I'm generally open to making this nullable, but would first like to understand how you ended up in this situation.

The reason why we decided that it's okay to make it non-nullable, is that the only way to end up with an uninitialized $queryString is if you don't create the PDOStatement through the PDO API and instead do something like "new PDOStatement". This is an illegal operation and the resulting object will throw "Error: PDO object is uninitialized" for all method calls. Now it will also throw an Error when accessing its only property, so that seems sensible and consistent. (Ideally we'd throw when you do "new PDOStatement" in the first place, but unfortunately this interferes with the ability to extend PDOStatement, thus we get the current situation where the object can be created but throws on all operations.)

Is my assumption about how you can end up with an uninitialized $queryString wrong? Is there some other pathway besides "new PDOStatement()"? If no, why are you creating / interacting with this kind of object?

We are not manually instantiating the PDOStatement. That was clearly just a simple snippet that would run in that environment.

The statement is generated by a call to "PDO::prepare()".

The use of $queryString in this instance is determining if certain drivers need a post-execute query such as SQLite for a row count.

Since we didn't know how to tell if the query string was available, we were using something like "if ($statement->queryString)". However, that now throws an error since it's uninitialized.

Would it be possible to share a snippet (not necessarily runnable on 3v4l) where this occurs when going through the PDO::prepare() API?

Based on my reading of the code we will always initialize $queryString when PDO creates the PDOStatement (https://github.com/php/php-src/blob/ae6c1b0c4ff3468cbc14ffaaaee4e5dc6e480427/ext/pdo/pdo_dbh.c#L457-L460) and I wasn't able to create an uninitialized $queryString via PDO::prepare() in some simple experiments. I'm probably missing something really basic here.

So, the actual issue code generating the issue is not what I thought. We do create PDOStatement through prepare(), but it seems these failures are from a mock object.

We assumed it was generating it for all queries. Your initial assessment is probably close to correct.

        $statement = $this->getMockBuilder('\PDOStatement')
            ->onlyMethods(['execute', 'rowCount', 'closeCursor', 'fetchAll'])
            ->getMock();
        $driver->getConnection()->expects($this->once())
            ->method('prepare')
            ->with('SELECT 1 FROM sqlite_master WHERE name = "sqlite_sequence"')
            ->will($this->returnValue($statement));
        $statement->expects($this->once())
            ->method('fetchAll')
            ->will($this->returnValue(['1']));
        $statement->method('execute')->will($this->returnValue(true));

        $table = new TableSchema('articles');
        $result = $table->truncateSql($connection);
        $this->assertCount(2, $result);
        $this->assertSame('DELETE FROM sqlite_sequence WHERE name="articles"', $result[0]);
        $this->assertSame('DELETE FROM "articles"', $result[1]);


It would be nice if we could somehow generate PDOStatement's in a mock scenario, but if that requires "isset()" because PDOStatement cannot initialize the property, then we can handle that.

Thanks for the explanation -- this coming up in a mock scenario makes sense to me.

Could it be that your original reason for the `if ($statement->queryString)` check was this mock scenerio as well? Because I don't think it would ever fail in other cases.

I think that the best solution here would be to construct the mock with the property initialized:

        $queryString = 'SELECT 1 FROM sqlite_master WHERE name = "sqlite_sequence"';
        $statement = $this->getMockBuilder('\PDOStatement')
            ->onlyMethods(['execute', 'rowCount', 'closeCursor', 'fetchAll'])
            ->getMock();
        $statement->queryString = $queryString;
        $driver->getConnection()->expects($this->once())
            ->method('prepare')
            ->with($queryString)
            ->will($this->returnValue($statement));

Because then, the code would see the same value it would see in a non-mock scenario.

Unfortunately, this will currently result in:

> Fatal error: Uncaught Error: Property queryString is read only

However, I think we can relax this to allow an assignment as long as the property is still uninitialized. Nowadays our understanding of "read-only" is really "init-once", so that would be in line.

Would this work for you?

The cake database wrapper is too large to show here, but the reason the queryString is checked in a mock scenario is the SQLite driver checks the query post-execute. There's just too much code to completely split up every line in a mock/test scenario so this kind of stuff still runs.

The mock scenario makes it hard to determine if $queryString "should" already be initialized in the driver. We tried tracking things like "ensure execute was called", but the mock scenario created an invalid PDOStatement so that made no difference.

I think the argument that properties can only be initialized by other code is contrary to other argument seen in php dev that all properties should be initialized by the constructor.

That said, allowing us to initialize the queryString as needed in a one time, "readonly" style  would definitely work. That would actually help us test code triggering this error in a mock up properly.

I've allowed initializing assignments to PDOStatement::$queryString in https://github.com/php/php-src/commit/91eb201fd881f7f897d0647113e4a99d4d4f59e3.

Looks like this should work for us in the future, thanks!

Can this ticket be closed now?

Ok with me. Maybe we can add an 8.1 specific note the PDOStatement docs.

Home Renovate Ideas are sharing latest news about home improvement, renovate, design, decoration, real estate etc. More info to visit: (https://homerenovateideas.com)github.com