Introducing the members

As we have seen in the introduction, members are used in the class definition of an atom object to define the fields that will exist on each instance of that class. As such, members are central to atom.

The following sections will shed some lights on the different members that come with atom and also how they work which will come handy when we will discuss how you can customize the behaviors of members later in this guide.


Starting with atom 0.7, atom ships with type hints allowing type checkers to resolve the values behind a member. More details about how typing works in atom and how to add custom type hints can be found in Static type checking explained

Member workings

From a technical point of view, members are descriptors like properties and they can do different things when you try to access or set the attribute.

Member reading

Let’s first look at what happen when you access an attribute:

class Custom(Atom):

    value = Int()

obj = Custom()

Since we did not pass a value for value when instantiating our class (we did not do obj=Custom(value=1)), when we first access value it does not have any value. As a consequence the framework will fetch the default value. As we have seen in the introduction, the default value can be specified in several ways, either as argument to the member, or using set_default or even by using a specially named method (more on that in Customimizing members: specially named methods).

Once the framework has fetched the default value it will validate it. In particular here, we are going to check that we did get an integer for example. The details of the validation will obviously depend on the member.

If the value is valid, next a post-validation method will be called that can some do further processing. By default this is a no-op and we will see in Customimizing members: specially named methods how this can be customized.

With this process complete, the state of our object has changed since we created the value stored in that instance. This corresponds to a create that will be sent to the observers if any is registered.

The observer called, the value can now be stored (so that we don’t go through this again) and is now ready to be returned to the sure, the get step is complete. However before doing that we will actually perform a post-gettatr step. Once again this is a no-op by default but can be customized.

On further accesses, since the value exists, we will go directly retrieve the value and perform the post-getattr, and no notification will be generated.

To summarize:

digraph getattr { a [label="A value was previously set?"]; a->b[label="Yes"]; a->c[label="No"]; b[label="get the value"]; c[label="retrieve the default value"]; c->d; d[label="validate the value"]; d->e; e[label="run post-validation"]; e->f; f[label="store the value"]; f->g; g[label="call observers"]; g->i; c->i; i[label="run post-getattr"]; i->j; j[label="return the result of post-getattr" ]; }

Member writing

Setting a value follows a very similar pattern. First the value is of course validated (and post-validated). It is then actually stored (set).

Next as for the get and validate operation, a post-setattr step is run. As for the other post by default this won’t do anything.

Finally is any observer is attached, the observers are notified.

To summarize:

digraph setattr { a [label="validate the value"]; a->b; b[label="run post-validation"]; b->c; c[label="store the value"]; c->d; d[label="run post-setattr"]; d->e; e[label="call observers"]; }

Members introduction

Now that the behavior of members is a bit less enigmatic let’s introduce the members that comes with atom.

Members for simple values

Atom provides the following members for basic scalars types:

  • Value: a member that can receives any value, no validation is performed

  • Int: an integer value. One can choose if it is allowed to cast the assigned values (float to int), the default is true.

  • Range: an integer value that is clamped to fall within a range.

  • Float: a floating point value. One can choose if it is allowed to cast the assigned values (int to float, …), the default is true.

  • FloatRange: a floating point value that is clamped to fall within a range.

  • Bytes, Str: bytes and unicode strings. One can choose if it is allowed to cast the assigned values (str to bytes, …), the default is false.

  • Enum: a value that can only take a finite set of values. Note that this is unrelated to the enum module.

Containers and type validation

Atom also provides members for three basic containers tuple, list and dictionaries: Tuple, List, Dict. In each case, you can specify the type of the values (key and value for dict), using members, as follows:

class MyAtom(Atom):

    t = Tuple(Int())
    l = List(Float())
    d = Dict(Str(), Int())

Alternatively, you can pass simple Python types. In this cases they will be wrapped in an Instance member that will be introduced in the next section.

class MyAtom(Atom):

    t = Tuple(int)
    l = List(float)
    d = Dict(str, int)


Note that you cannot (by default) enforce a specific number of items in a tuple.


In order to enforce type validation of container, atom has to use custom subclass. As a consequence, when assigning to a member, the original container is copied. This copy on assignment behavior can cause some surprises if you modify the original container after assigning it.

One additional important point, atom does not track the content of the container. As a consequence, in place modifications of the container do not trigger any notifications. One workaround can be to copy the container, modify it and re-assign it. Another option for lists is to use a ContainerList member, which uses a special list subclass sending notifications when the list is modified.

Enforcing custom types

Sticking to simple types can quickly be limiting and this is why atom provides member to enforce that the value is simply of a certain type or a subclass:

  • Instance: the value must pass isinstance(value, types)). Using Instance once can specify a tuple of types.

  • Typed: the value must of the specified type or a subtypes. Only one type can be specified. This check is equivalent to type(obj) in cls.mro(). It is less flexible but faster than Instance. Use Instance when allowing you need a tuple of types or (abstract) types relying on custom __isinstancecheck__and Typed when the value type is explicit.

  • Subclass: the value must be a class and a subclass of the specified type.


By default, Typed and Instance consider None to be a valid value if no way to build a default value was provided. One can explicitly specify if None is a valid value by using the optional keyword argument when creating the argument. New in atom 0.7.0, previously None was always a valid value.


If a Typed or Instance member is created with optional=False and no mean of creating a default value (no args, kwargs or factory), trying to access the member value before setting it will result in a ValueError.


Even though, generic aliases (i.e. list[int], introduced in PEP 585 ) are not proper types they can be used. Note however that just like isinstance(a, list[int]), a member Instance(list[int]) does not check the type of the items of a.

In some cases, the type is not accessible when the member is instantiated (because it will be created later in the same file for example), atom also provides ForwardTyped, ForwardInstance, ForwardSubclass. Those three members rather than taking a type or a tuple of type as argument, accept a callable taking no argument and returning the type(s) to use for validation.

class Leaf(Atom):

    node = ForwardTyped(lambda : Node)

class Node(Atom):

    parent = ForwardTyped(lambda : Node)

    leaves = List(Typed(Leaf))

In some cases, the same information may be conveniently represented either by a custom class or something simpler, like a tuple. One example of such a use case is a color: a color can be easily represented by the four components (red, green, blue, alpha) but in a library may be represented by a custom class. Atom provides the Coerced member to allow to enforce a particular type while also allowing seamless conversion from alternative representations.

The conversion can occur in two ways as illustrated below:

  • by calling the specified types on the provided value

  • by calling an alternative coercer function provided to the member

class Color(object):

    def __init__(self, components):,, self.alpha = components

def dict_to_color(color_dict):
    components = []
    for c in ('red', 'green', 'blue', 'alpha')
    return Color(components)

class MyAtom(Atom):

    color = Coerced(Color)

    color2 = Coerced(Color, coercer=dict_to_color)

Memory less members

Atom also provides two members that do not remember the value they are provided, but that can be used to fire notifications:

  • Event: this is a member to which each time a value is assigned to, a notification is fired. Additionally one can specify the type of value that are accepted. An alternative way to fire the notification is to call the object you get when accessing the member.

  • Signal: this member is similar to Qt signal. One cannot be assigned to it, however one can call it on instances, and when called the notifier will be called with the arguments and keyword arguments passed to the signal. Note that this is at odds with the general behavior of observers described in Notifications and observers.

    The example below illustrates how those members work:

    class MyAtom(Atom):
        s = Signal()
        e = Event()
        @observe('s', 'e')
        def print_value(self, change):
    obj = MyAtom()
    obj.e = 2


This last member is a bit special. It does not do anything by itself but can be used to copy the behaviors of another member. In addition, any observer attached to the delegator will also be attached to the delegate member.


The Property member is a special case and it will be discussed in details in The Property member.