cds_lfht_lookup -> check for nonzero return ? cds_lfht_iter_get_node :
>                                                     NULL -> check for NULL
> 
> actually adding one test into the caller code.
> 
> What use-case is it allowing to justify this added API complexity ?

I found that it is a pattern:
lookup() --> found --> done
	 --> failed --> create --> insert at the last search place