Avoid modifying managed RealmList when user executes a query.

[zaki50]

WIP

In model's constructor, calling mutator methods of managed RealmList must be ignored if getAcceptDefaultValue$realm() is false or the name of the field is in the ignore list.

However, RealmList has many methods which is meant to modify itself and that is very hard to provide a natural non-op behavior. Of course it breaks expected List behavior, too.

For example:

private RealmList<Dog> dogs;

// default constructor of DogOwner model
public DogOwner() {
    dogs = new RealmList<Dog>();
    dogs.add(new Dog());

    Dog dog = dogs.get(0); // this line throws an exception or returns unexpected object when a code above is non-op.
}

For that reason, I'm leaning toward adding a method public static boolean RealmObject#acceptsDefaultValue(RealmModel) which returns false when the assigned values are ignored in the constructor.
And when that method returns false, all mutator methods in RealmList throws an exception (this is not an breaking chance since we are already throwing IllegalStateException now).

After introducing acceptsDefaultValue(RealmModel), users can skip any assignment of default values like:

private RealmList<Dog> dogs;

// default constructor of DogOwner model
public DogOwner() {
    if (!acceptsDefaultValue(this)) {
        return;
    }
    dogs = new RealmList<Dog>();
    dogs.add(new Dog());

    Dog dog = dogs.get(0);
}

As Christian commented on #3812 (comment), I think we can automatically add the check by our transformer in most cases. We need to check that adding the code to the following constructor is valid.

public class Foo extends RealmObject {
    final int BAR;
    public Foo() {
        BAR = 1;
    }

    public void printBar() {
        Log.d("baz", "BAR is " + BAR ); // This must print "BAR is 1"
    }
}

fixes #3809

[beeender]

Remind me why do we need to transform all the getter/setters in the constructor? Can we just skip the transforming for the constructor (at least the default one)?

Ask user to have check of acceptsDefaultValue in the model constructor doesn't look good :(

[zaki50]

@beeender mainly for Realm.createObject(Class) to respect the default values.

[kneth]

You have provided a workaround to #3809, but I believe that Realm Java should be as easy to use as possible. From that point of view, I think this proposal makes sense.

[cmelchior]

I would really like to avoid anything like the above. It doesn't really solve the problem of being intuitive to use since you would need to read the documentation for both this and doing the work-around in #3809. If given a choice between the two I would probably prefer to just document the current bug and how to work around it instead of introducing new methods of questionable value?

However, could it be possible to inject that code using our transformer? That way people wouldn't need to do anything?

[zaki50]

@cmelchior That's a great idea.
I think it's possible.

[zaki50]

Adding if (!acceptsDefaultValue(this)) { return; } is possible but the following code does not work as expected.

public class Foo extends RealmObject {
    @Ignore
    final int BAR;
    public Foo() {
        BAR = 1;
    }

    public void printBar() {
        Log.d("baz", "BAR is " + BAR ); // This must print "BAR is 1"
    }
}

printBar() prints 0 instead of 1.

My current conclusion is that we should not inject if (!acceptsDefaultValue(this)) { return; } by our transformer.