Open_Ultra_SAST/src/openultrasast/config.py

from __future__ import annotations

import json
import os
import tomllib
from dataclasses import asdict, dataclass
from pathlib import Path

DEFAULT_VECTOR_STORE = "json-local"


def load_dotenv(path: Path | None = None, *, force: bool = False) -> None:
    """Load KEY=VALUE lines from .env into os.environ. Stdlib only; never overrides."""
    if not force and (os.environ.get("PYTEST_CURRENT_TEST") or os.environ.get("OPENULTRASAST_SKIP_DOTENV")):
        return
    env_path = path if path is not None else Path(".env")
    if not env_path.is_file():
        return
    for raw in env_path.read_text().splitlines():
        line = raw.strip()
        if not line or line.startswith("#") or "=" not in line:
            continue
        key, _, value = line.partition("=")
        key = key.strip()
        value = value.strip().strip("'").strip('"')
        if key and key not in os.environ:
            os.environ[key] = value


@dataclass(frozen=True)
class ModelConfig:
    ranker: str | None = None
    hunter: str | None = None
    verifier: str | None = None
    patcher: str | None = None
    judge: str | None = None  # second, independent judgement before anything is published as proven
    chat_base_url: str | None = None  # chat endpoint; independent of the embedding endpoint
    chat_api_key_env: str | None = None  # environment variable holding that endpoint's key


@dataclass(frozen=True)
class EmbeddingConfig:
    model: str | None = None
    store: str = DEFAULT_VECTOR_STORE


@dataclass(frozen=True)
class SandboxConfig:
    network: bool = False
    workspace_readonly: bool = True
    memory_mb: int = 2048
    timeout_seconds: int = 300
    pids_limit: int = 512


@dataclass(frozen=True)
class DynamicConfig:
    enabled: bool = False
    network_scope: tuple[str, ...] = ()


@dataclass(frozen=True)
class EvidenceConfig:
    minimum_report_verified: str = "static_corroboration"
    minimum_exploit: str = "crash_reproduced"
    minimum_patch: str = "root_cause_explained"


@dataclass(frozen=True)
class StaticAnalysisConfig:
    sarif_paths: tuple[str, ...] = ()


@dataclass(frozen=True)
class ScoreConfig:
    k: float = 60.0
    min_score: int = 80
    block_severity_reachable: int = 5
    blocking: bool = False


@dataclass(frozen=True)
class RulesetConfig:
    # Emission floors that replace "every regex match becomes a finding". The
    # defaults (0.0) emit everything, keeping output byte-identical by default.
    min_emit_priority: float = 0.0
    min_emit_precision: float = 0.0


@dataclass(frozen=True)
class HarnessxConfig:
    # Optional agentic plane (the openultrasast[harnessx] extra). `provider`
    # selects the HarnessX LLM provider used by both the hunter pool and the
    # llm-judge verifier; the budget caps bound per-task cost and token spend.
    # Inert unless the extra is installed and a model is configured in [models].
    provider: str = "anthropic"
    max_cost_usd: float = 2.0
    token_threshold: int = 120_000


@dataclass(frozen=True)
class HardeningConfig:
    # Release-readiness controls. redact_secrets masks credentials in traces/reports;
    # max_findings caps the reported finding count for bounded CI runs (0 = unlimited).
    redact_secrets: bool = True
    max_findings: int = 0


@dataclass(frozen=True)
class FusionConfig:
    # Two-panel fusion adjudication for triggered findings (standard mode). Runs
    # deterministically by default; setting panel_model routes the panels through
    # the configured HarnessX provider. high_assurance forces fusion on every finding.
    enabled: bool = True
    panel_model: str | None = None
    decider_model: str | None = None
    high_assurance: bool = False


@dataclass(frozen=True)
class ComplexityConfig:
    # Stage-2 map size and hunter attention. Small defaults keep PR/nightly map work cheap.
    top_k: int = 20
    max_hunter_hotspots: int = 8


@dataclass(frozen=True)
class RegressConfig:
    # Stage-3 candidate cap and optional per-language sandbox image pins. Isolation limits stay on SandboxConfig.
    max_candidates: int = 5
    images: tuple[tuple[str, str], ...] = ()


@dataclass(frozen=True)
class VariantsConfig:
    # corpus-seeded-mechanisms: structural variant search in MAP (standard/deep only).
    enabled: bool = True
    max_mechanisms: int = 500


@dataclass(frozen=True)
class LearningConfig:
    # learning-harness: the classified detector set and its rounds.
    enabled: bool = False  # off until round zero has written the family configurations this reads
    families_path: str | None = None  # taxonomy file; the shipped ruleset when unset
    configs_dir: str = ".openultrasast/learning/configs"
    k_runs: int = 5  # runs per pair; a single run misses most real changes
    round_cost_cap_usd: float = 10.0
    minibatch: int = 8  # train pairs in a round's first stage
    detector_model: str | None = None
    judge_model: str | None = None
    cutoff_date: str | None = None  # rows fixed after this date are reported as a separate slice
    # Whether the detector model reasons before answering. Off by default because that is what round zero was
    # measured under, not because it is known to be better: the provider silently ignores `temperature` while
    # thinking, and thinking was disabled to make `temperature: 0` mean something. It did not — the measured
    # disagreement between runs comes from the agentic path, not the decoder — so which mode wins is a question
    # for a second baseline rather than for a default (Req 8.3).
    thinking: bool = False


@dataclass(frozen=True)
class ObligationsConfig:
    # authorization-obligations: the obligation checker in MAP (standard/deep only).
    enabled: bool = True
    min_siblings: int = 3
    policy_path: str = ".openultrasast/obligations.toml"


@dataclass(frozen=True)
class ResolvedConfig:
    models: ModelConfig = ModelConfig()
    embeddings: EmbeddingConfig = EmbeddingConfig()
    sandbox: SandboxConfig = SandboxConfig()
    dynamic: DynamicConfig = DynamicConfig()
    evidence: EvidenceConfig = EvidenceConfig()
    static_analysis: StaticAnalysisConfig = StaticAnalysisConfig()
    score: ScoreConfig = ScoreConfig()
    ruleset: RulesetConfig = RulesetConfig()
    harnessx: HarnessxConfig = HarnessxConfig()
    fusion: FusionConfig = FusionConfig()
    hardening: HardeningConfig = HardeningConfig()
    complexity: ComplexityConfig = ComplexityConfig()
    regress: RegressConfig = RegressConfig()
    variants: VariantsConfig = VariantsConfig()
    obligations: ObligationsConfig = ObligationsConfig()
    learning: LearningConfig = LearningConfig()
    runs_dir: str = ".openultrasast/runs"


def load_config(config_path: Path | None = None) -> ResolvedConfig:
    load_dotenv()
    data: dict[str, object] = {}
    if config_path is not None and config_path.exists():
        with config_path.open("rb") as handle:
            data = tomllib.load(handle)

    return ResolvedConfig(
        models=_load_models(data.get("models", {})),
        embeddings=_load_embeddings(data.get("embeddings", {})),
        sandbox=_load_sandbox(data.get("sandbox", {})),
        dynamic=_load_dynamic(data.get("dynamic", {})),
        evidence=_load_evidence(data.get("evidence", {})),
        static_analysis=_load_static_analysis(data.get("static_analysis", {})),
        score=_load_score(data.get("score", {})),
        ruleset=_load_ruleset_config(data.get("ruleset", {})),
        harnessx=_load_harnessx(data.get("harnessx", {})),
        fusion=_load_fusion(data.get("fusion", {})),
        hardening=_load_hardening(data.get("hardening", {})),
        complexity=_load_complexity(data.get("complexity", {})),
        regress=_load_regress(data.get("regress", {})),
        variants=_load_variants(data.get("variants", {})),
        obligations=_load_obligations(data.get("obligations", {})),
        learning=_load_learning(data.get("learning", {})),
        runs_dir=os.environ.get("OPENULTRASAST_RUNS_DIR", ".openultrasast/runs"),
    )


def write_resolved_config(config: ResolvedConfig, path: Path) -> None:
    path.parent.mkdir(parents=True, exist_ok=True)
    path.write_text(json.dumps(asdict(config), indent=2, sort_keys=True) + "\n")


def _section(value: object) -> dict[str, object]:
    return value if isinstance(value, dict) else {}


def _string(value: object) -> str | None:
    return value if isinstance(value, str) and value else None


def _load_models(value: object) -> ModelConfig:
    data = _section(value)
    return ModelConfig(
        ranker=_string(data.get("ranker")),
        hunter=_string(data.get("hunter")),
        verifier=_string(data.get("verifier")),
        patcher=_string(data.get("patcher")),
        judge=_string(data.get("judge")),
        chat_base_url=_string(data.get("chat_base_url")),
        chat_api_key_env=_string(data.get("chat_api_key_env")),
    )


def _load_embeddings(value: object) -> EmbeddingConfig:
    data = _section(value)
    return EmbeddingConfig(
        model=_string(data.get("model")) or _string(os.environ.get("OPENROUTER_EMBEDDING_MODEL")),
        store=_string(data.get("store")) or DEFAULT_VECTOR_STORE,
    )


def _load_sandbox(value: object) -> SandboxConfig:
    data = _section(value)
    return SandboxConfig(
        network=bool(data.get("network", False)),
        workspace_readonly=bool(data.get("workspace_readonly", True)),
        memory_mb=_int_value(data.get("memory_mb"), 2048),
        timeout_seconds=_int_value(data.get("timeout_seconds"), 300),
        pids_limit=_int_value(data.get("pids_limit"), 512),
    )


def _load_dynamic(value: object) -> DynamicConfig:
    data = _section(value)
    scope = data.get("network_scope", [])
    if not isinstance(scope, list):
        scope = []
    return DynamicConfig(enabled=bool(data.get("enabled", False)), network_scope=tuple(str(item) for item in scope))


def _load_evidence(value: object) -> EvidenceConfig:
    data = _section(value)
    return EvidenceConfig(
        minimum_report_verified=str(data.get("minimum_report_verified", "static_corroboration")),
        minimum_exploit=str(data.get("minimum_exploit", "crash_reproduced")),
        minimum_patch=str(data.get("minimum_patch", "root_cause_explained")),
    )


def _int_value(value: object, default: int) -> int:
    if isinstance(value, bool):
        return default
    if isinstance(value, int | str | bytes | bytearray):
        return int(value)
    return default


def _load_static_analysis(value: object) -> StaticAnalysisConfig:
    data = _section(value)
    paths = data.get("sarif_paths", [])
    if not isinstance(paths, list):
        paths = []
    return StaticAnalysisConfig(sarif_paths=tuple(str(item) for item in paths))


def _load_score(value: object) -> ScoreConfig:
    data = _section(value)
    return ScoreConfig(
        k=_float_value(data.get("k"), 60.0),
        min_score=_int_value(data.get("min_score"), 80),
        block_severity_reachable=_int_value(data.get("block_severity_reachable"), 5),
        blocking=bool(data.get("blocking", False)),
    )


def _load_ruleset_config(value: object) -> RulesetConfig:
    data = _section(value)
    return RulesetConfig(
        min_emit_priority=_float_value(data.get("min_emit_priority"), 0.0),
        min_emit_precision=_float_value(data.get("min_emit_precision"), 0.0),
    )


def _load_harnessx(value: object) -> HarnessxConfig:
    data = _section(value)
    return HarnessxConfig(
        provider=_string(data.get("provider")) or "anthropic",
        max_cost_usd=_float_value(data.get("max_cost_usd"), 2.0),
        token_threshold=_int_value(data.get("token_threshold"), 120_000),
    )


def _load_hardening(value: object) -> HardeningConfig:
    data = _section(value)
    return HardeningConfig(
        redact_secrets=bool(data.get("redact_secrets", True)),
        max_findings=_int_value(data.get("max_findings"), 0),
    )


def _load_fusion(value: object) -> FusionConfig:
    data = _section(value)
    return FusionConfig(
        enabled=bool(data.get("enabled", True)),
        panel_model=_string(data.get("panel_model")),
        decider_model=_string(data.get("decider_model")),
        high_assurance=bool(data.get("high_assurance", False)),
    )


def _load_complexity(value: object) -> ComplexityConfig:
    data = _section(value)
    return ComplexityConfig(
        top_k=_int_value(data.get("top_k"), 20),
        max_hunter_hotspots=_int_value(data.get("max_hunter_hotspots"), 8),
    )


def _load_regress(value: object) -> RegressConfig:
    data = _section(value)
    return RegressConfig(
        max_candidates=_int_value(data.get("max_candidates"), 5),
        images=_string_map(data.get("images")),
    )


def _string_map(value: object) -> tuple[tuple[str, str], ...]:
    data = _section(value)
    images: list[tuple[str, str]] = []
    for key, item in data.items():
        pinned = _string(item)
        if pinned is not None:
            images.append((str(key), pinned))
    return tuple(sorted(images))


def _float_value(value: object, default: float) -> float:
    if isinstance(value, bool):
        return default
    if isinstance(value, int | float):
        return float(value)
    if isinstance(value, str):
        try:
            return float(value)
        except ValueError:
            return default
    return default


def _load_variants(value: object) -> VariantsConfig:
    data = _section(value)
    return VariantsConfig(
        enabled=bool(data.get("enabled", True)),
        max_mechanisms=_int_value(data.get("max_mechanisms"), 500),
    )


def _load_learning(value: object) -> LearningConfig:
    data = _section(value)
    return LearningConfig(
        enabled=bool(data.get("enabled", LearningConfig.enabled)),
        families_path=_string(data.get("families_path")),
        configs_dir=_string(data.get("configs_dir")) or LearningConfig.configs_dir,
        k_runs=_int_value(data.get("k_runs"), LearningConfig.k_runs),
        round_cost_cap_usd=_float_value(data.get("round_cost_cap_usd"), LearningConfig.round_cost_cap_usd),
        minibatch=_int_value(data.get("minibatch"), LearningConfig.minibatch),
        detector_model=_string(data.get("detector_model")),
        judge_model=_string(data.get("judge_model")),
        cutoff_date=_string(data.get("cutoff_date")),
        thinking=bool(data.get("thinking", False)),
    )


def _load_obligations(value: object) -> ObligationsConfig:
    data = _section(value)
    return ObligationsConfig(
        enabled=bool(data.get("enabled", True)),
        min_siblings=_int_value(data.get("min_siblings"), 3),
        policy_path=str(data.get("policy_path", ".openultrasast/obligations.toml")),
    )