On 2/27/24 12:13, jian he wrote:
> On Mon, Feb 12, 2024 at 7:47 PM Peter Eisentraut <peter@eisentraut.org> wrote:
>>
>>
>> In principle, hashRowType() could process all the fields that
>> equalRowTypes() does. But since it's only a hash function, it doesn't
>> have to be perfect. (This is also the case for the current
>> hashTupleDesc().) I'm not sure where the best tradeoff is.
>
> That's where my confusion comes from.
> hashRowType is used in record_type_typmod_hash.
> record_type_typmod_hash is within assign_record_type_typmod.
>> in assign_record_type_typmod:
> ------------------------------------------------
> if (RecordCacheHash == NULL)
> {
> /* First time through: initialize the hash table */
> HASHCTL ctl;
> ctl.keysize = sizeof(TupleDesc); /* just the pointer */
> ctl.entrysize = sizeof(RecordCacheEntry);
> ctl.hash = record_type_typmod_hash;
> ctl.match = record_type_typmod_compare;
> RecordCacheHash = hash_create("Record information cache", 64,
> &ctl,
> HASH_ELEM | HASH_FUNCTION | HASH_COMPARE);
> /* Also make sure CacheMemoryContext exists */
> if (!CacheMemoryContext)
> CreateCacheMemoryContext();
> }
> /*
> * Find a hashtable entry for this tuple descriptor. We don't use
> * HASH_ENTER yet, because if it's missing, we need to make sure that all
> * the allocations succeed before we create the new entry.
> */
> recentry = (RecordCacheEntry *) hash_search(RecordCacheHash,
> &tupDesc,
> HASH_FIND, &found);
> ------------------------------------------------
> based on the comments in hash_create. The above hash_search function
> would first use
> record_type_typmod_hash to find out candidate entries in a hash table
> then use record_type_typmod_compare to compare the given tupDesc with
> candidate entries.
>
> Is this how the hash_search in assign_record_type_typmod works?
>
Yes.
> equalRowTypes processed more fields than hashRowType,
> hashRowType comments mentioned equalRowTypes,
> maybe we should have some comments in hashRowType explaining why only
> hashing natts, tdtypeid, atttypid will be fine.
>
Not sure I understand what the confusion is - omitting fields with
little entropy is not uncommon, and collisions are inherent to hash
tables, and need to be handled anyway.
regards
--
Tomas Vondra
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company