Extending Teleport

Teleport’s specification does not define the totality of the t() function, just provides some instances of its inputs and outputs. Any implementation is allowed to extend it with new instances, inventing new concrete types, new generic types or other higher-level constructs.

This implementation provides a convenient interface for extending Teleport with new concrete and generic types. You can see how it works by reading through the following recipes, but first, in order to keep the global namespace clean, make your own personal instance of the t() function:

from teleport import TypeMap, ConcreteType, GenericType

t = TypeMap()

Save this in your project’s package, say, in types.py. To use your extended version of Teleport, simply import your version of t:

from jsonapp.types import t


All these recipes are taken verbatim from the teleport.examples module. You can play around with them in your REPL:

>>> from teleport.examples import t

Recipe: Color

Let’s add a concrete type that matches hex-encoded colors. Use the register() decorator to add a new type to your t instance:

class ColorType(ConcreteType):

    def contains(self, value):
        if not t("String").contains(value):
            return False

        return re.compile('^#[0-9a-f]{6}$').match(value) is not None

Once we have called register(), we can use the new type as a first-class citizen:

>>> t("Color").contains('#ffffff')
>>> t("Color").contains('yellow')
>>> t({"Array": "Color"}).contains(['#ffffff', '#000000']))

If you don’t provide your own from_json() and to_json() implementations, the default implementation assumes that the native form is the same as the JSON form:

>>> t("Color").from_json('#ffffff')

If your purpose for defining custom types is primarily type-checking, then you can forget about those methods altogether, serialization will still work.

Recipe: PythonObject

The pickle module from Python’s standard library provides generic serialization of Python objects. Even though pickle makes the author nervous, we will use it to give Teleport the same power:

import pickle

class PythonObjectType(ConcreteType):

    def from_json(self, json_value):
        if not t("String").contains(json_value):
            raise Undefined("PythonObject must be a string")
            return pickle.loads(json_value)
            raise Undefined("PythonObject could not be unpickled")

    def to_json(self, native_value):
        return pickle.dumps(native_value)

Note that if we implement from_json(), implementing contains() is not necessary, as long as the former behaves correctly by raising Undefined.

Now we can use it to serialize most Python objects:

>>> t("PythonObject").to_json(set([1, 2]))


Never unpickle data coming from an untrusted or unauthenticated source.

Recipe: Nullable

Teleport does not encourage using null unless there is a very good reason for it. One good reason is an existing format that uses it. Either way, the following generic type is a good way to introduce it:

class NullableType(GenericType):

    def process_param(self, param):
        self.child = self.t(param)

    def from_json(self, value):
        if value is None:
            return None
        return self.child.from_json(value)

    def to_json(self, value):
        if value is None:
            return None
        return self.child.to_json(value)

Now you can define weird types like this:

>>> s = t({"Array": {"Nullable": "String"}})
>>> s.contains(["sparse", None, "arrays", None, None, None, "what"])

More realistically, you might use it to deal with JSON objects with null values. The reason this type is not in Teleport core is to discourage us from creating these monsters:

>>> s = t({"Struct": {
...          "required": {"id": "Integer"},
...          "optional": {"name": {"Nullable": "String"},
...                       "age":  {"Nullable": "Integer"}}}})

Even though they may be useful for reading objects like these:

>>> s.contains({"id": 1, "name": "Jake", "age": 28})
>>> s.contains({"id": 1, "name": None, "age": 12})
>>> s.contains({"id": 1, "age": None})