contracts-flask/deploy-lucee/services/grouping.py

"""
Grouping service — match classified documents into contract groups.

Архитектурное решение (Opus, Q4 + Q6):
- Двухпроходный гибрид: LLM извлекает строки (classify.py), Python нормализует и группирует.
- Нормализация номеров: uppercase + только буквы/цифры.
  "МЭС-123-2024" == "МЭС 123/2024" после нормализации.
- Группировка на бэкенде (не на фронте): Python regex/unicode надёжнее JS.
- apply_groups(): создаёт contracts + supplements с авто-порядком по дате.
"""
import re
from db import documents as db_docs
from db import contracts as db_contracts
from db import supplements as db_supplements


def normalize_number(num):
    """
    Нормализация номера договора для сравнения.
    Убирает всё кроме букв и цифр, приводит к uppercase.
    Пример: "МЭС-123-2024" → "МЭС1232024", "МЭС 123/2024" → "МЭС1232024".
    """
    if not num:
        return ""
    return re.sub(r"[^A-Z0-9А-Я]", "", num.upper())


def group_documents(batch_id):
    """
    Сгруппировать классифицированные документы по контрактам.

    Алгоритм:
    1. Отделить contract от supplement/specification
    2. Каждый contract → якорь группы
    3. Для каждого supplement: найти contract по parent_number (нормализованный)
    4. Оставшиеся supplement/spec → виртуальные группы по parent_number/own_number
    5. Совсем без номеров → группа "__unresolved__"
    6. Внутри группы сортировка по doc_date

    Возвращает {ok, groups: [{contract_number, counterparty, documents: [...]}], total_docs}.
    """
    docs = db_docs.list_by_batch(batch_id)
    classified = [d for d in docs if d.get("classify_status") == "classified"]

    # Separate contracts and supplements
    contracts_list = []
    supplements_list = []

    for d in classified:
        if d.get("doc_type") == "contract":
            contracts_list.append(d)
        else:
            supplements_list.append(d)

    groups = []

    # Each contract becomes a group
    for c in contracts_list:
        group = {
            "contract_number": c.get("own_number") or c.get("filename", ""),
            "counterparty": c.get("counterparty") or "",
            "documents": [c],
        }
        # Find supplements matching this contract
        c_norm = normalize_number(c.get("own_number"))
        for s in supplements_list:
            parent = normalize_number(s.get("parent_number") or "")
            own = normalize_number(s.get("own_number") or "")
            if parent == c_norm or own == c_norm:
                if s not in group["documents"]:
                    group["documents"].append(s)

        # Sort by date
        group["documents"].sort(key=lambda x: x.get("doc_date") or "")

        # Remove matched supplements from the pool
        for s in group["documents"]:
            if s in supplements_list:
                supplements_list.remove(s)

        groups.append(group)

    # ── Virtual groups: unmatched supplements grouped by number ──────────
    # Group remaining supplements/specs by normalized parent_number (priority) or own_number
    virtual = {}
    for s in supplements_list:
        num = normalize_number(s.get("parent_number") or s.get("own_number") or "")
        if not num:
            continue  # no number → stays in supplements_list for unresolved
        if num not in virtual:
            # Use counterparty from first doc in group
            cp = s.get("counterparty") or ""
            virtual[num] = {
                "contract_number": s.get("parent_number") or s.get("own_number") or "?",
                "counterparty": cp,
                "documents": [],
            }
        virtual[num]["documents"].append(s)
        # Update counterparty if current doc has a better one
        if not virtual[num]["counterparty"] and s.get("counterparty"):
            virtual[num]["counterparty"] = s.get("counterparty")

    for vnum, vgroup in virtual.items():
        vgroup["documents"].sort(key=lambda x: x.get("doc_date") or "")
        groups.append(vgroup)
        # Remove grouped docs from supplements_list
        for s in vgroup["documents"]:
            supplements_list.remove(s)

    # ── Unresolved: everything left (no number, failed, pending, other) ──
    unmatched = [s for s in supplements_list]  # remaining after virtual grouping
    unmatched += [d for d in docs if d.get("classify_status") != "classified"]

    if unmatched:
        groups.append({
            "contract_number": "__unresolved__",
            "counterparty": "",
            "documents": unmatched,
        })

    return {"ok": True, "groups": groups, "total_docs": len(docs)}


def apply_groups(batch_id, groups_data):
    """
    Применить подтверждённые группы: создать contracts + supplements.

    Вызывается из POST /api/apply-groups.
    Для каждой группы (кроме __unresolved__):
    1. Создать запись в contracts (number, client)
    2. Для каждого документа создать supplement (type='initial' для первого, 'additional' для остальных)
    3. Порядок supplements соответствует порядку документов в группе (сортировка по дате уже сделана)

    Возвращает {ok, created: количество созданных supplements}.
    """
    created = 0
    contract_ids = []
    for g in groups_data:
        contract_number = g.get("contract_number", "")
        if contract_number == "__unresolved__":
            continue
        counterparty = g.get("counterparty", "")
        docs = g.get("documents", [])

        try:
            c = db_contracts.insert(contract_number, counterparty)
            contract_id = c["id"]
            contract_ids.append(contract_id)

            for i, d in enumerate(docs):
                doc_id = d.get("id")
                if not doc_id:
                    continue
                supp_type = "initial" if i == 0 else "additional"
                db_supplements.insert(contract_id, doc_id, supp_type)
                created += 1
        except Exception as e:
            return {"ok": False, "error": f"apply_groups failed at {contract_number}: {e}"}

    return {"ok": True, "created": created, "contract_ids": contract_ids}