graders/consistency.py

"""SecureCodeEnv - Consistency Grader v4 — clamped scores"""
from codegraph.graph import CodeGraph
from codegraph.extractor import extract_metadata
from graders.clamp import clamp

GOOD_PRACTICES = {
    "uses_type_hints":       0.15,
    "uses_docstrings":       0.15,
    "uses_try_catch":        0.10,
    "no_print_stmts":        0.10,
    "no_hardcoded_secrets":  0.10,
}


def grade_consistency(code: str, filename: str, graph: CodeGraph, step: int) -> dict:
    new_meta = extract_metadata(code, filename, step)
    conv     = new_meta.conventions

    if not graph.components:
        checks = {k: 1.0 if conv.get(k, False) else 0.0 for k in GOOD_PRACTICES}
        raw    = sum(checks.values()) / max(len(checks), 1)
        raw    = max(0.45, raw)   # floor so first step never crushes reward
        return {"score": clamp(raw), "checks": checks,
                "feedback": _first_feedback(raw, checks)}

    established = graph.conventions
    checks      = {}

    naming = established.get("naming")
    if naming and naming != "mixed" and new_meta.functions:
        fns = new_meta.functions
        if naming == "snake_case":
            correct = sum(1 for f in fns if "_" in f["name"] or f["name"].islower())
        else:
            correct = sum(1 for f in fns if f["name"] and f["name"][0].islower()
                          and any(c.isupper() for c in f["name"]))
        checks["naming_convention"] = correct / len(fns)

    if established.get("error_handling") == "try_catch":
        checks["error_handling"] = 1.0 if conv.get("uses_try_catch") else 0.3
    if established.get("uses_type_hints"):
        checks["type_hints"] = 1.0 if conv.get("uses_type_hints") else 0.4
    if established.get("uses_docstrings"):
        checks["docstrings"] = 1.0 if conv.get("uses_docstrings") else 0.5

    existing_no_print = all(c.conventions.get("no_print_stmts", True)
                            for c in graph.components.values())
    if existing_no_print:
        checks["no_print_drift"] = 1.0 if conv.get("no_print_stmts", True) else 0.3

    reuse_opp = reuse_taken = 0
    for name in graph.components:
        if name.lower() in code.lower():
            reuse_opp += 1
            if name in code:
                reuse_taken += 1
    if reuse_opp > 0:
        checks["component_reuse"] = reuse_taken / reuse_opp

    raw   = sum(checks.values()) / max(len(checks), 1) if checks else 0.7
    return {"score": clamp(raw), "checks": checks,
            "feedback": _feedback(raw, checks)}


def _first_feedback(score, checks):
    missing = [k for k, v in checks.items() if v == 0.0]
    if not missing:
        return f"Good conventions established ({score:.2f})"
    return f"Missing practices: {', '.join(missing)}"


def _feedback(score, checks):
    if score >= 0.85: return "Excellent consistency with codebase"
    failing = [k for k, v in checks.items() if isinstance(v, float) and v < 0.5]
    return f"Convention drift in: {', '.join(failing)}" if failing else f"Minor drift ({score:.2f})"