Composing JSON Data

We wish to achieve composability and avoid repetition when specifying "things" to be "done". Cider-CI provides the inheritance and means to sharing data to achieve this. Both require a mechanism to merge data. This page discusses the inner workings of this.

The project configuration is encoded in JSON format. JSON provides maps, arrays and some primitive types. The obvious solution when composing JSON data is to use some kind of "recursive merging". This works well for maps. But it is impossible to specify a general merging strategy for arrays. Arrays are ordered and provide indexes. A meaningful composition must consider the means of the data in the array to produce a sensible composition.

We avoid the overhead of considering semantics of arrays by favoring maps in the project configuration of Cider-CI. Whenever some set, or collection like structure is the natural encoding for some entities we will use a map. The following extract gives a simple example.

 2  prepare:
 3    body: do some preparing
 4  test:
 5    body: test
 6    start_when:
 7      'prepared':
 8        script_key: prepare
 9        states: [ "passed" ]

Here the contents of scripts and start_when could have been arrays. In the case of scripts the keys are important to be used as reference.

The Deep-Merge Strategy

The canonical definition of deep-merge is mnemonic and easy to understand. The following almost formal definition will help to clarify doubts.


Let m1 and m2 be a maps, and let m be the result of (m1 deep-merge m2) Then the following holds true:

  1. If the key k1 with the value v1 is present in m1 but not in m2, then the key value pair (k1,v1) is be present in m.

  2. If the key k2 with the value v2 is present in m2 but not in m1, then the key value pair (k2,v2) is be present in m.

  3. If k is present in m1 and m2

    1. and v1 and v2 are both maps, then the pair (k, ( v1 deep-merge v2)) is present in m.

    2. otherwise the pair (k,v2) is present in m.

Structural Properties

We expand into some structural properties of deep-merge. It is not important to understand following but it deepens the understanding and emphasizes the composability of the chosen method in Cider-CI to aggregate JSON data.

Let S be the set of all JSON encoded data which has a map as its top level data structure, let {} denote the empty map, and the shorthand symbol for deep-merge. Then (a ⊕ (b ⊕ c)) = (a ⊕ b) ⊕ c)) ∈ S and (a ⊕ {}) = ({} ⊕ a) = a for all a, b, c ∈ S. In other words: the deep-merge operation and the set of all maps in JSON form a monoid (S,{},⊕). However, it is clearly not commutative and there exists only the trivial inverse element {}.