Add the PgSQL driver

[derrabus]

Q	A
Type	feature
Fixed issues	N/A

Summary

This PR introduces a new Postgres driver that leverages PHP's pgsql extension. Postgres is currently the last database that we only support through PDO. This driver closes the gap.

ext-pgsql has a rather verbose API. Unlike other database extensions, it does not even bother to cast result types: It returns everything as a string. Except for null values: Those are really null!

However, we get information on the database type of all result fields, so we are able to perform those casts ourselves. This is why I decided to perform very basic type mappings in the driver already. Otherwise, the result sets produced by ext-pgsql would have been a permanent edge case: Booleans come as 't' and 'f', binary data is returned as hex string prefixed with 0x. Not casting those would cause a lot of functional tests to fail.

The type casts that I perform in the driver should be similar to what PDO_pgsql does internally as well. That should make a potential switch from PDO to the native extension rather seamless.

In PHP 8.1, all resource types used by the extensions have been changed to objects. This is why you will see a couple of type annotations with a union of resource and some class from the PgSql namespace. This PR currently targets DBAL 3.6 and PHP 7.4. Once this code hits the 4.x branch, we can simplify a lot of type declarations.

The one big limitation though is that I haven't implemented support for loading BLOBs from resources yet. That topic is a bit complex, so I'd like to work on that in a follow-up. For now, I think we can ship this driver without BLOB support.

[derrabus]

Psalm actually runs better using its own ext-pgsql stubs. 🧐

[derrabus]

Running the ORM test suite uncovered a nice little bug in the type mapping logic. 😓 On it.

[derrabus]

I've added Doctrine\DBAL\Tests\Functional\Driver\PgSQL\ResultTest::testTypeConversionWithDuplicateFieldNames() to reproduce the problem found by the ORM's test suite and fixed the issue.

The ORM's test suite passes now, but it's significantly slower with the new driver than it is with PDO. 😕

But I can investigate this in a follow-up, I guess.

[greg0ire]

but it's significantly slower with the new driver than it is with PDO

Should we somehow document the new driver as experimental because of this?

[derrabus]

Done.

[andrew-demb]

The ORM's test suite passes now, but it's significantly slower with the new driver than it is with PDO. 😕

It may be related to "real" prepared statements, in comparison to emulated prepared statements in PDO.

[derrabus]

You mean because those real prepared statements mean an extra round-trip to the Posgres server? But that is nevertheless desirable, isn't it? I'm not sure I want to resolve prepared statements in the driver. 🙈