Re: Cube extension point support // GSoC'13
От | Alexander Korotkov |
---|---|
Тема | Re: Cube extension point support // GSoC'13 |
Дата | |
Msg-id | CAPpHfdu-6COA6VA+50Q9cF=rEGn3DnzvhnUjbkpqXj4m_1zAVg@mail.gmail.com обсуждение исходный текст |
Ответ на | Re: Cube extension point support // GSoC'13 (Heikki Linnakangas <hlinnakangas@vmware.com>) |
Ответы |
Re: Cube extension point support // GSoC'13
|
Список | pgsql-hackers |
On Fri, Oct 11, 2013 at 5:56 PM, Heikki Linnakangas <hlinnakangas@vmware.com> wrote:
#define SIGLENINT 63 /* >122 => key will toast, so very slow!!! */
2. I didn't understand this change:@@ -422,24 +439,14 @@ g_cube_union(PG_FUNCTION_ARGS)
Datum
g_cube_compress(PG_FUNCTION_ARGS)
{
- PG_RETURN_DATUM(PG_GETARG_DATUM(0));
+ GISTENTRY *entry = (GISTENTRY *) PG_GETARG_POINTER(0);
+ PG_RETURN_POINTER(entry);
}
Datum
g_cube_decompress(PG_FUNCTION_ARGS)
{
GISTENTRY *entry = (GISTENTRY *) PG_GETARG_POINTER(0);
- NDBOX *key = DatumGetNDBOX(PG_DETOAST_DATUM(entry->key));
-
- if (key != DatumGetNDBOX(entry->key))
- {
- GISTENTRY *retval = (GISTENTRY *) palloc(sizeof(GISTENTRY));
-
- gistentryinit(*retval, PointerGetDatum(key),
- entry->rel, entry->page,
- entry->offset, FALSE);
- PG_RETURN_POINTER(retval);
- }
PG_RETURN_POINTER(entry);
}
What did the removed code do, and why isn't it needed anymore?
I don't understand this place even more generally. What decompress function do is to detoast key and return new gist entry with it if needed. But can GiST key ever be toasted? My experiments show that answer is no, it can't. When too long key is passed then GiST falls during inserting key into page. And I didn't find any code doing GiST key toast in git.
However taking care about key toast is sequentially repeated in GiST related code. For example, in contrib/intarray/_int.h
#define SIGLENINT 63 /* >122 => key will toast, so very slow!!! */
Any ideas?
------
With best regards,
Alexander Korotkov.
With best regards,
Alexander Korotkov.
В списке pgsql-hackers по дате отправления: