Fixes for insertid issues with postgresql driver

[carlopires]

Also includes:

• Fixes to postgresql sql script
• Fixes to database adapters

Gabriele,

First, thanks for your hard work on make joomla run with postgresql. I'm using it right now.

I had to make some changes to make my components install with postgresql_v2 branch. I'm pulling to you to see if these fixes are useful to you.

[gpongelli]

This is JDatabase::insertObject, base class for all drivers, so this is common for all driver while "RETURNING" clause works only on PostgreSQL and maybe on Oracle, but surely it doesn't work on MySQL .
This change, and the following on this file, can be useful for JDatabasePostgresql::insertObject, or change this to use implementation merged inside platform .
Anyway, thank you to submit this change I think I'll push inside platform's driver.

[carlopires]

Oh yeah... I think this code must go to platform's driver. The current code did not work to me (postgresql 9.1)

[gpongelli]

It does not work because there's no JDatabasePostgresqlQuery for that INSERT INTO statement, so "insertid" will crash.
If you change this code with that already present inside platform it will work.

[gpongelli]

Interesting change, could be better implemented without same code replication, maybe doing what you've proposed for database.php

[gpongelli]

The real effort of this block is replacement of "insertid" with "returning" JDatabasePostgresqlQuery element.
I'm thinking a better solution than existing and than this change.

Ps
platform code has to pass some code style check, this change will not pass them.

[carlopires]

Gabriele,

I changed JDatabase::insertObject and JDatabase::updateObject like you suggested and could revert the changes about insertid on installer adapters.

I was wondering if JDatabasePostgreSQL::inserid could just be just:

return $this->loadResult()

As JDatabasePostgreSQL::insertObject are using RETURNING clause.

The problem is that JDatabasePostgreSQL::insertObject can be called null key parameter and this could crash insertid in some point. I decided to keep the current code and now everything is working fine. The final diffs are now much less impactive.

Thanks a lot for you help.

[gpongelli]

I'm reading pg_num_rows doc, it accepts only resources so casting null to integer doesn't seem to be a resource.
I think it's better something like ( !isnull($cur) ? $cur : $this->cursor) , don't you?

[carlopires]

$cur parameter comes with boolean value (false) and is causing php to issue warnings... What about this:

return pg_num_rows( is_numeric($cur) ? (int)$cur : $this->cursor);

[gpongelli]

I think the error is $cur value set to false, it has to be a resource link .

[carlopires]

I think so, but to fix this I will need to revisit all getNumRows calls and this is out of my possibilities, now.

[gpongelli]

These changes are what I wanted to do after merging this pull, thank you.
+1

[gpongelli]

Really good job, thank you very much of this pull!
I'll do same change inside platform code to keep them.

Eng. Gabriele Pongelli.