Losing metadata when updating an object implementing Persistable

[gquemener]

Hello,

I'm having trouble updating one document that implements the Persistable interface.
It's unserialization metadata get lost every time I replace the document.

Here's an exemple showing the problem:

<?php

require 'vendor/autoload.php';

class A implements \MongoDB\BSON\Persistable
{
    private $name;
    public function __construct($name)
    {
        $this->name = $name;
    }

    public function setName($name)
    {
        $this->name = $name;
    }

    public function bsonSerialize()
    {
        return [
            'name' => $this->name
        ];
    }

    public function bsonUnserialize(array $data)
    {
        $this->name = $data['name'];
    }
}

$instance = new A('foo');

$manager = new \MongoDB\Driver\Manager('mongodb://mongo:27017');
$collection = new \MongoDB\Collection($manager, 'test', 'coll');
$result = $collection->insertOne($instance);

/**
 * > db.coll.find()
 * {
 *         "_id" : ObjectId("56a269e0046d5f0014645471"),
 *         "__pclass" : BinData(128,"QQ=="),
 *         "name" : "foo"
 * }
 */

$instance->setName('bar');
$collection->replaceOne(['_id' => $result->getInsertedId()], $instance);

/**
 * > db.coll.find()
 * { "_id" : ObjectId("56a26a97046d5f0039107a21"), "name" : "bar" }
 */

What am I doing wrong here please?

I'm running mongodb server v3.2.1.

Thanks in advance!

[gquemener]

Hello!

I've found a way to do it, but I'm not sure this is the correct one:

$instance->setName('bar');
$collection->updateOne(['_id' => $result->getInsertedId()], [
    '$set' => $instance->bsonSerialize()
]);
/**
 * > db.coll.findOne()
 * {
 *         "_id" : ObjectId("56a366dc046d5f0055094a91"),
 *         "__pclass" : BinData(128,"QQ=="),
 *         "name" : "bar"
 * }
 */

What do you think?

[gquemener]

No.
After second tought, this is definitly not a good idea. bsonSerialize() is used internally by the MongoDB driver. Anyway it doesn't work with embedded document.

[jmikola]

I believe the problem is due to your class not handling the _id field.

When inserting the instance, bsonSerialize() returns a document containing only a name field, which leaves the driver to generate an ObjectId on its own (the same that you later retrieve from the write result). When replacing the document, bsonSerialize() is again invoked and returns another document containing only name but with a different value. In my local test, that replacement does achieve the desired goal of changing the name field of the inserted document. I verified this using the shell to dump the document after the insertOne() and replaceOne() calls and there is but one document in the collection:

> db.coll.find()
{ "_id" : ObjectId("56a6a61c6118fd39a911c6e1"), "name" : "bar" }

I would suggest the following to properly handle the _id property on your class:

• Add an $_id property on the class (or whatever you'd like the name it)
• Ensure the property is included in bsonSerialize() if set. Unless you plan to initialize $_id to a new ObjectId (or some other value) during construction, I'd suggest you not include it in bsonSerialize() if it's null, as that will happy create a document in MongoDB with a null ID value (and naturally start throwing duplicate key exceptions after the first insertion).
• Ensure the property is picked up in bsonUnserialize().

For example:

class A implements MongoDB\BSON\Persistable
{
    private $_id;
    private $name;

    public function __construct($name)
    {
        $this->name = $name;
    }

    public function setName($name)
    {
        $this->name = $name;
    }

    public function bsonSerialize()
    {
        $data = [
            'name' => $this->name,
        ];

        if ($this->_id !== null) {
            $data['_id'] = $this->_id;
        }

        return $data;
    }

    public function bsonUnserialize(array $data)
    {
        $this->name = $data['name'];
        $this->_id = $data['_id'];
    }
}

That said, I think it would be easiest to assign a new MongoDB\BSON\ObjectID to $this->_id in your constructor and simplify bsonSerialize(). I expect this may end up being a common mistake for new users, so I will get this added to our documentation examples soon (both on PHP.net and in the upcoming library documentation for using Persistable and the other two interfaces).

---

That said, there is a driver problem here (below the library). The __pclass property is disappearing after the replacement is executed. Curiously, it is present when we debug the replacement document through the BSON serialization functions:

echo MongoDB\BSON\toJSON(MongoDB\BSON\fromPHP($instance));
// outputs: { "__pclass" : { "$type" : "80", "$binary" : "QQ==" }, "name" : "bar" }

Looking into the extension code, I believe we're missing PHONGO_BSON_ADD_ODS|PHONGO_BSON_ADD_CHILD_ODS flags when encoding the replacement arguments in BulkWrite.c. Those are flags we use when inserting documents, and I believe missing them for replacements was an oversight. I'll look into that and cross-reference this with a JIRA ticket on the extension soon.

Thanks for the report!

[jmikola]

Cross-referencing:

• https://jira.mongodb.org/browse/PHPC-549 for documentation improvements
• https://jira.mongodb.org/browse/PHPC-550 for replacement doc encoding bug

[gquemener]

FYI, while waiting for the fix, one must specificaly serialize the __pclass property to prevent metadata to be le lost on document update:

    public function bsonSerialize()
    {
        return [
            '_id' => $this->id,
            '__pclass' => new BSON\Binary(get_class($this), BSON\Binary::TYPE_USER_DEFINED),
            //... whatever other fields you want to persist
        ];
    }

[jmikola]

FYI: PHPC-550 has been merged and is slated for a 1.1.3 release this week.

[jmikola]

1.1.3 has been released, so I think this can be closed. Feel free to follow up if the issue persists.