The provider can also live in @aws-cdk/aws-route53, where it might make more sense. I agree that it does not make a lot of sense for @aws-cdk/core to know about Route53.

The fact that the "server" part of the implementation lives in aws-cdk is an implementation detail that's born out of pragmatism, but there's no reason that couldn't be moved somewhere else either. We can move those without impact to the user, so I'm not too worried (it's not a one-way door, in that way).

The second part that doesn't feel quite right is the args[] array.

You're right that this might as well have been a map. What we need from it is a consistent serialization to a string (in particular, with order), which an array satisfies quite easily. We could do something like "key1=value:key2=value:key3=value" for a map, if the keys are ordered. In fact, we can probably merge the two disparate arrays for scope and args together.

If you feel like making that change, you're very welcome to it. Otherwise feel free to make an issue for us.

I wonder if the right API for this type of context provider is a static methods similar to HostedZone.import, so users can naturally find it:

const zone: HostedZoneRef = HostedZone.load(this, 'MyExistingHostedZone', { 
  /* filter */ 
});

It seems like this mechanism can be very relevant for certain resources (e.g. #506). Maybe we should think about a common pattern/idiom for supporting those in a discoverable and a clean way.

@eladb -- I think load or lookup makes a lot of sense for me. I think that filter is another flexible data structure. I don't fully understand @rix0rrr comment about ordering in the array being required yet. I need to go look at the code some more. I was hoping to have more of the duck typing, similar to golang passing around interface{} and casting.

I don't fully understand @rix0rrr comment about ordering in the array being required

These are context values, and context values have a controlled update path. They're cached upon first lookup and then kept the same until the user consciously refreshes them, to make sure templates don't change out from under users without them expecting so.

That means that they need to be stored, and that means they need to be serialized--in practice, to a key-value object in cdk.json. The key is based on the "context query", which right now is the name of the of context provider and the stringified value of each of the arguments (combined in long string with : between them).

Maybe an example is simpler:

{
   "context": {
    "availability-zones:1234567:us-east-1": ["us-east-1a", "us-east-1b"],
    "ssm:1234567:us-east-1:/MyParam": "SomeValue"
  }
}

The only thing we need to guarantee is that every query that "means the same" (in effect, set of parameters) leads to the same key in the context object. With an array with concatenation, that's easy! Combine values and it's automatically true. For a dictionary, TYPICALLY that's not so easy since there is no well-defined order for the keys in the dictionary, and it COULD be that every iteration over the keys of a dictionary returns them in a different order. The only thing that means is we need to impose a consistent order on the keys ourselves, so that every serialization is the same.

(Now in practice, in JavaScript it so happens that it maintains insertion order, but for example you would also want the following to be true:

key({ foo: 'foo', bar: 'bar' }) == key({ bar: 'bar', foo: 'foo' })

Which is for example not true if we simply used JSON.stringify as the key function, or did a naive iteration over the keys.

Oh and I'd prefer for the key to be a little aesthetically pleasing as well, which is why I suggested

bar=bar:foo=foo

as a map serialization.

@moofish32 any updates on this? Should we put it on hold for now? Do you need help getting this aligned with master after the indentation change?

@eladb I would like to come back to this in about a week. I would like to propose some changes to the ContextProvider framework to use some typed inputs. There were many things I didn't understand in this first attempt.

However, if the team is set on this pattern I will update and stay in the current flow. Do I have time to propose something in the next 2 weeks?

I don't think the indent will be a problem, I just changed my settings back to 2, this was only 4 project.

closing and moving to #823