Introducing denial: a marker for omitted values.

Python’s None is usually enough. Library authors, however, sometimes need to distinguish two different events: the caller explicitly passed None, or the caller passed nothing at all. If both cases use None, an API may ignore a requested reset as though no value was supplied, making defaults, partial updates, and layered settings ambiguous.

A local MISSING = object() works inside one module; dataclasses.MISSING and attrs.NOTHING belong to their own frameworks. Cooperating libraries instead need one framework-neutral marker with a public type.

denial is a small Python library that gives the second state a dedicated marker value, often called a sentinel. It provides InnerNone to mean “not supplied” inside library code, while ordinary None remains available as real input. The release is available on GitHub.

How InnerNone is used🔗

A common use case is an API where omission means “inherit the current value” while None means “clear it explicitly”:

from typing import Union

from denial import InnerNone, InnerNoneType

Timeout = Union[int, None, InnerNoneType]

def resolve_timeout(value: Timeout = InnerNone):
    if value is InnerNone:
        return inherited_timeout()
    return value  # None means explicitly disabled.

The same distinction appears in patch APIs, layered configuration, caches, and defaults. InnerNone is internal library state, not application data.

The API is deliberately small: InnerNone is the shared value, while InnerNoneType can be used in annotations and isinstance checks. Identity checks use value is InnerNone.

Code should compare the marker with is, as in the example. It is one shared marker for cooperating libraries, not a tool for creating many unrelated marker values.

When a sentinel factory is a better fit🔗

sentinel-value and sentinels provide factories and stronger serialization, making them better for projects that need several sentinels. denial standardizes one marker and public type for cooperating libraries.

PEP 661, the proposal for a standard sentinel API, has Deferred status. A future standard could provide a common API, while older Python versions and project-specific semantics would still require separate implementations.