View Single Post

   
  #1 (permalink)  
Old 04-15-2008, 11:48 PM
Dimitri Fontaine
 
Posts: n/a
Default GiST opclass and varlena

Hi,

I'm trying to code a GiST opclass to index prefix searches (select ... fromt
where t.prefix @> query), now using a prefix_range datatype. This datatype is
a varlena one, and storing it to disk and indexing it with BTrees work ok,
but I'm failing to have my GiST opclass working, here's the problem:

postgres=# create index idx_prefix on ranges using gist(prefix
gist_prefix_range_ops);
NOTICE: gpr_picksplit(): entryvec->n= 234 maxoff= 232 l= 176 r= 56 l+r= 232
unionL='01[0-7]' unionR='01[4-7]'
NOTICE: gpr_picksplit(): v->spl_ldatum='01[0-7]' v->spl_rdatum='01[4-7]'
ERROR: invalid memory alloc request size 3049878020

The code is available at pgfoundry here:
http://cvs.pgfoundry.org/cgi-bin/cvs...prefix/prefix/

The previous support for prefixes as text is still there (but will get
deprecated soon --- or so I hope), and the new datatype and its usage not
well commented nor documented currenlty. If this show up as a requirement to
get your attention, please state it and I'll work on documenting prefix_range
first.

I'm looking for some help on how to resolve the shown index creation problem,
which I think is related to how I give data to GiST in its spl_ldatum and
spl_rdatum from the user defined picksplit() method, lines 1101 and 1102 in
prefix.c (version 1.26).

Regards,
--
dim

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)

iD8DBQBH6SRplBXRlnbh1bkRAuGWAKC9cnM7zNQq/5fRClZPSCjAiYUN+QCglf3Z
GuIEVYi5p2J/fqYRVPTMCbw=
=qB7V
-----END PGP SIGNATURE-----

Reply With Quote