import camelot
import pandas as pd
import re
import csv

def clean_amount(val):
    """Removes '=' and whitespace from amount strings."""
    if not val: return ""
    return val.replace('=', '').strip()

def extract_account(text):
    """Extracts IBAN (SI56...) from a text block."""
    match = re.search(r'[A-Z]{2}\d{2}\s?(?:\d{4}\s?){4}\d{1,3}', text)
    return match.group(0).strip() if match else ""

def process_bank_statement(pdf_path, output_csv):
    # Load all pages. 'stream' flavor is ideal for this layout.
    tables = camelot.read_pdf(pdf_path, pages='all', flavor='stream', split_text=True)
    
    all_transactions = []
    current_tx = None
    
    # Regex to identify a date (DD.MM.YY) which anchors a new transaction
    date_pattern = re.compile(r'\d{2}\.\d{2}\.\d{2}')

    for table in tables:
        df = table.df
        for index, row in df.iterrows():
            # Skip the header rows (usually contains 'Creditor' or 'value date')
            if "Creditor" in row[0] or "value date" in str(row.iloc[-1]):
                continue
                
            # Detect if this row contains a date in the last column (Column 4)
            date_val = str(row.iloc[-1]).strip()
            is_new_tx = bool(date_pattern.search(date_val))

            if is_new_tx:
                # If we were already building a transaction, save it before starting a new one
                if current_tx:
                    all_transactions.append(current_tx)
                
                # Initialize a new transaction object
                current_tx = {
                    'value date': date_pattern.search(date_val).group(0),
                    'raw_text': row[0],
                    'debit': clean_amount(row[1]),
                    'credit': clean_amount(row[2])
                }
            else:
                # If it's not a new date, it's a continuation of the previous transaction's text
                if current_tx and row[0].strip():
                    current_tx['raw_text'] += " " + row[0].strip()

    # Append the last transaction processed
    if current_tx:
        all_transactions.append(current_tx)

    # Final formatting: Split raw_text into Name, Account, and Details
    final_data = []
    for tx in all_transactions:
        text = tx['raw_text']
        account = extract_account(text)
        
        # Simple split logic: First part is Name, rest is Details (excluding account)
        # Note: This logic can be refined based on specific name lengths
        details = text.replace(account, "").strip()
        parts = details.split("  ", 1) # Try to find a larger gap
        name = parts[0].strip()
        payment_details = parts[1].strip() if len(parts) > 1 else details
        
        final_data.append({
            "value date": tx['value date'],
            "Creditor / Debtor": name,
            "account": account,
            "payment details": payment_details,
            "debit": tx['debit'],
            "credit": tx['credit']
        })

    # Export to CSV with semicolon delimiter
    output_df = pd.DataFrame(final_data)
    output_df.to_csv(output_csv, index=False, sep=';', encoding='utf-8-sig', quoting=csv.QUOTE_ALL)
    print(f"Successfully exported {len(final_data)} transactions to {output_csv}")

# Execution
process_bank_statement("MSDataReport (1).pdf", "Processed_Transactions.csv")