Interface metadata

The amaranth.lib.meta module provides a way to annotate objects in an Amaranth design and exchange these annotations with external tools in a standardized format.

Introduction

Many Amaranth designs stay entirely within the Amaranth ecosystem, using the facilities it provides to define, test, and build hardware. In this case, the design is available for exploration using Python code, and metadata is not necessary. However, if an Amaranth design needs to fit into an existing ecosystem, or, conversely, to integrate components developed for another ecosystem, metadata can be used to exchange structured information about the design.

Consider a simple component:

class Adder(wiring.Component):
    a: In(unsigned(32))
    b: In(unsigned(32))
    o: Out(unsigned(33))

    def elaborate(self, platform):
        m = Module()
        m.d.comb += self.o.eq(self.a + self.b)
        return m

While it can be easily converted to Verilog, external tools will find the interface of the resulting module opaque unless they parse its Verilog source (a difficult and unrewarding task), or are provided with a description of it. Components can describe their signature with JSON-based metadata:

>>> adder = Adder()
>>> adder.metadata 
<amaranth.lib.wiring.ComponentMetadata for ...Adder object at ...>
>>> adder.metadata.as_json() 
{
    'interface': {
        'members': {
            'a': {
                'type': 'port',
                'name': 'a',
                'dir': 'in',
                'width': 32,
                'signed': False,
                'init': '0'
            },
            'b': {
                'type': 'port',
                'name': 'b',
                'dir': 'in',
                'width': 32,
                'signed': False,
                'init': '0'
            },
            'o': {
                'type': 'port',
                'name': 'o',
                'dir': 'out',
                'width': 33,
                'signed': False,
                'init': '0'
            }
        },
        'annotations': {}
    }
}

All metadata in Amaranth must adhere to a schema in the JSON Schema language, which is integral to its definition, and can be used to validate the generated JSON:

>>> wiring.ComponentMetadata.validate(adder.metadata.as_json())

The built-in component metadata can be extended to provide arbitrary information about an interface through user-defined annotations. For example, a memory bus interface could provide the layout of any memory-mapped peripherals accessible through that bus.

Defining annotations

Consider a simple control and status register (CSR) bus that provides the memory layout of the accessible registers via an annotation:

class CSRLayoutAnnotation(meta.Annotation):
    schema = {
        "$schema": "https://json-schema.org/draft/2020-12/schema",
        "$id": "https://amaranth-lang.org/schema/example/0/csr-layout.json",
        "type": "object",
        "properties": {
            "registers": {
                "type": "object",
                "patternProperties": {
                    "^.+$": {
                        "type": "integer",
                        "minimum": 0,
                    },
                },
            },
        },
        "requiredProperties": [
            "registers",
        ],
    }

    def __init__(self, origin):
        self._origin = origin

    @property
    def origin(self):
        return self._origin

    def as_json(self):
        instance = {
            "registers": self.origin.registers,
        }
        # Validating the value returned by `as_json()` ensures its conformance.
        self.validate(instance)
        return instance


class CSRSignature(wiring.Signature):
    def __init__(self):
        super().__init__({
            "addr":     Out(16),
            "w_en":     Out(1),
            "w_data":   Out(32),
            "r_en":     Out(1),
            "r_data":   In(32),
        })

    def annotations(self, obj, /):
        # Unfortunately `super()` cannot be used in `wiring.Signature` subclasses;
        # instead, use a direct call to a superclass method. In this case that is
        # `wiring.Signature` itself, but in a more complex class hierarchy it could
        # be different.
        return wiring.Signature.annotations(self, obj) + (CSRLayoutAnnotation(obj),)

A component that embeds a few CSR registers would define their addresses:

class MyPeripheral(wiring.Component):
    csr_bus: In(CSRSignature())

    def __init__(self):
        super().__init__()
        self.csr_bus.registers = {
            "control": 0x0000,
            "status":  0x0004,
            "data":    0x0008,
        }

>>> peripheral = MyPeripheral()
>>> peripheral.metadata.as_json() 
{
    'interface': {
        'members': {
            'csr_bus': {
                'type': 'interface',
                'members': {
                    'addr': {
                        'type': 'port',
                        'name': 'csr_bus__addr',
                        'dir': 'in',
                        'width': 16,
                        'signed': False,
                        'init': '0'
                    },
                    'w_en': {
                        'type': 'port',
                        'name': 'csr_bus__w_en',
                        'dir': 'in',
                        'width': 1,
                        'signed': False,
                        'init': '0'
                    },
                    'w_data': {
                        'type': 'port',
                        'name': 'csr_bus__w_data',
                        'dir': 'in',
                        'width': 32,
                        'signed': False,
                        'init': '0'
                    },
                    'r_en': {
                        'type': 'port',
                        'name': 'csr_bus__r_en',
                        'dir': 'in',
                        'width': 1,
                        'signed': False,
                        'init': '0'
                    },
                    'r_data': {
                        'type': 'port',
                        'name': 'csr_bus__r_data',
                        'dir': 'out',
                        'width': 32,
                        'signed': False,
                        'init': '0'
                    },
                },
                'annotations': {
                    'https://amaranth-lang.org/schema/example/0/csr-layout.json': {
                        'registers': {
                            'control': 0,
                            'status':  4,
                            'data':    8
                        }
                    }
                }
            }
        },
        'annotations': {}
    }
}

Identifying schemas

An Annotation schema must have a "$id" property, whose value is a URL that serves as its globally unique identifier. The suggested format of this URL is:

<protocol>://<domain>/schema/<package>/<version>/<path>.json

where:

<domain> is a domain name registered to the person or entity defining the annotation;

<package> is the name of the Python package providing the Annotation subclass;

<version> is the version of that package;

<path> is a non-empty string specific to the annotation.

Note

Annotations used in the Amaranth project packages are published under https://amaranth-lang.org/schema/ according to this URL format, and are covered by the usual compatibility commitment.

Other projects that define additional Amaranth annotations are encouraged, but not required, to make their schemas publicly accessible; the only requirement is for the URL to be globally unique.

Reference

exception amaranth.lib.meta.InvalidSchema: Exception raised when a subclass of Annotation is defined with a non-conformant schema.

exception amaranth.lib.meta.InvalidAnnotation: Exception raised by Annotation.validate() when the JSON representation of an annotation does not conform to its schema.

class amaranth.lib.meta.Annotation

Interface annotation.

Annotations are containers for metadata that can be retrieved from an interface object using the Signature.annotations method.

Annotations have a JSON representation whose structure is defined by the JSON Schema language.

classmethod __init_subclass__()

Defining a subclass of Annotation causes its schema to be validated.

Raises:

InvalidSchema – If schema doesn’t conform to the 2020-12 draft of JSON Schema.
InvalidSchema – If schema doesn’t have a “$id” keyword at its root. This requirement is specific to Annotation schemas.

schema = { "$id": "...", ... }

Schema of this annotation, expressed in the JSON Schema language.

Subclasses of Annotation must define this class attribute.

Type:: dict

abstract property origin

Python object described by this Annotation instance.

Subclasses of Annotation must implement this property.

abstract as_json()

Convert to a JSON representation.

Subclasses of Annotation must implement this method.

JSON representation returned by this method must adhere to schema and pass validation by validate().

Returns:: JSON representation of this annotation, expressed in Python primitive types (dict, list, str, int, bool).
Return type:: dict

classmethod validate(instance)

Validate a JSON representation against schema.

Parameters:: instance (dict) – JSON representation to validate, either previously returned by as_json() or retrieved from an external source.
Raises:: InvalidAnnotation – If instance doesn’t conform to schema.