bugfix: deallocate mysqli prepared statement

[petermein]

Long running processes might hit the max_prepared_stmt_count due not deallocating the statement correctly.

Q	A
Type	bug
Fixed issues	-

Summary

For postgres there was an implementation to deallocate prepared statements to prevent hitting a any limits. This was not the case for the mysqli drivers. Which can suffer from the same problem.

In this PR a mysqli_stmt_close is added to properly cleanup any open prepared statement.

A similar approach was implemented for PostgreSQL in this PR #5893

[derrabus]

Thank you. Can you try to rebase your bugfix onto the 3.9.x branch?

[petermein]

Thank you. Can you try to rebase your bugfix onto the 3.9.x branch?

Re-based the PR to point to 3.9.x and changed to code to not include any still unsupported language features (property promotion and attributes).

Please have a look if you like the wording of the comments.

[derrabus]

Please have a look at the PHPCS report.

[derrabus]

Can you look into the test failures on PHP 7.4?

[petermein]

Can you look into the test failures on PHP 7.4?

I think the error messages has changed. I pushed a change to the test to expect a different error message <8.0.

Hard to test, I cannot replicate that setup locally easily.

[derrabus]

Still failing.

Hard to test, I cannot replicate that setup locally easily.

What setup? PHP 7.4? What OS are you on?

[petermein]

Still failing.

Hard to test, I cannot replicate that setup locally easily.

What setup? PHP 7.4? What OS are you on?

I found a docker way to replicate it. With some docker magic.

My current attempt is with a deprecated PHPUnit function. Which will throw an E_DEPRECATION. @derrabus do you have any suggestion on how to make this prettier?

        if (PHP_VERSION_ID < 80000) {
            $this->expectError();
            $this->expectErrorMessage('mysqli_stmt::execute(): Couldn\'t fetch mysqli_stmt');
        } else {
            $this->expectException(Error::class);
            $this->expectExceptionMessage('mysqli_stmt object is already closed');
        }

[petermein]

I would propose marking the test as skipped for 7.4, not ideal but I don't see a way to do properly.

[derrabus]

I've pushed a fix for your test.

[derrabus]

Thank you, @petermein!

[petermein]

Thank you for the merge, I contemplated overriding the error reporting but I deemed it too rigorous 😅