extract_red_text.py

#!/usr/bin/env python3
"""
extract_red_text.py
Improved version that reuses hf_utils for shared heuristics while preserving
the original schema logic, logging and behavior.
"""

import re
import json
import sys
from docx import Document
from docx.oxml.ns import qn

# master schema & patterns (unchanged)
from master_key import TABLE_SCHEMAS, HEADING_PATTERNS, PARAGRAPH_PATTERNS

# canonical helpers (from your new hf_utils.py)
from hf_utils import (
    is_red_font,
    normalize_text,
    normalize_header_text,
    flatten_json,
    find_matching_json_key_and_value,
    get_clean_text,
    has_red_text,
    extract_red_text_segments,
    replace_red_text_in_cell,
    key_is_forbidden_for_position,
)

# -------------------------------------------------------------------
# Small XML helper (kept exactly as before — low-level)
# -------------------------------------------------------------------
def _prev_para_text(tbl):
    """Get text from previous paragraph before table"""
    prev = tbl._tbl.getprevious()
    while prev is not None and not prev.tag.endswith("}p"):
        prev = prev.getprevious()
    if prev is None:
        return ""
    return "".join(node.text for node in prev.iter() if node.tag.endswith("}t") and node.text).strip()

# -------------------------------------------------------------------
# Table context helpers (use normalize_text from hf_utils)
# -------------------------------------------------------------------
def fuzzy_match_heading(heading, patterns):
    """Check if heading matches any pattern with fuzzy matching"""
    if not heading:
        return False
    heading_norm = normalize_text(heading).upper()
    for pattern in patterns:
        try:
            if re.search(pattern, heading_norm, re.IGNORECASE):
                return True
        except re.error:
            # fallback simple substring if pattern isn't a valid re
            if pattern.upper() in heading_norm:
                return True
    return False

def get_table_context(tbl):
    """Get comprehensive context information for table"""
    heading = normalize_text(_prev_para_text(tbl))
    # first row headers
    headers = [normalize_text(c.text) for c in tbl.rows[0].cells if c.text.strip()] if tbl.rows else []
    col0 = [normalize_text(r.cells[0].text) for r in tbl.rows if r.cells and r.cells[0].text.strip()]
    first_cell = normalize_text(tbl.rows[0].cells[0].text) if tbl.rows else ""
    all_cells = []
    for row in tbl.rows:
        for cell in row.cells:
            text = normalize_text(cell.text)
            if text:
                all_cells.append(text)
    return {
        'heading': heading,
        'headers': headers,
        'col0': col0,
        'first_cell': first_cell,
        'all_cells': all_cells,
        'num_rows': len(tbl.rows),
        'num_cols': len(tbl.rows[0].cells) if tbl.rows else 0
    }

# -------------------------------------------------------------------
# Scoring / matching logic (kept your behavior but using normalize_text)
# -------------------------------------------------------------------
def calculate_schema_match_score(schema_name, spec, context):
    """Enhanced calculate match score - IMPROVED for Vehicle Registration tables"""
    score = 0
    reasons = []

    # VEHICLE REGISTRATION BOOST
    if "Vehicle Registration" in schema_name:
        vehicle_keywords = ["registration", "vehicle", "sub-contractor", "weight verification", "rfs suspension"]
        table_text = " ".join(context['headers']).lower() + " " + context['heading'].lower()
        keyword_matches = sum(1 for keyword in vehicle_keywords if keyword in table_text)
        if keyword_matches >= 2:
            score += 150
            reasons.append(f"Vehicle Registration keywords: {keyword_matches}/5")
        elif keyword_matches >= 1:
            score += 75
            reasons.append(f"Some Vehicle Registration keywords: {keyword_matches}/5")

    # SUMMARY TABLE BOOST (existing logic)
    if "Summary" in schema_name and "details" in " ".join(context['headers']).lower():
        score += 100
        reasons.append(f"Summary schema with DETAILS column - perfect match")

    if "Summary" not in schema_name and "details" in " ".join(context['headers']).lower():
        score -= 75
        reasons.append(f"Non-summary schema penalized for DETAILS column presence")

    # Context exclusions
    if spec.get("context_exclusions"):
        table_text = " ".join(context['headers']).lower() + " " + context['heading'].lower()
        for exclusion in spec["context_exclusions"]:
            if exclusion.lower() in table_text:
                score -= 50
                reasons.append(f"Context exclusion penalty: '{exclusion}' found")

    # Context keywords
    if spec.get("context_keywords"):
        table_text = " ".join(context['headers']).lower() + " " + context['heading'].lower()
        keyword_matches = 0
        for keyword in spec["context_keywords"]:
            if keyword.lower() in table_text:
                keyword_matches += 1
        if keyword_matches > 0:
            score += keyword_matches * 15
            reasons.append(f"Context keyword matches: {keyword_matches}/{len(spec['context_keywords'])}")

    # Direct first cell match
    if context['first_cell'] and context['first_cell'].upper() == schema_name.upper():
        score += 100
        reasons.append(f"Direct first cell match: '{context['first_cell']}'")

    # Heading pattern matching
    if spec.get("headings"):
        for h in spec["headings"]:
            if fuzzy_match_heading(context['heading'], [h.get("text", "")]):
                score += 50
                reasons.append(f"Heading match: '{context['heading']}'")
                break

    # Column header matching
    if spec.get("columns"):
        cols = [normalize_text(col) for col in spec["columns"]]
        matches = 0
        for col in cols:
            if any(col.upper() in h.upper() for h in context['headers']):
                matches += 1
        if matches == len(cols):
            score += 60
            reasons.append(f"All column headers match: {cols}")
        elif matches > 0:
            score += matches * 20
            reasons.append(f"Partial column matches: {matches}/{len(cols)}")

    # Label matching for left-oriented tables
    if spec.get("orientation") == "left":
        labels = [normalize_text(lbl) for lbl in spec["labels"]]
        matches = 0
        for lbl in labels:
            if any(lbl.upper() in c.upper() or c.upper() in lbl.upper() for c in context['col0']):
                matches += 1
        if matches > 0:
            score += (matches / len(labels)) * 30
            reasons.append(f"Left orientation label matches: {matches}/{len(labels)}")

    # Enhanced Label matching for row1-oriented tables (Vehicle Registration)
    elif spec.get("orientation") == "row1":
        labels = [normalize_text(lbl) for lbl in spec["labels"]]
        matches = 0
        for lbl in labels:
            if any(lbl.upper() in h.upper() or h.upper() in lbl.upper() for h in context['headers']):
                matches += 1
            elif any(word.upper() in " ".join(context['headers']).upper() for word in lbl.split() if len(word) > 3):
                matches += 0.5
        if matches > 0:
            score += (matches / len(labels)) * 40
            reasons.append(f"Row1 orientation header matches: {matches}/{len(labels)}")

    # Special handling for Declaration tables
    if schema_name == "Operator Declaration" and context['first_cell'].upper() == "PRINT NAME":
        if "OPERATOR DECLARATION" in context['heading'].upper():
            score += 80
            reasons.append("Operator Declaration context match")
        elif any("MANAGER" in cell.upper() for cell in context['all_cells']):
            score += 60
            reasons.append("Manager found in cells (likely Operator Declaration)")

    if schema_name == "NHVAS Approved Auditor Declaration" and context['first_cell'].upper() == "PRINT NAME":
        if any("MANAGER" in cell.upper() for cell in context['all_cells']):
            score -= 50
            reasons.append("Penalty: Manager found (not auditor)")

    return score, reasons

def match_table_schema(tbl):
    """Improved table schema matching with scoring system"""
    context = get_table_context(tbl)
    best_match = None
    best_score = 0
    for name, spec in TABLE_SCHEMAS.items():
        score, reasons = calculate_schema_match_score(name, spec, context)
        if score > best_score:
            best_score = score
            best_match = name
    if best_score >= 20:
        return best_match
    return None

# -------------------------------------------------------------------
# Multi-schema detection & extraction (kept behavior)
# -------------------------------------------------------------------
def check_multi_schema_table(tbl):
    """Check if table contains multiple schemas and split appropriately"""
    context = get_table_context(tbl)
    operator_labels = ["Operator name (Legal entity)", "NHVAS Accreditation No.", "Registered trading name/s", 
                      "Australian Company Number", "NHVAS Manual"]
    contact_labels = ["Operator business address", "Operator Postal address", "Email address", "Operator Telephone Number"]
    has_operator = any(any(op_lbl.upper() in cell.upper() for op_lbl in operator_labels) for cell in context['col0'])
    has_contact = any(any(cont_lbl.upper() in cell.upper() for cont_lbl in contact_labels) for cell in context['col0'])
    if has_operator and has_contact:
        return ["Operator Information", "Operator contact details"]
    return None

def extract_multi_schema_table(tbl, schemas):
    """Extract data from table with multiple schemas"""
    result = {}
    for schema_name in schemas:
        if schema_name not in TABLE_SCHEMAS:
            continue
        spec = TABLE_SCHEMAS[schema_name]
        schema_data = {}
        for ri, row in enumerate(tbl.rows):
            if ri == 0:
                continue
            row_label = normalize_text(row.cells[0].text)
            belongs_to_schema = False
            matched_label = None
            for spec_label in spec["labels"]:
                spec_norm = normalize_text(spec_label).upper()
                row_norm = row_label.upper()
                if spec_norm == row_norm or spec_norm in row_norm or row_norm in spec_norm:
                    belongs_to_schema = True
                    matched_label = spec_label
                    break
            if not belongs_to_schema:
                continue
            for ci, cell in enumerate(row.cells):
                red_txt = "".join(run.text for p in cell.paragraphs for run in p.runs if is_red_font(run)).strip()
                if red_txt:
                    if matched_label not in schema_data:
                        schema_data[matched_label] = []
                    if red_txt not in schema_data[matched_label]:
                        schema_data[matched_label].append(red_txt)
        if schema_data:
            result[schema_name] = schema_data
    return result

# -------------------------------------------------------------------
# Table extraction for schemas (kept your specialized vehicle handling)
# -------------------------------------------------------------------
def extract_table_data(tbl, schema_name, spec):
    """Extract red text data from table based on schema - ENHANCED for Vehicle Registration"""

    # Special handling for vehicle registration tables
    if "Vehicle Registration" in schema_name:
        print(f"    🚗 EXTRACTION FIX: Processing Vehicle Registration table")
        labels = spec["labels"]
        collected = {lbl: [] for lbl in labels}
        seen = {lbl: set() for lbl in labels}

        if len(tbl.rows) < 2:
            print(f"    ❌ Vehicle table has less than 2 rows")
            return {}

        header_row = tbl.rows[0]
        column_mapping = {}

        print(f"    📋 Mapping {len(header_row.cells)} header cells to labels")

        for col_idx, cell in enumerate(header_row.cells):
            header_text = normalize_text(cell.text).strip()
            if not header_text:
                continue

            print(f"      Column {col_idx}: '{header_text}'")

            best_match = None
            best_score = 0

            for label in labels:
                if header_text.upper() == label.upper():
                    best_match = label
                    best_score = 1.0
                    break

                header_words = set(word.upper() for word in header_text.split() if len(word) > 2)
                label_words = set(word.upper() for word in label.split() if len(word) > 2)

                if header_words and label_words:
                    common_words = header_words.intersection(label_words)
                    if common_words:
                        score = len(common_words) / max(len(header_words), len(label_words))
                        if score > best_score and score >= 0.4:
                            best_score = score
                            best_match = label

            if best_match:
                column_mapping[col_idx] = best_match
                print(f"        ✅ Mapped to: '{best_match}' (score: {best_score:.2f})")
            else:
                print(f"        ⚠️ No mapping found for '{header_text}'")

        print(f"    📊 Total column mappings: {len(column_mapping)}")

        # Extract red text from data rows (skip header)
        for row_idx in range(1, len(tbl.rows)):
            row = tbl.rows[row_idx]
            print(f"      📌 Processing data row {row_idx}")
            for col_idx, cell in enumerate(row.cells):
                if col_idx in column_mapping:
                    label = column_mapping[col_idx]
                    red_txt = "".join(run.text for p in cell.paragraphs for run in p.runs if is_red_font(run)).strip()
                    if red_txt:
                        print(f"        🔴 Found red text in '{label}': '{red_txt}'")
                        if red_txt not in seen[label]:
                            seen[label].add(red_txt)
                            collected[label].append(red_txt)
        result = {k: v for k, v in collected.items() if v}
        print(f"    ✅ Vehicle Registration extracted: {len(result)} columns with data")
        return result

    # FALLBACK: original extraction logic for other tables
    labels = spec.get("labels", []) + [schema_name]
    collected = {lbl: [] for lbl in labels}
    seen = {lbl: set() for lbl in labels}
    by_col = (spec.get("orientation") == "row1")
    start_row = 1 if by_col else 0
    rows = tbl.rows[start_row:]

    for ri, row in enumerate(rows):
        for ci, cell in enumerate(row.cells):
            red_txt = "".join(run.text for p in cell.paragraphs for run in p.runs if is_red_font(run)).strip()
            if not red_txt:
                continue
            if by_col:
                if ci < len(spec.get("labels", [])):
                    lbl = spec["labels"][ci]
                else:
                    lbl = schema_name
            else:
                raw_label = normalize_text(row.cells[0].text)
                lbl = None
                for spec_label in spec.get("labels", []):
                    if normalize_text(spec_label).upper() == raw_label.upper():
                        lbl = spec_label
                        break
                if not lbl:
                    for spec_label in spec.get("labels", []):
                        spec_norm = normalize_text(spec_label).upper()
                        raw_norm = raw_label.upper()
                        if spec_norm in raw_norm or raw_norm in spec_norm:
                            lbl = spec_label
                            break
                if not lbl:
                    lbl = schema_name
            if red_txt not in seen[lbl]:
                seen[lbl].add(red_txt)
                collected[lbl].append(red_txt)
    return {k: v for k, v in collected.items() if v}

# -------------------------------------------------------------------
# Main extraction: iterate tables & paragraphs
# -------------------------------------------------------------------
def extract_red_text(input_doc):
    """
    input_doc: docx.Document object or file path
    returns: dict
    """
    if isinstance(input_doc, str):
        doc = Document(input_doc)
    else:
        doc = input_doc
    out = {}
    table_count = 0

    for tbl in doc.tables:
        table_count += 1
        # Check multi-schema table first
        multi_schemas = check_multi_schema_table(tbl)
        if multi_schemas:
            multi_data = extract_multi_schema_table(tbl, multi_schemas)
            for schema_name, schema_data in multi_data.items():
                if schema_data:
                    if schema_name in out:
                        for k, v in schema_data.items():
                            if k in out[schema_name]:
                                out[schema_name][k].extend(v)
                            else:
                                out[schema_name][k] = v
                    else:
                        out[schema_name] = schema_data
            continue

        schema = match_table_schema(tbl)
        if not schema:
            # keep scanning for tables even if no schema matched
            continue
        spec = TABLE_SCHEMAS[schema]
        data = extract_table_data(tbl, schema, spec)
        if data:
            if schema in out:
                for k, v in data.items():
                    if k in out[schema]:
                        out[schema][k].extend(v)
                    else:
                        out[schema][k] = v
            else:
                out[schema] = data

    # paragraphs
    paras = {}
    for idx, para in enumerate(doc.paragraphs):
        red_txt = "".join(r.text for r in para.runs if is_red_font(r)).strip()
        if not red_txt:
            continue

        # find context heading by scanning backward
        context = None
        for j in range(idx-1, -1, -1):
            txt = normalize_text(doc.paragraphs[j].text)
            if txt:
                all_patterns = HEADING_PATTERNS["main"] + HEADING_PATTERNS["sub"]
                if any(re.search(p, txt, re.IGNORECASE) for p in all_patterns):
                    context = txt
                    break

        # if it's date-like and matches date pattern, set context to Date
        if not context and re.fullmatch(PARAGRAPH_PATTERNS["date_line"], red_txt):
            context = "Date"

        if not context:
            context = "(para)"
        paras.setdefault(context, []).append(red_txt)

    if paras:
        out["paragraphs"] = paras
    return out

# -------------------------------------------------------------------
# File-like wrapper (keeps API used elsewhere)
# -------------------------------------------------------------------
def extract_red_text_filelike(input_file, output_file):
    """
    Accepts:
      input_file: file-like object (BytesIO/File) or path
      output_file: file-like object (opened for writing text) or path
    """
    if hasattr(input_file, "seek"):
        input_file.seek(0)
    doc = Document(input_file)
    result = extract_red_text(doc)
    if hasattr(output_file, "write"):
        json.dump(result, output_file, indent=2, ensure_ascii=False)
        output_file.flush()
    else:
        with open(output_file, "w", encoding="utf-8") as f:
            json.dump(result, f, indent=2, ensure_ascii=False)
    return result

# -------------------------------------------------------------------
# CLI entrypoint (preserve original UX)
# -------------------------------------------------------------------
if __name__ == "__main__":
    if len(sys.argv) == 3:
        input_docx = sys.argv[1]
        output_json = sys.argv[2]
        doc = Document(input_docx)
        word_data = extract_red_text(doc)
        with open(output_json, 'w', encoding='utf-8') as f:
            json.dump(word_data, f, indent=2, ensure_ascii=False)
        print(json.dumps(word_data, indent=2, ensure_ascii=False))
    else:
        print("To use as a module: extract_red_text_filelike(input_file, output_file)")