Re: Sequences in DB/pgsql (once again)

Alexey Borzov wrote: > > Greetings! > > Well, I made an update of PEAR and found sequence implementation... > First of all, I honestly don't think that > > if ($ondemand && DB::isError($result) && > $result->getMessage() == "DB Error: ERROR: db_seq_test.nextval: sequence does not > exist\n") { > > is the intended behaviour... ;] > Relying on error messages' format is probably a Bad Thing itself... I know, consider it a temporary hack. :-) Anyway, as we run the pgsql regression tests on different postgres versions, we'll find out what works and not. I'm thinking of addressing postgres error mapping (so we can use DB's DB_ERROR_BLAH error codes for things like these) by regular expression matching on pg_errormessage() output. There are really no other ways to do it as far as I can see. Again, the implementation will be supported by regression tests so we can verify that it works in different postgres versions. > As for 'db_seq_{$seq_name}' - I don't think that prepending *anything* to > table (or sequence) name is good. The reasons are simple - > prefixes are usually used for hiding something or grouping (if a table > name begins with 'pg_' in Postgres - it is a system table, if it > begins with 'pga_' - it is a PgAccess table and is not shown in > PgAccess). Besides, 'db_' prefix is misleading. I don't have any strong feelings about this, the reason I chose "db_seq_" was to indicate that "this is a sequence created by DB". As long as it is possible, I would like to see a consistent naming scheme for all databases, what it actually ends up being is less important. "_seq" suffix is fine with me. :-) - Stig