Often application can be split into two parts declarative models (DSLs) and imperative interpreter (engines).
Models can be expressed with data - Data DSLs. Why? Because data is easily composable, regular and introspectable.
There are a lot of successful Data DSLs in clojure ecosystem. But often they do not play together :(.
Zen is a framework to unify Data DSLs, make them composable and take model driven design to the next level.
Zen separate models source code from interpreters code, but keep it in source path using same layout as clojure.
Models are grouped into namespaces and namespaces are stored in edn files.
Namespaces can be published as reusable packages.
Each namespace contains one or multiple models described with data.
Namespace is a map (in terms of edn) with two special symbol keys - 'ns and 'import
ns
- defines name of namespaceimports
- imported namespaces
Namespaces may refer other namespaces by imports
.
Core zen
namespace is imported implicitly.
That's how starting from one entry point namespace, your project can import only used modules and models from other packages.
Rest of symbolic keys in namespace define models.
Just like in clojure namespace you may refer one model from another located in one namespace
by short name (symbol
) and refer between namespaces by full name - (namespace.name/symbol
)
Example namespace:
{ns myapp.module ;; namespace name
imports #{http pg model auth} ;; imports - TODO: think about aliases
db
{:zen/tags #{pg/config}
:connection {:host "..." :port "..."}
:storages #{model/patient-store}}
;; model
web
{:zen/tags #{http/server} ;; tags set
:port 8080
:workers 8
:db db
:apis #{api http/admin-api}}
api
{:zen/tags #{http/api}
:zen/desc "API definition"
:middleware [{:type http/cors :allow #{"https://myapp.io"}}
{:type auth/authorize}]
:routes {
"Patient" {:post {:op create-pt}
:get {:op search-pt}}
"meta" {:get {:op http/meta}}}}
create-pt
{:zen/tags #{http/create}
:operation http/create
:schemas #{model/patient}
:response {
201 {:confirms #{model/patient}}
422 {:confirms #{http/error}}}}
search-pt
{:zen/tags #{http/search}
:operation http/search
:store #{model/patient-store}
:params {:query {:name {:zen/desc "Search by name"
:engine http/fhir-search-param
:type "string" :expression "Patient.name"}
:ilike {:engine http/sql-search-param
:query "resource::text ilike {{param.value}}"}}}
:response {
201 {:confirms #{model/search-bundle}}
422 {:confirms #{model/search-errors}}}}}
Instead of introducing any kind of types and type hierarchies, zen uses tag system to classify models.
You may think about tag system as non-hierarchical multidimensional classification (just like java interfaces). Or as a function of meta store - get all models labeled with specific tag.
Model project may be loaded into store. You start loading from entry point namespace. All imports will be resolved, validated and loaded into store.
Store functions:
- get-symbol
ns/sym
- get-tag
tag/name
- read-ns
my/ns
Zen has built-in schema engine deeply integrated with meta-store.
The key features of zen/schema is:
- open world design - i.e. each schema validates only what it knows
- support of RDF inspired property schema - i.e. schema attached to key (only namespaced keys) - not to a key container
{ns myapp
Contact
{:zen/tags #{zen/schema}
:type zen/map
:keys
{:system {:type zen/string
:enum [{:value "phone"}
{:value "email"}]
:value {:type zen/string}}}
Contactable
{:zen/tags #{zen/schema}
:type zen/map
:keys {:contacts {:type zen/vector
:every {:type map :confirms #{Contact}}}}}
User
{:zen/tags #{zen/schema}
:type zen/map
:confirms #{Contactable}
:require #{:id :email}
:keys {
:id {:type zen/string}
:email {:type zen/string :regex ".*@.*"}
:password {:type zen/string }}}
;; example of property schema
human-name
{:zen/tags #{zen/property zen/schema}
:type zen/map
:keys {:family {:type zen/string}
:given {:type zen/vector :every {:type zen/string}}}}}
instance
{:id "niquola"
:myapp/human-name {:given ["Nikolai"] :family "Ryzhikov"}
:password "secret"}
Schema statements are described with maps.
Map may have a :type key, which defines how this
map is interpreted. For example :type zen/string
will check for string, zen/map describes map validation.
Here is list of built-in types:
- primitives
- zen/symbol
- zen/keyword
- zen/string
- zen/number
- zen/integer
- zen/boolean
- zen/date
- zen/datetime
- collections
- zen/vector
- zen/set
- zen/map
- zen/list
- zen/case
You can get all schema types by query meta-store for zen/type tag. User can extend schema with new types - TBD
All schema maps may have common a keys:
:confirms
- set of other schemas to confirm (this is not inheritance!):enum
- polymorphic enumeration of possible values (TODO: think about terminology - reference semantic?):constant
- polymorphic fixed value validation:valuesets
- like enum, but using valueset zen protocol (TBD)
Depending on type schema map may have type specific keys. For example :minLength and :regex for zen/string or :keys for zen/map.
zen/case
is alternative to union type,
it is more advanced and may be applied to different maps
path
{:zen/tags #{'zen/schema}
:type 'zen/vector
:every
{:type 'zen/case
:case [{:when {:type 'zen/string}}
{:when {:type 'zen/map}
:then {:type 'zen/map
:require #{:name}
:keys {:name {:type 'zen/string}}}}]}}
- :tags - constraint to only symbols with tags
For example zen/map
type defines following validation keys:
:values
schema - schema to apply to all values:keys
{ key: schema } - enumeration of keys and schema for each key:require
#{:key,...} - list of required keys in map:schema-key
{:key :some-key } - key to resolve schema from data on fly
Apply clojure.spec regular expressions for collections!!!!
- :every schema - apply schema to every element in collection
- :nth {integer: schema} - apply schema to nth element
- :minItems & :maxItems - min/max items in collection
- :filter - TODO: apply filter to collection, then apply schema to