Idea for mocking relational data about graphql-tools

[helfer]

A refinement of the idea above:
Mock objects are created only when they are requested by a resolver, but once they have been created, they get inserted into a pool of objects of that type. Whenever an object of this type needs to be returned by a mock resolver, we check the pool to see if we find something that matches the requirements of the kind of object that that mock resolver generates. If we do, we return that object instead of creating a new one.
Lists (and lists of lists) would get existing objects from the pool until the pool is depleted, then generate new ones to get to the expected number of items in the list.

Generating mock objects in this way would have the benefit that data returned in subsequent queries would be consistent while not giving up the convenience of creating mock data on the fly.

The main challenge for implementing this is creating an easy to use interface for creating mock scalars that we can search against if some randomness is desired. One solution that comes to mind is to create classes for all scalars like we did for MockList. A random int between 10 and 20 for example would be represented as new MockInt(10,20). If type Person has an age, and a mock resolver for the Person type returns people with ages between 10 and 20, we can then use the MockInt instance to search the store of existing Person and return a matching one if it exists, or create a new one if it doesn't.

Strings could be represented by their length, a regexp that they have to match etc. If we go down this path, my guess is that further down the road it will lead to duplicating much of the functionality of boo1ean/casual, but in a way that lets us search our pool for a matching randomly generated thing.

[helfer]

Actually, thinking about this a little bit more, it seems attractive at first, but doesn't really get us all the way there. If you want mocked data to be completely realistic and support almost anything, I think the best way to go is to generate that data at startup according to some rules which can be defined in a similar fashion to the mock data before.

For example, let's say you want to mock some users who have todo lists with tasks. Some lists have owners, some don't. Some tasks are completed, some are not.

For this scenario, you'd like to be able to say how many users you want to generate, then how many lists each user should have, then how many tasks each list should have and how many of them should be complete vs. incomplete. Then you probably also want to create some lists that don't have owners. You can do that separately.

Next you'll have to define actual resolve functions that work against this JSON store, which is basically like writing a server, but just a slightly simpler one because you don't have to write the loaders. What that really boils down to is that we've basically written a server. Now we're mocking the backends and the loaders, but you have to write all the resolve functions. That's useful too, but it requires a lot more effort than before.

My intuition here is to see how far people can go with stateless mocking and then find other solutions for more complex tasks, if that's desired.

[gbisheimer]

Great work you are doing with apollostack! Keep going!

Just wondering if you know this tool for creating a mock api. The UI is nice and configuring the mock data pretty fast and easy.

[helfer]

Thanks! Yes, I checked mockapi.io when I was writing the medium post. The UI is really nice, but I think we could do much better with creating mock data, especially relational one. When I tried it, I couldn't even figure out how to mock an array of integers, which should be trivial imho.

The other downside of mockapi is that you can't run it locally, so it's not as useful for unit testing.

With relatively little work, some could build a really good mocking service for GraphQL with a great UI like mockapi, but much more powerful, I think.

[helfer]

Closing this issue for the time being, it can still be referenced if people have questions.