Open_Ultra_SAST/src/openultrasast/core/provenance.py

"""Legacy stamps and strict v2 input/decision provenance contracts."""

from __future__ import annotations

import hashlib
import json
import re
from collections.abc import Mapping
from dataclasses import dataclass, fields
from enum import Enum

from openultrasast.core import canonical
from openultrasast.core.canonical import ContentRef, ContentResolver

NAMED_INPUTS_SCHEMA = "openultrasast.detection-core.named-inputs.v2"
PROVENANCE_POLICY_SCHEMA = "openultrasast.detection-core.provenance-policy.v2"
_DIGEST = re.compile(r"[0-9a-f]{64}")


class ProvenanceMismatch(ValueError):
    """A legacy recorded digest does not match its re-derived identity."""


_LEGACY_DIGEST_FIELDS = (
    "analyzer_revision",
    "fact_set_digest",
    "acquisition_record_digest",
    "configuration_digest",
    "population_digest",
    "input_digest",
)


def _input_digest(change_id: str, population_digest: str) -> str:
    payload = json.dumps([population_digest, change_id], sort_keys=True, separators=(",", ":"))
    return hashlib.sha256(payload.encode()).hexdigest()[:12]


def stamp(
    *,
    analyzer_revision: str,
    fact_set_digest: str,
    acquisition_record_digest: str,
    configuration_digest: str,
    population_digest: str,
    change_id: str,
) -> dict[str, str]:
    """Preserve the v1 provenance stamp during migration."""
    return {
        "analyzer_revision": analyzer_revision,
        "fact_set_digest": fact_set_digest,
        "acquisition_record_digest": acquisition_record_digest,
        "configuration_digest": configuration_digest,
        "population_digest": population_digest,
        "input_digest": _input_digest(change_id, population_digest),
    }


def verify_stamp(recorded: Mapping[str, object], rederived: Mapping[str, object]) -> None:
    """Preserve v1 refusal when a legacy digest cannot be re-derived."""
    for field_name in _LEGACY_DIGEST_FIELDS:
        if recorded.get(field_name) != rederived.get(field_name):
            raise ProvenanceMismatch(f"{field_name}: recorded {recorded.get(field_name)!r} != rederived {rederived.get(field_name)!r}")


class ProvenanceValidationError(ValueError):
    """A provenance identity cannot be reproduced from its named inputs."""


class ProvenanceInputRole(Enum):
    ANALYZER = "analyzer"
    FAMILY_CONTRACT = "family_contract"
    FACT_SET = "fact_set"
    ACQUISITION = "acquisition"
    CONFIGURATION = "configuration"
    POPULATION = "population"
    CHANGE_INPUT = "change_input"


def _require_digest(value: object, field_name: str) -> None:
    if not isinstance(value, str) or _DIGEST.fullmatch(value) is None:
        raise canonical.DigestFormatError(f"{field_name} must be a full lowercase SHA-256 digest")


def _require_text(value: object, field_name: str) -> None:
    if not isinstance(value, str) or not value:
        raise ValueError(f"{field_name} must be nonempty text")


def _require_maximum(value: object) -> None:
    if isinstance(value, bool) or not isinstance(value, int) or value <= 0:
        raise ValueError("maximum_input_bytes must be a positive integer")


def _require_stores(stores: object) -> None:
    if not isinstance(stores, tuple) or not stores:
        raise ValueError("trusted_store_identities must be a nonempty tuple")
    for store in stores:
        _require_text(store, "trusted_store_identity")
    if len(set(stores)) != len(stores):
        raise ValueError("trusted_store_identities must be unique")


@dataclass(frozen=True)
class ProvenanceStamp:
    analyzer_revision: str
    family_contract_digest: str
    fact_set_digest: str
    acquisition_record_digest: str
    configuration_digest: str
    population_digest: str
    change_input_digest: str

    def __post_init__(self) -> None:
        for item in fields(self):
            _require_digest(getattr(self, item.name), item.name)


@dataclass(frozen=True)
class ProvenanceTrustPolicy:
    trusted_store_identities: tuple[str, ...]
    maximum_input_bytes: int
    policy_digest: str

    def __post_init__(self) -> None:
        _require_stores(self.trusted_store_identities)
        _require_maximum(self.maximum_input_bytes)
        _require_digest(self.policy_digest, "policy_digest")


@dataclass(frozen=True)
class NamedInput:
    role: ProvenanceInputRole
    identity: str
    content: ContentRef

    def __post_init__(self) -> None:
        _require_digest(self.identity, "named input identity")
        if not isinstance(self.content, ContentRef):
            raise TypeError("named input content must be a ContentRef")


@dataclass(frozen=True)
class NamedInputs:
    inputs: tuple[NamedInput, ...]
    manifest_digest: str

    def __post_init__(self) -> None:
        if not isinstance(self.inputs, tuple) or not all(isinstance(item, NamedInput) for item in self.inputs):
            raise TypeError("named inputs must be a tuple of NamedInput values")
        _require_digest(self.manifest_digest, "manifest_digest")


@dataclass(frozen=True)
class ModelCallIdentity:
    provider: str
    model: str
    program_digest: str
    prompt_digest: str
    configuration_digest: str
    adapter_revision: str

    def __post_init__(self) -> None:
        for field_name in ("provider", "model", "adapter_revision"):
            _require_text(getattr(self, field_name), field_name)
        for field_name in ("program_digest", "prompt_digest", "configuration_digest"):
            _require_digest(getattr(self, field_name), field_name)


@dataclass(frozen=True)
class DecisionProvenance:
    model_identity: ModelCallIdentity | None
    model_result_digest: str | None
    model_declaration_digest: str | None
    proof_record_digest: str | None
    proof_producer_digest: str | None
    lineage_manifest_digest: str | None

    def __post_init__(self) -> None:
        if self.model_identity is not None and not isinstance(self.model_identity, ModelCallIdentity):
            raise TypeError("model_identity must be a ModelCallIdentity")
        for item in fields(self)[1:]:
            value = getattr(self, item.name)
            if value is not None:
                _require_digest(value, item.name)


def _content_ref_value(reference: ContentRef) -> canonical.JsonValue:
    return {
        "artifact_id": reference.artifact_id,
        "content_digest": reference.content_digest,
        "media_type": reference.media_type,
        "byte_length": reference.byte_length,
        "retention_policy_digest": reference.retention_policy_digest,
        "storage_receipt_digest": reference.storage_receipt_digest,
    }


def _input_value(item: NamedInput) -> canonical.JsonValue:
    if not isinstance(item.role, ProvenanceInputRole):
        raise ProvenanceValidationError(f"unknown named input role: {item.role!r}")
    return {"role": item.role.value, "identity": item.identity, "content": _content_ref_value(item.content)}


def derive_named_inputs_digest(inputs: tuple[NamedInput, ...]) -> str:
    """Derive an order-independent identity over every named input reference."""
    ordered = tuple(sorted(inputs, key=lambda item: item.role.value))
    payload: canonical.JsonValue = {
        "schema_version": NAMED_INPUTS_SCHEMA,
        "inputs": tuple(_input_value(item) for item in ordered),
    }
    return canonical.content_digest(payload)


def derive_trust_policy_digest(trusted_store_identities: tuple[str, ...], maximum_input_bytes: int) -> str:
    """Derive the trust and size policy identity."""
    _require_stores(trusted_store_identities)
    _require_maximum(maximum_input_bytes)
    payload: canonical.JsonValue = {
        "schema_version": PROVENANCE_POLICY_SCHEMA,
        "trusted_store_identities": tuple(sorted(trusted_store_identities)),
        "maximum_input_bytes": maximum_input_bytes,
    }
    return canonical.content_digest(payload)


_STAMP_FIELD = {
    ProvenanceInputRole.ANALYZER: "analyzer_revision",
    ProvenanceInputRole.FAMILY_CONTRACT: "family_contract_digest",
    ProvenanceInputRole.FACT_SET: "fact_set_digest",
    ProvenanceInputRole.ACQUISITION: "acquisition_record_digest",
    ProvenanceInputRole.CONFIGURATION: "configuration_digest",
    ProvenanceInputRole.POPULATION: "population_digest",
    ProvenanceInputRole.CHANGE_INPUT: "change_input_digest",
}


def _index_inputs(inputs: tuple[NamedInput, ...]) -> dict[ProvenanceInputRole, NamedInput]:
    indexed: dict[ProvenanceInputRole, NamedInput] = {}
    for item in inputs:
        if not isinstance(item.role, ProvenanceInputRole):
            raise ProvenanceValidationError(f"unknown named input role: {item.role!r}")
        if item.role in indexed:
            raise ProvenanceValidationError(f"duplicate named input role: {item.role.value}")
        indexed[item.role] = item
    missing = set(ProvenanceInputRole) - set(indexed)
    if missing:
        names = ", ".join(sorted(role.value for role in missing))
        raise ProvenanceValidationError(f"missing named input roles: {names}")
    return indexed


def _validate_policy(trust: ProvenanceTrustPolicy) -> None:
    expected = derive_trust_policy_digest(trust.trusted_store_identities, trust.maximum_input_bytes)
    if trust.policy_digest != expected:
        raise ProvenanceValidationError("provenance trust policy digest cannot be re-derived")


def _validate_manifest(named_inputs: NamedInputs) -> None:
    expected = derive_named_inputs_digest(named_inputs.inputs)
    if named_inputs.manifest_digest != expected:
        raise ProvenanceValidationError("named input manifest digest cannot be re-derived")


def validate_provenance(
    stamp: ProvenanceStamp,
    named_inputs: NamedInputs,
    resolver: ContentResolver,
    trust: ProvenanceTrustPolicy,
) -> None:
    """Resolve every closed-role input and reproduce each immutable stamp identity."""
    _validate_policy(trust)
    indexed = _index_inputs(named_inputs.inputs)
    _validate_manifest(named_inputs)
    mismatches: list[str] = []
    for role in ProvenanceInputRole:
        item = indexed[role]
        canonical.resolve_content(
            item.content,
            resolver,
            trusted_store_identities=trust.trusted_store_identities,
            maximum_bytes=trust.maximum_input_bytes,
        )
        if item.identity != item.content.content_digest:
            raise ProvenanceValidationError(f"{role.value} named identity does not match resolved content")
        field_name = _STAMP_FIELD[role]
        if getattr(stamp, field_name) != item.identity:
            mismatches.append(field_name)
    if mismatches:
        raise ProvenanceValidationError(f"stamp identities do not match named inputs: {', '.join(mismatches)}")


__all__ = [
    "DecisionProvenance",
    "ModelCallIdentity",
    "NAMED_INPUTS_SCHEMA",
    "NamedInput",
    "NamedInputs",
    "PROVENANCE_POLICY_SCHEMA",
    "ProvenanceInputRole",
    "ProvenanceMismatch",
    "ProvenanceStamp",
    "ProvenanceTrustPolicy",
    "ProvenanceValidationError",
    "derive_named_inputs_digest",
    "derive_trust_policy_digest",
    "stamp",
    "validate_provenance",
    "verify_stamp",
]