Lazy Evaluation

spock supports lazy evaluation for both dependencies between classes and class inheritance.

Lazy Class Dependencies

When we create a new spock class that depends on another spock class we can be 'lazy' and pass only the top-level class(es) to the SpockBuilder instead of all the dependent classes. This tends to make your code much less verbose when there are many @spock decorated classes. We do this by setting the lazy flag to True within the SpockBuilder. This will tell spock to look for any missing references to @spock decorated classes within sys.modules["spock"].backend.config and add them to the underlying list of classes within graph construction. For instance,

from spock import spockfrom spock import SpockBuilder
@spockclass NestedStuffDefault:    away: str = "arsenal"    goals: int = 0

@spockclass Maybe:    r: int = 4    t: float = 1.9    nested_no_conf_def: NestedStuffDefault = NestedStuffDefault

# Set lazy to True to make sure spock lazily finds @spock decorated classes # Not we only pass the Maybe class here instead of both Maybe and NestedStuffDefaultconfig = SpockBuilder(Maybe, desc='Lazy Example', lazy=True).generate()

The above is equivalent to passing both Maybe and NestedStuffDefault to the SpockBuilder.

Lazy Class Inheritance

When we create a new spock class that inherits from another spock class we can be 'lazy' and leave off the @spock decorator for any parent classes. We do this by setting the dynamic flag within the @spock decorator to True. With this flag, spock will traverse the MRO and automatically cast parent classes to the equivalent of a @spock decorated class. With lazy inheritance you only need to pass the child class to the SpockBuilder instead of all parents and children. Additionally, set the lazy flag to True within the SpockBuilder to lazily find and return the parent classes to the generated Spockspace. For instance:

from spock import spockfrom spock import SpockBuilderfrom typing import Optional
class OptimizerConfig:    lr: float = 0.01    n_epochs: int = 2    grad_clip: Optional[float]

class DataConfig:    batch_size: int = 2

# Set dynamic=True to allow for lazy inheritance@spock(dynamic=True)class SGDConfig(OptimizerConfig, DataConfig):    weight_decay: float    momentum: float    nesterov: bool
# Set lazy to True to make sure the parent classes are returned from the generate callconfig = SpockBuilder(SGDConfig, desc='Lazy Example', lazy=True).generate()

The above is equivalent to decorating each class with the @spock decorator and additionally passing DataConfig and OptimizerConfig to the SpockBuilder.