Thank you @MatTheCat.

Hello!

Today, my colleague @samsonradu and I noticed an issue regarding the DoctrineDbalAdapter, potentially introduced by this PR.

Basically, by making some parameters lazily injected, the order of those parameters are no more preserved, and we end up with totally broken SQL queries. Here are more details:

Failed to save key "ulm_ac_rbac_action_roles" of type array: 

An exception occurred while executing

MERGE INTO cache_items 
USING DUAL ON (item_id = ?) 
WHEN NOT MATCHED THEN INSERT (item_id, item_data, item_lifetime, item_time) 
VALUES (?, ?, ?, ?) 
WHEN MATCHED THEN UPDATE SET item_data = ?, item_lifetime = ?, item_time = ? 

with params [
  3600, 
  1709126855, 
  3600, 
  1709126855, 
  "EeA4QGjhnZ:ulm_ac_rbac_action_roles", 
  "EeA4QGjhnZ:ulm_ac_rbac_action_roles", 
  "\x00...\x69..."
]

The introduction of a lazy loading mechanism in this PR has led to issues with the ordering of bound values in SQL statements. New keys added by the lazy loading process are appended at the end of the parameters array, disrupting the intended logical order. See it here:

/app/vendor/symfony/cache/Adapter/DoctrineDbalAdapter.php:354:
object(Doctrine\DBAL\Statement)[1878]
  protected 'sql' => string 'MERGE INTO cache_items USING DUAL ON (item_id = ?) WHEN NOT MATCHED THEN INSERT (item_id, item_data, item_lifetime, item_time) VALUES (?, ?, ?, ?) WHEN MATCHED THEN UPDATE SET item_data = ?, item_lifetime = ?, item_time = ?' (length=223)
  protected 'params' => 
    array (size=8)
      4 => int 3600
      5 => int 1709713419
      7 => int 3600
      8 => int 1709713419
      1 => string 'q55PWWD1R0:ulm_ac_rbac_action_roles' (length=35)
      2 => string 'q55PWWD1R0:ulm_ac_rbac_action_roles' (length=35)
      3 => string 'value1'
      6 => string 'value1'

In here, you can see that the params are not correctly ordered and this is breaking the SQL query.

IMHO, I think this issue should be fixed in doctrine/dbal with a simple ksort on params, but I'm not sure, that's the reason why I comment in here.
Alternatively, we could implement explicit parameter mapping instead of positional indexes to ensure each bound value maintains its correct order, regardless of when it's added.

From the doctrine/dbal doc, I see:

Support for positional and named prepared statements varies between the different database extensions. PDO implements its own client side parser so that both approaches are feasible for all PDO drivers. OCI8/Oracle only supports named parameters, but Doctrine implements a client side parser to allow positional parameters also.

Maybe this is something to fix somewhere else?

I created a draft PR to show you what I have in mind at: #54172, just let me know what you think.

Hey all,

I wanted to provide an update regarding the issue I reported.

After further investigation and numerous attempts to reproduce the issue in doctrine/dbal#6328, including starting from a blank application state and trying to replicate the conditions under which the issue was initially observed, my colleague and I have come to a conclusion that we are now unable to reproduce the issue in any form. Perhaps, this indicates that the problem may have been influenced by factors or states within our specific project setup that we could not accurately replicate in isolated testing, but as of today, we are totally unable to reproduce it.

Edit: We've discovered a rational explanation for the behaviour we observed. The root cause of the confusion stems from the cache_items table being manually created with incorrect column types in our environment. This discrepancy was not immediately evident, leading to our initial belief that there was a problem with the DoctrineDbalAdapter. With this new understanding, I want to clarify that the issue was not with the doctrine/dbal or the changes introduced by the mentioned PR. The anomaly was due to our project-specific database schema misconfiguration.

Given this, the issue I reported is a false alarm. The DoctrineDbalAdapter behaves as expected. I want to extend my apologies for any confusion or concern my initial report may have caused.

I believe it is now appropriate to close the associated PRs (#54172 and doctrine/dbal#6328), unless you think replacing positional parameters with named parameters worth it.

Thanks.