classes.py

import shutil
import bm25s
from bm25s.hf import BM25HF
import threading, re, time, concurrent.futures, requests, os, hashlib, traceback, io, zipfile, subprocess, tempfile, json, fitz
import pandas as pd
import numpy as np

from bs4 import BeautifulSoup
from datasets import load_dataset, Dataset
from datasets.data_files import EmptyDatasetError
from dotenv import load_dotenv

load_dotenv()

class TDocIndexer:
    def __init__(self, max_workers=33):
        self.indexer_length = 0
        self.dataset = "OrganizedProgrammers/3GPPTDocLocation"

        self.indexer = self.load_indexer()
        self.main_ftp_url = "https://3gpp.org/ftp"
        self.valid_doc_pattern = re.compile(r'^(S[1-6P]|C[1-6P]|R[1-6P])-\d+', flags=re.IGNORECASE)
        self.max_workers = max_workers
        
        self.print_lock = threading.Lock()
        self.indexer_lock = threading.Lock()
        
        self.total_indexed = 0
        self.processed_count = 0
        self.total_count = 0

    def load_indexer(self):
        self.indexer_length = 0
        all_docs = {}
        tdoc_locations = load_dataset(self.dataset)
        tdoc_locations = tdoc_locations["train"].to_list()
        for doc in tdoc_locations:
            self.indexer_length += 1
            all_docs[doc["doc_id"]] = doc["url"]

        return all_docs
    
    def save_indexer(self):
        """Save the updated index"""
        data = []
        for doc_id, url in self.indexer.items():
            data.append({"doc_id": doc_id, "url": url})
        
        dataset = Dataset.from_list(data)
        dataset.push_to_hub(self.dataset, token=os.environ["HF"])
        self.indexer = self.load_indexer()

    def get_docs_from_url(self, url):
        try:
            response = requests.get(url, verify=False, timeout=10)
            soup = BeautifulSoup(response.text, "html.parser")
            return [item.get_text() for item in soup.select("tr td a")]
        except Exception as e:
            with self.print_lock:
                print(f"Erreur lors de l'accès à {url}: {e}")
            return []

    def is_valid_document_pattern(self, filename):
        return bool(self.valid_doc_pattern.match(filename))

    def is_zip_file(self, filename):
        return filename.lower().endswith('.zip')

    def extract_doc_id(self, filename):
        if self.is_valid_document_pattern(filename):
            match = self.valid_doc_pattern.match(filename)
            if match:
                # Retourner le motif complet (comme S1-12345)
                full_id = filename.split('.')[0]  # Enlever l'extension si présente
                return full_id.split('_')[0]  # Enlever les suffixes après underscore si présents
        return None

    def process_zip_files(self, files_list, base_url, workshop=False):
        """Traiter une liste de fichiers pour trouver et indexer les ZIP valides"""
        indexed_count = 0
        
        for file in files_list:
            if file in ['./', '../', 'ZIP/', 'zip/']:
                continue
                
            # Vérifier si c'est un fichier ZIP et s'il correspond au motif
            if self.is_zip_file(file) and (self.is_valid_document_pattern(file) or workshop):
                file_url = f"{base_url}/{file}"
                
                # Extraire l'ID du document
                doc_id = self.extract_doc_id(file)
                if doc_id is None:
                    doc_id = file.split('.')[0]
                if doc_id:
                    # Vérifier si ce fichier est déjà indexé
                    with self.indexer_lock:
                        if doc_id in self.indexer and self.indexer[doc_id] == file_url:
                            continue
                            
                        # Ajouter ou mettre à jour l'index
                        self.indexer[doc_id] = file_url
                        indexed_count += 1
                        self.total_indexed += 1
                        
        return indexed_count

    def process_meeting(self, meeting, wg_url, workshop=False):
        """Traiter une réunion individuelle avec multithreading"""
        try:
            if meeting in ['./', '../']:
                return 0
                
            meeting_url = f"{wg_url}/{meeting}"
            
            with self.print_lock:
                print(f"Vérification du meeting: {meeting}")
                
            # Vérifier le contenu de la réunion
            meeting_contents = self.get_docs_from_url(meeting_url)
            
            key = None
            if "docs" in [x.lower() for x in meeting_contents]:
                key = "docs"
            elif "tdocs" in [x.lower() for x in meeting_contents]:
                key = "tdocs"
            elif "tdoc" in [x.lower() for x in meeting_contents]:
                key = "tdoc"
                
            if key is not None:
                docs_url = f"{meeting_url}/{key}"
                
                with self.print_lock:
                    print(f"Vérification des documents présent dans {docs_url}")
                    
                # Récupérer la liste des fichiers dans le dossier Docs
                docs_files = self.get_docs_from_url(docs_url)
                
                # 1. Indexer les fichiers ZIP directement dans le dossier Docs
                docs_indexed_count = self.process_zip_files(docs_files, docs_url, workshop)
                
                if docs_indexed_count > 0:
                    with self.print_lock:
                        print(f"{docs_indexed_count} fichiers trouvés")
                
                # 2. Vérifier le sous-dossier ZIP s'il existe
                if "zip" in [x.lower() for x in docs_files]:
                    zip_url = f"{docs_url}/zip"
                    
                    with self.print_lock:
                        print(f"Vérification du dossier ./zip: {zip_url}")
                        
                    # Récupérer les fichiers dans le sous-dossier ZIP
                    zip_files = self.get_docs_from_url(zip_url)
                    
                    # Indexer les fichiers ZIP dans le sous-dossier ZIP
                    zip_indexed_count = self.process_zip_files(zip_files, zip_url, workshop)
                    
                    if zip_indexed_count > 0:
                        with self.print_lock:
                            print(f"{zip_indexed_count} fichiers trouvés")
            
            # Mise à jour du compteur de progression
            with self.indexer_lock:
                self.processed_count += 1
                
            # Affichage de la progression
            with self.print_lock:
                progress = (self.processed_count / self.total_count) * 100 if self.total_count > 0 else 0
                print(f"\rProgression: {self.processed_count}/{self.total_count} réunions traitées ({progress:.1f}%)")
                
            return 1  # Réunion traitée avec succès
            
        except Exception as e:
            with self.print_lock:
                print(f"\nErreur lors du traitement de la réunion {meeting}: {str(e)}")
            return 0

    def process_workgroup(self, wg, main_url):
        """Traiter un groupe de travail avec multithreading pour ses réunions"""
        if wg in ['./', '../']:
            return
            
        wg_url = f"{main_url}/{wg}"
        
        with self.print_lock:
            print(f"Vérification du working group: {wg}")
            
        # Récupérer les dossiers de réunion
        meeting_folders = self.get_docs_from_url(wg_url)
        
        # Ajouter au compteur total
        self.total_count += len([m for m in meeting_folders if m not in ['./', '../']])
        
        # Utiliser ThreadPoolExecutor pour traiter les réunions en parallèle
        with concurrent.futures.ThreadPoolExecutor(max_workers=self.max_workers) as executor:
            futures = [executor.submit(self.process_meeting, meeting, wg_url) 
                      for meeting in meeting_folders if meeting not in ['./', '../']]
            
            total = len(futures)
            done_count = 0
            yield f"event: get-maximum\ndata: {total}\n\n"

            for future in concurrent.futures.as_completed(futures):
                done_count += 1
                yield f"event: progress\ndata: {done_count}\n\n"

    def index_all_tdocs(self):
        """Indexer tous les documents ZIP dans la structure FTP 3GPP avec multithreading"""
        print("Démarrage de l'indexation des TDocs 3GPP complète")
        
        start_time = time.time()
        docs_count_before = self.indexer_length
        
        # Principaux groupes TSG
        main_groups = ["tsg_sa", "tsg_ct", "tsg_ran"]  # Ajouter d'autres si nécessaire
        
        for main_tsg in main_groups:
            print(f"Indexation de {main_tsg.upper()}...")
            
            main_url = f"{self.main_ftp_url}/{main_tsg}"
            
            # Récupérer les groupes de travail
            workgroups = self.get_docs_from_url(main_url)
            
            # Traiter chaque groupe de travail séquentiellement
            # (mais les réunions à l'intérieur seront traitées en parallèle)
            for wg in workgroups:
                yield f"event: info\ndata: {main_tsg}-{wg}\n\n"
                for content in self.process_workgroup(wg, main_url):
                    yield content
        
        docs_count_after = len(self.indexer)
        new_docs_count = abs(docs_count_after - docs_count_before)
        
        print(f"Indexation terminée en {time.time() - start_time:.2f} secondes")
        print(f"Nouveaux documents ZIP indexés: {new_docs_count}")
        print(f"Total des documents dans l'index: {docs_count_after}")
        
        return self.indexer

    def index_all_workshops(self):
        print("Démarrage de l'indexation des workshops ZIP 3GPP...")
        start_time = time.time()
        docs_count_before = len(self.indexer)

        print("\nIndexation du dossier 'workshop'")
        main_url = f"{self.main_ftp_url}/workshop"
            
        # Récupérer les dossiers de réunion
        meeting_folders = self.get_docs_from_url(main_url)
        
        # Ajouter au compteur total
        self.total_count += len([m for m in meeting_folders if m not in ['./', '../']])
        
        # Utiliser ThreadPoolExecutor pour traiter les réunions en parallèle
        with concurrent.futures.ThreadPoolExecutor(max_workers=self.max_workers) as executor:
            futures = [executor.submit(self.process_meeting, meeting, main_url, workshop=True) 
                      for meeting in meeting_folders if meeting not in ['./', '../']]
            total = len(futures)
            done_count = 0

            yield f"event: get-maximum\ndata: {total}\n\n"

            for future in concurrent.futures.as_completed(futures):
                done_count += 1
                yield f"event: progress\ndata: {done_count}\n\n"
        
        docs_count_after = len(self.indexer)
        new_docs_count = docs_count_after - docs_count_before
        
        print(f"\nIndexation terminée en {time.time() - start_time:.2f} secondes")
        print(f"Nouveaux documents ZIP indexés: {new_docs_count}")
        print(f"Total des documents dans l'index: {docs_count_after}")
        
        return self.indexer

class Spec3GPPIndexer:
    def __init__(self, max_workers=16):
        self.spec_contents = load_dataset("OrganizedProgrammers/3GPPSpecContent")["train"].to_list()
        self.documents_by_spec_num = self._make_doc_index(self.spec_contents)
        self.indexed_specifications = {}
        self.specifications_passed = set()
        self.processed_count = 0
        self.total_count = 0

        self.DICT_LOCK = threading.Lock()
        self.DOCUMENT_LOCK = threading.Lock()
        self.STOP_EVENT = threading.Event()
        self.max_workers = max_workers
        self.LIBREOFFICE_SEMAPHORE = threading.Semaphore(self.max_workers)

    def _make_doc_index(self, specs):
        doc_index = {}
        for section in specs:
            if section["doc_id"] not in doc_index:
                doc_index[section["doc_id"]] = {"content": {section["section"]: section["content"]}, "hash": section["hash"]}
            else:
                doc_index[section["doc_id"]]["content"][section["section"]] = section["content"]
        return doc_index

    @staticmethod
    def version_to_code(version_str):
        chars = "0123456789abcdefghijklmnopqrstuvwxyz"
        parts = version_str.split('.')
        if len(parts) != 3:
            return None
        try:
            x, y, z = [int(p) for p in parts]
        except ValueError:
            return None
        if x < 36 and y < 36 and z < 36:
            return f"{chars[x]}{chars[y]}{chars[z]}"
        else:
            return f"{str(x).zfill(2)}{str(y).zfill(2)}{str(z).zfill(2)}"

    @staticmethod
    def hasher(specification, version_code):
        return hashlib.md5(f"{specification}{version_code}".encode()).hexdigest()

    @staticmethod
    def get_scope(content):
        for title, text in content.items():
            if title.lower().endswith("scope"):
                return text
        return ""

    def get_text(self, specification, version_code):
        if self.STOP_EVENT.is_set():
            return []

        doc_id = specification
        series = doc_id.split(".")[0]
        url = f"https://www.3gpp.org/ftp/Specs/archive/{series}_series/{doc_id}/{doc_id.replace('.', '')}-{version_code}.zip"

        try:
            response = requests.get(url, verify=False)
            if response.status_code != 200:
                return []

            zip_bytes = io.BytesIO(response.content)
            with zipfile.ZipFile(zip_bytes) as zip_file:
                # Filtrer uniquement fichiers .doc et .docx
                docx_files = [f for f in zip_file.namelist() if f.lower().endswith(('.doc', '.docx'))]
                if not docx_files:
                    return []

                full_text = []

                for doc_file in docx_files:
                    with tempfile.TemporaryDirectory() as tmpdir:
                        extracted_path = os.path.join(tmpdir, os.path.basename(doc_file))
                        with open(extracted_path, 'wb') as f:
                            f.write(zip_file.read(doc_file))

                        # Profil libreoffice temp dédié
                        profile_dir = tempfile.mkdtemp(prefix="libreoffice_profile_")

                        try:
                            with self.LIBREOFFICE_SEMAPHORE:
                                cmd = [
                                    'soffice',
                                    '--headless',
                                    f'-env:UserInstallation=file://{profile_dir}',
                                    '--convert-to', 'txt:Text',
                                    '--outdir', tmpdir,
                                    extracted_path
                                ]
                                subprocess.run(cmd, check=True, timeout=60*5, stdout=subprocess.PIPE, stderr=subprocess.PIPE)

                                txt_file = os.path.splitext(extracted_path)[0] + '.txt'
                                if os.path.exists(txt_file):
                                    with open(txt_file, 'r', encoding='utf-8', errors='ignore') as ftxt:
                                        full_text.extend(ftxt.readlines())
                        finally:
                            shutil.rmtree(profile_dir, ignore_errors=True)

                return full_text

        except Exception as e:
            print(f"Error getting text for {specification} v{version_code}: {e}")
            return []

    def get_spec_content(self, specification, version_code):
        if self.STOP_EVENT.is_set():
            return {}

        text = self.get_text(specification, version_code)
        if not text:
            return {}

        chapters = []
        chapter_regex = re.compile(r"^(\d+[a-z]?(?:\.\d+)*)\t[A-Z0-9][\ \S]+[^\.]$")
        for i, line in enumerate(text):
            if chapter_regex.fullmatch(line):
                chapters.append((i, line))

        document = {}
        for i in range(len(chapters)):
            start_index, chapter_title = chapters[i]
            end_index = chapters[i+1][0] if i+1 < len(chapters) else len(text)
            content_lines = text[start_index + 1:end_index]
            document[chapter_title.replace("\t", " ")] = "\n".join(content_lines)

        return document

    def fetch_spec_table(self):
        response = requests.get(
            'https://www.3gpp.org/dynareport?code=status-report.htm',
            headers={"User-Agent": 'Mozilla/5.0'},
            verify=False
        )
        dfs = pd.read_html(io.StringIO(response.text))
        for x in range(len(dfs)):
            dfs[x] = dfs[x].replace({np.nan: None})
        columns_needed = [0, 1, 2, 3, 4]
        extracted_dfs = [df.iloc[:, columns_needed] for df in dfs]
        columns = [x.replace("\xa0", "_") for x in extracted_dfs[0].columns]
        specifications = []
        for df in extracted_dfs:
            for index, row in df.iterrows():
                doc = row.to_list()
                doc_dict = dict(zip(columns, doc))
                specifications.append(doc_dict)
        return specifications

    def process_specification(self, spec):
        if self.STOP_EVENT.is_set():
            return
        try:
            doc_id = str(spec['spec_num'])
            version_code = self.version_to_code(str(spec['vers']))
            if not version_code:
                with self.DICT_LOCK:
                    self.processed_count += 1
                return

            document = None
            already_indexed = False
            with self.DOCUMENT_LOCK:
                doc_in_cache = doc_id in self.documents_by_spec_num and \
                               self.documents_by_spec_num[doc_id]["hash"] == self.hasher(doc_id, version_code)

            if doc_in_cache and doc_id not in self.specifications_passed:
                document = self.documents_by_spec_num[doc_id]
                self.specifications_passed.add(doc_id)
                already_indexed = True
            elif doc_id not in self.specifications_passed:
                doc_content = self.get_spec_content(doc_id, version_code)
                if doc_content:
                    document = {"content": doc_content, "hash": self.hasher(doc_id, version_code)}
                    with self.DOCUMENT_LOCK:
                        self.documents_by_spec_num[doc_id] = document
                        self.specifications_passed.add(doc_id)
                    already_indexed = False

            if document:
                url = f"https://www.3gpp.org/ftp/Specs/archive/{doc_id.split('.')[0]}_series/{doc_id}/{doc_id.replace('.', '')}-{version_code}.zip"
                metadata = {
                    "id": doc_id,
                    "title": spec.get("title", ""),
                    "type": spec.get("type", ""),
                    "version": str(spec.get("vers", "")),
                    "working_group": spec.get("WG", ""),
                    "url": url,
                    "scope": self.get_scope(document["content"])
                }
                key = f"{doc_id}+-+{spec.get('title', '')}+-+{spec.get('type', '')}+-+{spec.get('vers', '')}+-+{spec.get('WG', '')}"
                with self.DICT_LOCK:
                    self.indexed_specifications[key] = metadata

            with self.DICT_LOCK:
                self.processed_count += 1
                status = "already indexed" if already_indexed else "indexed now"
                print(f"Spec {doc_id} ({spec.get('title', '')}): {status} - Progress {self.processed_count}/{self.total_count}")

        except Exception as e:
            traceback.print_exc()
            print(f"Error processing spec {spec.get('spec_num')} v{spec.get('vers')}: {e}")
            with self.DICT_LOCK:
                self.processed_count += 1
                print(f"Progress: {self.processed_count}/{self.total_count} specs processed")
    
    def get_document(self, spec_id: str, spec_title: str):
        text = [f"{spec_id} - {spec_title}\n"]
        for section in self.spec_contents:
            if spec_id == section["doc_id"]:
                text.extend([f"{section['section']}\n\n{section['content']}"])
        return text

    def create_bm25_index(self):
        dataset_metadata = self.indexed_specifications.values()
        unique_specs = set()
        corpus_json = []

        for specification in dataset_metadata:
            if specification['id'] in unique_specs: continue
            for section in self.spec_contents:
                if specification['id'] == section['doc_id']:
                    corpus_json.append({"text": f"{section['section']}\n{section['content']}", "metadata": {
                "id": specification['id'],
                "title": specification['title'],
                "section_title": section['section'],
                "version": specification['version'],
                "type": specification['type'],
                "working_group": specification['working_group'],
                "url": specification['url'],
                "scope": specification['scope']
            }})
        
        corpus_text = [doc["text"] for doc in corpus_json]
        corpus_tokens = bm25s.tokenize(corpus_text, stopwords="en")

        print("Indexing BM25")
        retriever = BM25HF(corpus=corpus_json)
        retriever.index(corpus_tokens)

        retriever.save_to_hub("OrganizedProgrammers/3GPPBM25IndexSections", token=os.environ.get("HF"))

        unique_specs = set()
        corpus_json = []

        for specification in dataset_metadata:
            if specification['id'] in unique_specs: continue
            text_list = self.get_document(specification['id'], specification['title'])
            text = "\n".join(text_list)
            if len(text_list) == 1: continue
            corpus_json.append({"text": text, "metadata": specification})
            unique_specs.add(specification['id'])
                
        corpus_text = [doc["text"] for doc in corpus_json]
        corpus_tokens = bm25s.tokenize(corpus_text, stopwords="en")

        print("Indexing BM25")
        retriever = BM25HF(corpus=corpus_json)
        retriever.index(corpus_tokens)

        retriever.save_to_hub("OrganizedProgrammers/3GPPBM25IndexSingle", token=os.environ.get("HF"))

    def run(self):
        print("Fetching specification tables from 3GPP...")
        yield "event: info\ndata: Indexing 3GPP specs ...\n\n"
        specifications = self.fetch_spec_table()
        self.total_count = len(specifications)
        print(f"Processing {self.total_count} specs with {self.max_workers} threads...")
        with concurrent.futures.ThreadPoolExecutor(max_workers=self.max_workers) as executor:
            futures = [executor.submit(self.process_specification, spec) for spec in specifications]
            total = len(futures)
            done_count = 0
            yield f"event: get-maximum\ndata: {total}\n\n"

            for future in concurrent.futures.as_completed(futures):
                done_count += 1
                yield f"event: progress\ndata: {done_count}\n\n"
                if self.STOP_EVENT.is_set():
                    break
        print("All specs processed.")

    # Sauvegarde (identique au script original)
    def save(self):
        print("Saving indexed data...")
        flat_metadata = [metadata for metadata in self.indexed_specifications.values()]
        flat_docs = []
        print("Flatting doc contents")
        for doc_id, data in self.documents_by_spec_num.items():
            for title, content in data["content"].items():
                flat_docs.append({"hash": data["hash"], "doc_id": doc_id, "section": title, "content": content})
        print("Creating datasets ...")
        push_spec_content = Dataset.from_list(flat_docs)
        push_spec_metadata = Dataset.from_list(flat_metadata)
        # Token handling assumed set in environment
        print("Pushing ...")
        push_spec_content.push_to_hub("OrganizedProgrammers/3GPPSpecContent", token=os.environ["HF"])
        push_spec_metadata.push_to_hub("OrganizedProgrammers/3GPPSpecMetadata", token=os.environ["HF"])
        
        self.spec_contents = load_dataset("OrganizedProgrammers/3GPPSpecContent")["train"].to_list()
        self.documents_by_spec_num = self._make_doc_index(self.spec_contents)
        print("Save finished.")

class SpecETSIIndexer:
    def __init__(self, max_workers=16):
        self.session = requests.Session()
        self.session.verify = False

        self.spec_contents = load_dataset("OrganizedProgrammers/ETSISpecContent")["train"].to_list()
        self.documents_by_spec_num = self._make_doc_index(self.spec_contents)
        self.indexed_specifications = {}
        self.specifications_passed = set()
        self.processed_count = 0
        self.total_count = 0

        self.DICT_LOCK = threading.Lock()
        self.DOCUMENT_LOCK = threading.Lock()
        self.STOP_EVENT = threading.Event()
        self.max_workers = max_workers

        self.df = self._fetch_spec_table()
    
    def _make_doc_index(self, specs):
        doc_index = {}
        for section in specs:
            if section["doc_id"] not in doc_index:
                doc_index[section["doc_id"]] = {"content": {section["section"]: section["content"]}, "hash": section["hash"]}
            else:
                doc_index[section["doc_id"]]["content"][section["section"]] = section["content"]
        return doc_index
    
    def _fetch_spec_table(self):
        # Connexion login et récupération CSV TS/TR
        print("Connexion login ETSI...")
        self.session.post(
            "https://portal.etsi.org/ETSIPages/LoginEOL.ashx",
            verify=False,
            headers={"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) ..."},
            data=json.dumps({"username": os.environ.get("EOL_USER"), "password": os.environ.get("EOL_PASSWORD")}),
        )

        print("Récupération des métadonnées TS/TR …")
        url_ts = "https://www.etsi.org/?option=com_standardssearch&view=data&format=csv&includeScope=1&page=1&search=&title=1&etsiNumber=1&content=0&version=0&onApproval=0&published=1&withdrawn=0&historical=0&isCurrent=1&superseded=0&harmonized=0&keyword=&TB=&stdType=TS&frequency=&mandate=&collection=&sort=1"
        url_tr = url_ts.replace("stdType=TS", "stdType=TR")
        data_ts = self.session.get(url_ts, verify=False).content
        data_tr = self.session.get(url_tr, verify=False).content
        df_ts = pd.read_csv(io.StringIO(data_ts.decode('utf-8')), sep=";", skiprows=1, index_col=False)
        df_tr = pd.read_csv(io.StringIO(data_tr.decode('utf-8')), sep=";", skiprows=1, index_col=False)

        backup_ts = df_ts["ETSI deliverable"]
        backup_tr = df_tr["ETSI deliverable"]
        df_ts["ETSI deliverable"] = df_ts["ETSI deliverable"].str.extract(r"\s*ETSI TS (\d+ \d+(?:-\d+(?:-\d+)?)?)")
        df_tr["ETSI deliverable"] = df_tr["ETSI deliverable"].str.extract(r"\s*ETSI TR (\d+ \d+(?:-\d+(?:-\d+)?)?)")
        version1 = backup_ts.str.extract(r"\s*ETSI TS \d+ \d+(?:-\d+(?:-\d+)?)? V(\d+\.\d+\.\d+)")
        version2 = backup_tr.str.extract(r"\s*ETSI TR \d+ \d+(?:-\d+(?:-\d+)?)? V(\d+\.\d+\.\d+)")
        df_ts["Version"] = version1[0]
        df_tr["Version"] = version2[0]

        def ver_tuple(v):
            return tuple(map(int, v.split(".")))
        df_ts["temp"] = df_ts["Version"].apply(ver_tuple)
        df_tr["temp"] = df_tr["Version"].apply(ver_tuple)
        df_ts["Type"] = "TS"
        df_tr["Type"] = "TR"
        df = pd.concat([df_ts, df_tr])
        unique_df = df.loc[df.groupby("ETSI deliverable")["temp"].idxmax()]
        unique_df = unique_df.drop(columns="temp")
        unique_df = unique_df[(~unique_df["title"].str.contains("3GPP", case=True, na=False))]
        df = df.drop(columns="temp")
        df = df[(~df["title"].str.contains("3GPP", case=True, na=False))]
        return df
    
    @staticmethod
    def hasher(specification: str, version: str):
        return hashlib.md5(f"{specification}{version}".encode()).hexdigest()

    @staticmethod
    def get_scope(content):
        for title, text in content.items():
            if title.lower().endswith("scope"):
                return text
        return ""
    
    def get_document(self, spec_id: str, spec_title: str):
        text = [f"{spec_id} - {spec_title}\n"]
        for section in self.spec_contents:
            if spec_id == section["doc_id"]:
                text.extend([f"{section['section']}\n\n{section['content']}"])
        return text
    
    def get_text(self, specification: str):
        if self.STOP_EVENT.is_set():
            return None, []
        print(f"\n[INFO] Tentative de récupération de la spécification {specification}", flush=True)
        try:
            # Récupérer la ligne avec le bon lien PDF
            row = self.df[self.df["ETSI deliverable"] == specification]
            if row.empty:
                print(f"[WARN] Spécification {specification} absente du tableau")
                return None, []

            pdf_link = row.iloc[0]["PDF link"]
            response = self.session.get(
                pdf_link,
                headers={"User-Agent": 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) ...'}
            )
            if response.status_code != 200:
                print(f"[ERREUR] Echec du téléchargement du PDF pour {specification}.")
                return None, []
            pdf = fitz.open(stream=response.content, filetype="pdf")
            return pdf, pdf.get_toc()
        except Exception as e:
            print(f"[ERROR] Échec get_text pour {specification} : {e}", flush=True)
            return None, []
    
    def get_spec_content(self, specification: str):
        def extract_sections(text, titles):
            sections = {}
            sorted_titles = sorted(titles, key=lambda t: text.find(t))
            for i, title in enumerate(sorted_titles):
                start = text.find(title)
                if i + 1 < len(sorted_titles):
                    end = text.find(sorted_titles[i + 1])
                    sections[re.sub(r"\s+", " ", title)] = re.sub(r"\s+", " ", text[start:end].replace(title, "").strip().rstrip())
                else:
                    sections[re.sub(r"\s+", " ", title)] = re.sub(r"\s+", " ", text[start:].replace(title, "").strip().rstrip())
            return sections

        if self.STOP_EVENT.is_set():
            return {}
        print(f"[INFO] Extraction du contenu de {specification}", flush=True)
        pdf, doc_toc = self.get_text(specification)
        text = []
        if not pdf or not doc_toc:
            print("[ERREUR] Pas de texte ou table of contents trouvé !")
            return {}
        # On prend à partir de la première réelle page référencée
        first_page = 0
        for level, title, page in doc_toc:
            first_page = page - 1
            break
        for page in pdf[first_page:]:
            text.append("\n".join([line.strip() for line in page.get_text().splitlines()]))
        text = "\n".join(text)
        if not text or not doc_toc or self.STOP_EVENT.is_set():
            print("[ERREUR] Pas de texte/table of contents récupéré !")
            return {}
        titles = []
        for level, title, page in doc_toc:
            if self.STOP_EVENT.is_set():
                return {}
            if title and title[0].isnumeric() and '\n'.join(title.strip().split(" ", 1)) in text:
                titles.append('\n'.join(title.strip().split(" ", 1)))
        return extract_sections(text, titles)

    def process_specification(self, spec):
        if self.STOP_EVENT.is_set():
            return
        try:
            version = spec.get('Version')
            if not version: return
            doc_id = str(spec.get("ETSI deliverable"))
            document = None
            already_indexed = False

            with self.DOCUMENT_LOCK:
                if (doc_id in self.documents_by_spec_num
                    and self.documents_by_spec_num[doc_id]["hash"] == self.hasher(doc_id, version)
                    and doc_id not in self.specifications_passed):
                    document = self.documents_by_spec_num[doc_id]
                    self.specifications_passed.add(doc_id)
                    already_indexed = True
                elif doc_id in self.specifications_passed:
                    document = self.documents_by_spec_num[doc_id]
                    already_indexed = True
                else:
                    document_content = self.get_spec_content(doc_id)
                    if document_content:
                        self.documents_by_spec_num[doc_id] = {"content": document_content, "hash": self.hasher(doc_id, version)}
                        document = {"content": document_content, "hash": self.hasher(doc_id, version)}
                        self.specifications_passed.add(doc_id)
                        already_indexed = False

            if document:
                string_key = f"{doc_id}+-+{spec['title']}+-+{spec['Type']}+-+{spec['Version']}"
                metadata = {
                    "id": str(doc_id),
                    "title": spec["title"],
                    "type": spec["Type"],
                    "version": version,
                    "url": spec["PDF link"],
                    "scope": "" if not document else self.get_scope(document["content"])
                }
                with self.DICT_LOCK:
                    self.indexed_specifications[string_key] = metadata

            with self.DICT_LOCK:
                self.processed_count += 1
                status = "already indexed" if already_indexed else "indexed now"
                print(f"Spec {doc_id} ({spec.get('title', '')}): {status} - Progress {self.processed_count}/{self.total_count}")

        except Exception as e:
            traceback.print_exc()
            print(f"\n[ERREUR] Échec du traitement de {doc_id} {spec.get('Version')}: {e}", flush=True)
            with self.DICT_LOCK:
                self.processed_count += 1
                print(f"Progress: {self.processed_count}/{self.total_count} specs processed")
    
    def run(self):
        print("Démarrage indexation ETSI…")
        yield "event: info\ndata: Indexing ETSI specs ...\n\n"
        specifications = self.df.to_dict(orient="records")
        self.total_count = len(specifications)
        print(f"Traitement de {self.total_count} specs avec {self.max_workers} threads...\n")

        with concurrent.futures.ThreadPoolExecutor(max_workers=self.max_workers) as executor:
            futures = [executor.submit(self.process_specification, spec) for spec in specifications]
            total = len(futures)
            done_count = 0
            yield f"event: get-maximum\ndata: {total}\n\n"

            for future in concurrent.futures.as_completed(futures):
                done_count += 1
                yield f"event: progress\ndata: {done_count}\n\n"
                if self.STOP_EVENT.is_set():
                    break

        print(f"\nAll {self.processed_count}/{self.total_count} specs processed.")

    def save(self):
        print("\nSauvegarde en cours...", flush=True)
        flat_metadata = [metadata for metadata in self.indexed_specifications.values()]
        flat_docs = []
        for doc_id, data in self.documents_by_spec_num.items():
            for title, content in data["content"].items():
                flat_docs.append({"hash": data["hash"], "doc_id": doc_id, "section": title, "content": content})
        push_spec_content = Dataset.from_list(flat_docs)
        push_spec_metadata = Dataset.from_list(flat_metadata)
        push_spec_content.push_to_hub("OrganizedProgrammers/ETSISpecContent", token=os.environ["HF"])
        push_spec_metadata.push_to_hub("OrganizedProgrammers/ETSISpecMetadata", token=os.environ["HF"])
        
        self.spec_contents = load_dataset("OrganizedProgrammers/ETSISpecContent")["train"].to_list()
        self.documents_by_spec_num = self._make_doc_index(self.spec_contents)
        print("Sauvegarde terminée.")
    
    def create_bm25_index(self):
        dataset_metadata = self.indexed_specifications.values()
        unique_specs = set()
        corpus_json = []

        for specification in dataset_metadata:
            if specification['id'] in unique_specs: continue
            for section in self.spec_contents:
                if specification['id'] == section['doc_id']:
                    corpus_json.append({"text": f"{section['section']}\n{section['content']}", "metadata": {
                "id": specification['id'],
                "title": specification['title'],
                "section_title": section['section'],
                "version": specification['version'],
                "type": specification['type'],
                "url": specification['url'],
                "scope": specification['scope']
            }})
        
        corpus_text = [doc["text"] for doc in corpus_json]
        corpus_tokens = bm25s.tokenize(corpus_text, stopwords="en")

        print("Indexing BM25")
        retriever = BM25HF(corpus=corpus_json)
        retriever.index(corpus_tokens)

        retriever.save_to_hub("OrganizedProgrammers/ETSIBM25IndexSections", token=os.environ.get("HF"))

        unique_specs = set()
        corpus_json = []

        for specification in dataset_metadata:
            if specification['id'] in unique_specs: continue
            text_list = self.get_document(specification['id'], specification['title'])
            text = "\n".join(text_list)
            if len(text_list) == 1: continue
            corpus_json.append({"text": text, "metadata": specification})
            unique_specs.add(specification['id'])
                
        corpus_text = [doc["text"] for doc in corpus_json]
        corpus_tokens = bm25s.tokenize(corpus_text, stopwords="en")

        print("Indexing BM25")
        retriever = BM25HF(corpus=corpus_json)
        retriever.index(corpus_tokens)

        retriever.save_to_hub("OrganizedProgrammers/ETSIBM25IndexSingle", token=os.environ.get("HF"))