[BUG]: change zend_inline_hash_func to avoid collisions and make parser_cache setting working

[Jurigag]

Describe the bug zend_inline_hash_func creates collisions http://www.phpinternalsbook.com/php5/hashtables/hash_algorithm.html#hash-collisions

In our specific case Phalcon\Mvc\Model\Query::parse() creates exactly same ast for different phql causing us to return same values for different queries because we store ast representation in cache, which hash we build using zend_inline_hash_func.

We have pretty long ids(for elasticsearch, custom implementation) but we use parse method of phalcon query anyway. I guess this is possible to reproduce as well to use use same namespaces etc.

I know that partially the issue here is that we put id value in query - but still we should maybe use better algortihm for this hashing.

Also the issue here is that we tried to set phalcon.orm.parser_cache in php.ini to disable this functionality but none of values worked, like -1, php ini_get always returns false

I'am going also to provide PR with simple test, hopefully it will fail too.

To Reproduce

Steps to reproduce the behavior:

Run code below

Provide minimal script to reproduce the issue

        $ast1 = \Phalcon\Mvc\Model\Query\Lang::parsePHQL("SELECT [TagMe\Modules\Identity\Infrastructure\User\Model\Elasticsearch\Users].* FROM [TagMe\Modules\Identity\Infrastructure\User\Model\Elasticsearch\Users] WHERE _id = 'GcbN9FIoI8IC3LqBIuCu50AC02TpI8ruKOjyzd1fqOCuna' LIMIT :APL0:");
        $ast2 = \Phalcon\Mvc\Model\Query\Lang::parsePHQL("SELECT [TagMe\Modules\Identity\Infrastructure\User\Model\Elasticsearch\Users].* FROM [TagMe\Modules\Identity\Infrastructure\User\Model\Elasticsearch\Users] WHERE _id = 'GcbN9FIoI8IC3LpcIuCu50AC02TpI8ruKOjyzd1fqOCuna' LIMIT :APL0:");
// it should be the same ast representations

Expected behavior Return different ast representation so query will work.

Details

• Phalcon version: 3.4, but should happen on 4.0 as well (php --ri phalcon)
• PHP Version: 7.2.26
• Operating System: Ubuntu 16.04
• Installation type: Compiling from source
• Zephir version (if any): not related, we use zend function internally
• Server: Nginx
• Other related info (Database, table schema): elasticsearch, but not related in that case

[Jurigag]

For now i solved it in our project just to not include full id in query, but still, is there any possibility we could use some better algorithm to avoid collisions? Or maybe add some way to choose which one we want to use?

[ruudboon]

Gonna add this to 4.1 where we're planning more parser fixes

[zsilbi]

This issue will reopened later and fixed for 5.0