from __future__ import annotations import json import math from collections import defaultdict from sqlalchemy import and_, func, or_, select from app.modules.accounting.gstr2b_models import AccountingGSTR2BPurchase from app.modules.accounting.gstr2b_service import nature_lookup, review_purchase from app.modules.accounting.historical_learning_service import active_natures, ledger_mappings from app.modules.accounting.purchase_enrichment_models import AccountingPurchaseEnrichmentRecord from app.modules.accounting.ai_service import ai_assist_purchase as run_ai_assist_purchase, mark_review_outcome from app.modules.accounting.internal_model_service import predict_purchase, record_prediction_review VALID_STATUSES = {"all", "pending_analysis", "suggested", "review_required", "reviewed"} VALID_CONFIDENCE = {"all", "high", "medium", "low"} VALID_SOURCE = {"all", "gstr2b_only", "enriched"} PER_PAGE_CHOICES = {10, 25, 50, 100} def _safe_page(value): try: return max(1, int(value)) except Exception: return 1 def _safe_per_page(value): try: value = int(value) except Exception: value = 25 return value if value in PER_PAGE_CHOICES else 25 def _base_stmt(tenant_id, client_id): enriched = ( select( AccountingPurchaseEnrichmentRecord.gstr2b_purchase_id.label("purchase_id"), func.count(AccountingPurchaseEnrichmentRecord.id).label("enrichment_count"), ) .where( AccountingPurchaseEnrichmentRecord.gstr2b_purchase_id.is_not(None), AccountingPurchaseEnrichmentRecord.match_status == "linked", ) .group_by(AccountingPurchaseEnrichmentRecord.gstr2b_purchase_id) .subquery() ) stmt = ( select( AccountingGSTR2BPurchase, func.coalesce(enriched.c.enrichment_count, 0).label("enrichment_count"), ) .outerjoin(enriched, enriched.c.purchase_id == AccountingGSTR2BPurchase.id) .where( AccountingGSTR2BPurchase.tenant_id == tenant_id, AccountingGSTR2BPurchase.client_id == client_id, ) ) return stmt, enriched def review_queue(db, *, tenant_id, client_id, tally_guid="", status="all", confidence="all", source="all", supplier="", return_period="", page=1, per_page=25): status = status if status in VALID_STATUSES else "all" confidence = confidence if confidence in VALID_CONFIDENCE else "all" source = source if source in VALID_SOURCE else "all" page = _safe_page(page); per_page = _safe_per_page(per_page) stmt, enriched = _base_stmt(tenant_id, client_id) if tally_guid: stmt = stmt.where(AccountingGSTR2BPurchase.tally_guid == tally_guid) if status != "all": stmt = stmt.where(AccountingGSTR2BPurchase.review_status == status) if confidence == "high": stmt = stmt.where(AccountingGSTR2BPurchase.suggested_confidence >= 85) elif confidence == "medium": stmt = stmt.where(and_(AccountingGSTR2BPurchase.suggested_confidence >= 60, AccountingGSTR2BPurchase.suggested_confidence < 85)) elif confidence == "low": stmt = stmt.where(AccountingGSTR2BPurchase.suggested_confidence < 60) if source == "enriched": stmt = stmt.where(func.coalesce(enriched.c.enrichment_count, 0) > 0) elif source == "gstr2b_only": stmt = stmt.where(func.coalesce(enriched.c.enrichment_count, 0) == 0) supplier = (supplier or "").strip() if supplier: pattern = f"%{supplier}%" stmt = stmt.where(or_( AccountingGSTR2BPurchase.supplier_name.ilike(pattern), AccountingGSTR2BPurchase.supplier_gstin.ilike(pattern), AccountingGSTR2BPurchase.invoice_number.ilike(pattern), )) if return_period: stmt = stmt.where(AccountingGSTR2BPurchase.return_period == return_period.strip()) total = int(db.execute(select(func.count()).select_from(stmt.order_by(None).subquery())).scalar_one() or 0) pages = max(1, math.ceil(total / per_page)) if total else 1 page = min(page, pages) rows = list(db.execute(stmt.order_by(AccountingGSTR2BPurchase.invoice_date.desc(), AccountingGSTR2BPurchase.id.desc()).offset((page-1)*per_page).limit(per_page)).all()) return {"rows": rows, "total": total, "page": page, "per_page": per_page, "pages": pages} def review_counts(db, *, tenant_id, client_id): base = [AccountingGSTR2BPurchase.tenant_id == tenant_id, AccountingGSTR2BPurchase.client_id == client_id] total = int(db.execute(select(func.count(AccountingGSTR2BPurchase.id)).where(*base)).scalar_one() or 0) reviewed = int(db.execute(select(func.count(AccountingGSTR2BPurchase.id)).where(*base, AccountingGSTR2BPurchase.review_status == "reviewed")).scalar_one() or 0) high = int(db.execute(select(func.count(AccountingGSTR2BPurchase.id)).where(*base, AccountingGSTR2BPurchase.review_status != "reviewed", AccountingGSTR2BPurchase.suggested_nature_id.is_not(None), AccountingGSTR2BPurchase.suggested_confidence >= 85)).scalar_one() or 0) exceptions = int(db.execute(select(func.count(AccountingGSTR2BPurchase.id)).where(*base, AccountingGSTR2BPurchase.review_status != "reviewed", or_(AccountingGSTR2BPurchase.review_status == "review_required", AccountingGSTR2BPurchase.suggested_nature_id.is_(None), AccountingGSTR2BPurchase.suggested_confidence < 60))).scalar_one() or 0) enriched_ids = select(AccountingPurchaseEnrichmentRecord.gstr2b_purchase_id).where(AccountingPurchaseEnrichmentRecord.tenant_id == tenant_id, AccountingPurchaseEnrichmentRecord.client_id == client_id, AccountingPurchaseEnrichmentRecord.gstr2b_purchase_id.is_not(None), AccountingPurchaseEnrichmentRecord.match_status == "linked").distinct() enriched = int(db.execute(select(func.count(AccountingGSTR2BPurchase.id)).where(*base, AccountingGSTR2BPurchase.id.in_(enriched_ids))).scalar_one() or 0) return {"total": total, "reviewed": reviewed, "pending": max(0,total-reviewed), "high": high, "exceptions": exceptions, "enriched": enriched} def return_periods(db, *, tenant_id, client_id): return list(db.execute(select(AccountingGSTR2BPurchase.return_period).where(AccountingGSTR2BPurchase.tenant_id == tenant_id, AccountingGSTR2BPurchase.client_id == client_id, AccountingGSTR2BPurchase.return_period != "").distinct().order_by(AccountingGSTR2BPurchase.return_period.desc())).scalars().all()) def nature_maps_for_rows(db, rows): ids=[] for row, _count in rows: ids.extend([row.suggested_nature_id,row.final_nature_id]) return nature_lookup(db, ids) def explanations_for_rows(rows): result={} for row,_count in rows: try: result[row.id]=json.loads(row.suggestion_explanation_json or "[]") except Exception: result[row.id]=[] return result def mappings_by_nature(db, tenant_id, client_id, tally_guid=""): result=defaultdict(list) for row in ledger_mappings(db,tenant_id,client_id,tally_guid): result[int(row.nature_id)].append(row) for values in result.values(): values.sort(key=lambda x:(str(x.ledger_name or "").casefold(),x.id)) return dict(result) def _validate_choice(db, *, tenant_id, client_id, purchase, nature_id, ledger_name): valid_natures={int(n.id) for n in active_natures(db,tenant_id)} if int(nature_id) not in valid_natures: raise ValueError("Select an active accounting nature.") ledger_name=(ledger_name or "").strip() if not ledger_name: return allowed=ledger_mappings(db,tenant_id,client_id,purchase.tally_guid) if not any(m.ledger_name == ledger_name and int(m.nature_id)==int(nature_id) for m in allowed): raise ValueError("Selected Tally ledger is not mapped to the selected accounting nature for this client/company.") def review_one(db, *, tenant_id, client_id, purchase_id, final_nature_id, final_ledger_name, user_id): row=db.execute(select(AccountingGSTR2BPurchase).where(AccountingGSTR2BPurchase.id==purchase_id,AccountingGSTR2BPurchase.tenant_id==tenant_id,AccountingGSTR2BPurchase.client_id==client_id)).scalar_one_or_none() if not row: raise ValueError("Purchase record was not found.") _validate_choice(db,tenant_id=tenant_id,client_id=client_id,purchase=row,nature_id=final_nature_id,ledger_name=final_ledger_name) reviewed = review_purchase(db,row=row,final_nature_id=int(final_nature_id),final_ledger_name=(final_ledger_name or "").strip(),user_id=user_id) mark_review_outcome( db, tenant_id=tenant_id, source_type="gstr2b", source_record_id=reviewed.id, final_nature_id=reviewed.final_nature_id, final_ledger_name=reviewed.final_ledger_name, user_id=user_id, ) record_prediction_review( db, tenant_id=tenant_id, source_type="gstr2b", source_record_id=reviewed.id, final_nature_id=reviewed.final_nature_id, user_id=user_id, ) return reviewed def bulk_confirm(db, *, tenant_id, client_id, purchase_ids, user_id): ids=sorted({int(x) for x in purchase_ids if str(x).isdigit() and int(x)>0}) if not ids: raise ValueError("Select at least one high-confidence purchase.") rows=list(db.execute(select(AccountingGSTR2BPurchase).where(AccountingGSTR2BPurchase.tenant_id==tenant_id,AccountingGSTR2BPurchase.client_id==client_id,AccountingGSTR2BPurchase.id.in_(ids))).scalars().all()) by_id={r.id:r for r in rows} if len(by_id)!=len(ids): raise ValueError("One or more selected purchases are outside this client scope.") completed=0 for pid in ids: row=by_id[pid] if row.review_status=="reviewed": continue if not row.suggested_nature_id or int(row.suggested_confidence or 0)<85: raise ValueError(f"Invoice {row.invoice_number} is not eligible for high-confidence bulk confirmation.") _validate_choice(db,tenant_id=tenant_id,client_id=client_id,purchase=row,nature_id=row.suggested_nature_id,ledger_name=row.suggested_ledger_name or "") reviewed = review_purchase(db,row=row,final_nature_id=int(row.suggested_nature_id),final_ledger_name=row.suggested_ledger_name or "",user_id=user_id) mark_review_outcome( db, tenant_id=tenant_id, source_type="gstr2b", source_record_id=reviewed.id, final_nature_id=reviewed.final_nature_id, final_ledger_name=reviewed.final_ledger_name, user_id=user_id, ) record_prediction_review( db, tenant_id=tenant_id, source_type="gstr2b", source_record_id=reviewed.id, final_nature_id=reviewed.final_nature_id, user_id=user_id, ) completed+=1 return completed def ai_assist_one(db, *, tenant_id, client_id, purchase_id, user_id): row = db.execute(select(AccountingGSTR2BPurchase).where( AccountingGSTR2BPurchase.id == int(purchase_id), AccountingGSTR2BPurchase.tenant_id == int(tenant_id), AccountingGSTR2BPurchase.client_id == int(client_id), )).scalar_one_or_none() if not row: raise ValueError("Purchase record was not found.") return run_ai_assist_purchase(db, row=row, user_id=user_id) def internal_model_predict_one(db, *, tenant_id, client_id, purchase_id, force_shadow=True): row = db.execute(select(AccountingGSTR2BPurchase).where( AccountingGSTR2BPurchase.id == int(purchase_id), AccountingGSTR2BPurchase.tenant_id == int(tenant_id), AccountingGSTR2BPurchase.client_id == int(client_id), )).scalar_one_or_none() if not row: raise ValueError("Purchase record was not found.") return predict_purchase(db, row=row, force_shadow=force_shadow)