import multiprocessing import pathlib import re import time import traceback import orjson import tqdm import typer from bs4 import BeautifulSoup, Tag from markdownify import MarkdownConverter, chomp CONCURRENT = 64 class WikiConverter(MarkdownConverter): def convert_a(self, el, text, convert_as_inline): prefix, suffix, text = chomp(text) if not text: return "" return "%s%s%s" % (prefix, text, suffix) integer_rgx = re.compile("^[0-9]*$") @staticmethod def is_intable(string: str): if not string or not string.isdigit(): return False if WikiConverter.integer_rgx.match(string): return True def convert_img(self, el, text, convert_as_inline): convert_as_inline = True alt = el.attrs.get("alt", None) or "" if ( convert_as_inline and el.parent.name not in self.options["keep_inline_images_in"] ): return alt return alt def convert_li(self, el, text, convert_as_inline): parent = el.parent if parent is not None and parent.name == "ol": start = parent.get("start") if start and WikiConverter.is_intable(start.strip()): start = int(start.strip()) else: start = 1 bullet = "%s." % (start + parent.index(el)) else: depth = -1 while el: if el.name == "ul": depth += 1 el = el.parent bullets = self.options["bullets"] bullet = bullets[depth % len(bullets)] return "%s %s\n" % (bullet, (text or "").strip()) class MultilangWikipediaProcessor: def __init__(self) -> None: self.md = WikiConverter() def is_stub(self, soup: BeautifulSoup): for plainlinks in soup.select(".metadata.plainlinks"): if "stub" in plainlinks.get("id", "") or "stub" in plainlinks.get( "class", [] ): return True return False def rital_ambox(self, input_soup: BeautifulSoup): ambox_classes = [] for ambox in selects: if ambox is not None: ambox_classes.append(ambox.get("class")) for ambox in selects: ambox.decompose() return input_soup, ambox_classes def rital_ombox(self, input_soup: BeautifulSoup): ombox_classes = [] selects = input_soup.select('table[class~="ombox"]') for ombox in selects: if ombox is not None: ombox_classes.append(ombox["class"]) for ombox in selects: ombox.decompose() return input_soup, ombox_classes def table_filtration(self, input_soup: BeautifulSoup, title): for table in input_soup.select("table"): tds = len(table.find_all("td")) texsize = len(table.get_text().replace(" ", "")) if tds >= texsize and texsize < 50: print(table.get_text().replace(" ", "")) print("Removing table from", title, ". TD exceeds Content") return input_soup all_selectors = [ "style", # Remove styling "sup.reference", # Seems to still exist across "table.nomobile", # Seems to still exist across ] def process_infobox(self, infobox: Tag): return str(infobox) def process_figures(self, figure: Tag): figure_data = {} fig_a = figure.find("a") fig_cap = figure.find("figcaption") if fig_a: figure_data["file_url"] = fig_a.get("href", None) else: figure_data["file_url"] = None if fig_cap: figure_data["caption"] = fig_cap.get_text() else: figure_data["caption"] = None if figure_data["caption"] == figure_data["file_url"] is None: return None return figure_data def convert_soup(self, input_soup: BeautifulSoup): [i.unwrap() for i in input_soup.select('[data-mw^="interface"]')] [i.decompose() for i in input_soup.select(", ".join(self.all_selectors))] title = input_soup.select_one("title").extract() # titletext = title.get_text() for i in input_soup.select(".mw-collapsible"): hidden = i.select_one("div.hidden-content") if hidden: # Expose collapsed content hidden["class"].remove("hidden-content") for i in input_soup.select("[data-mw]"): i["data-mw"] = "" ifbs = [i.extract() for i in input_soup.select("table.infobox")] ifbs += [i.extract() for i in input_soup.select("table.sidebar.vcard.hlist")] ifbs += [i.extract() for i in input_soup.select("table.infobox.vcard")] ifbs = [self.process_infobox(ifb) for ifb in ifbs] figures = [ self.process_figures(fig.extract()) for fig in input_soup.select('figure[typeof^="mw:File/Thumb"]') ] return input_soup, ifbs, figures, title def convert(self, wiki_data: bytes): data = orjson.loads(wiki_data.rstrip(b"\n")) try: templates = [ ":".join(template["name"].split(":")[1:]) for template in data.get("templates", []) ] categories = [ ":".join(category["name"].split(":")[1:]) for category in data.get("categories", []) ] if not data["article_body"].get("wikitext"): return None soup = BeautifulSoup(data["article_body"]["html"], "lxml") is_stub = self.is_stub(soup) soup, infobox, figures, title = self.convert_soup(soup) # soup, issues, issue_selectors = self.remove_templates(soup, all_templates) soup, amboxes = self.rital_ambox(soup) soup, omboxes = self.rital_ombox(soup) soup = self.table_filtration(soup, title) text = ( self.md.convert_soup(soup) .strip() .replace("\n\n", "\n") .replace("\n\n\n", "\n\n") ) return orjson.dumps( { "id": data["identifier"], "title": data["name"], "url": data["url"], "stub": is_stub, "template": templates, "category": categories, "license": [lic["name"] for lic in data["license"]], "text": text, "wikitext": data["article_body"].get("wikitext"), "lang": data["in_language"]["identifier"], "abstract": data.get("abstract", ""), "boxes_filters": amboxes + omboxes, "infobox_html": infobox, "figures_dict": figures, } ) except Exception as e: print(f"Exception at Soup Conversion: {e} [{data['name']}]") return None queue = multiprocessing.Queue(maxsize=1024 * 1024) def worker(worker_idx: int, prefix_path: pathlib.Path): if prefix_path.parent: prefix_path.parent.mkdir(exist_ok=True, parents=True) processor = MultilangWikipediaProcessor() prefix_path = prefix_path.with_name( f"{prefix_path.name}-{str(worker_idx).zfill(2)}.jsonl" ) with open(prefix_path, "wb") as f: while True: data = queue.get() if data is None: break # print(data[:16]) parsed = processor.convert(data) if parsed: f.write(parsed) f.write(b"\n") app = typer.Typer() def err_cb(err: Exception): print(traceback.format_exception(err)[0]) @app.command() def main(folder: pathlib.Path, prefix_path: pathlib.Path, processes: int = 64): # orjson.loads() with multiprocessing.Pool(processes=processes) as pool: workers = [ pool.apply_async( worker, args=(worker_idx, prefix_path), error_callback=err_cb ) for worker_idx in range(processes) ] for file in pathlib.Path(folder).glob("*.ndjson"): with open(file, "rb") as fp: pbar = tqdm.tqdm(desc=f"{file.name}") for line in fp: # print(line[:16]) queue.put(line) pbar.update(1) print("processed", file.name) pbar.close() for _ in range(processes): queue.put(None) while not queue.empty(): print("Waiting for empty queue") time.sleep(5) print("Queue is empty. Waiting for threads to finish.") patience = 120 # processor_thr = 0.05 while True: workers_done = 0 for running_worker in workers: if running_worker.ready(): workers_done += 1 if workers_done == processes: break if patience <= 0: # Screw it. pool.terminate() break else: if workers_done > processes // 1.5: print("Waiting for dangling processes to complete.") time.sleep(10) patience -= 1 # pool.join() if __name__ == "__main__": app()