#!/usr/bin/env python3 ### NZBGET SCAN SCRIPT # Extract filenames from subjects containing [PRiVATE]-[WtFnZb] # # This extensions extracts obfuscated filenames from .nzb files # created by WtFnZb. # # Supported subject formats: # # - [PRiVATE]-[WtFnZb]-[filename]-[1/5] - "" yEnc 0 (1/1)" # # - [PRiVATE]-[WtFnZb]-[5]-[1/filename] - "" yEnc # # # NOTE: Requires Python and lxml (sudo apt install python3-lxml python-lxml) # ### NZBGET SCAN SCRIPT import sys import os import re # Exit codes used by NZBGet POSTPROCESS_SUCCESS = 93 POSTPROCESS_NONE = 95 POSTPROCESS_ERROR = 94 try: from lxml import etree except ImportError: print(u'[ERROR] Python lxml required. Please install with "sudo apt install python-lxml" or "pip install lxml".') sys.exit(POSTPROCESS_ERROR) patterns = ( re.compile(r'^(?P<prefix>.*\[PRiVATE\]-\[WtFnZb\]-)' r'\[(?P<total>\d+)\]-\[(?P<segment>\d+)\/(?P<filename>.{3,}?)\]' r'\s+-\s+""\s+yEnc\s+', re.MULTILINE | re.UNICODE), re.compile(r'^(?P<prefix>.*\[PRiVATE\]-\[WtFnZb\]-)' r'\[(?P<filename>.{3,}?)\]-\[(?P<segment>\d+)/(?P<total>\d+)\]' r'\s+-\s+""\s+yEnc\s+', re.MULTILINE | re.UNICODE)) nzb_dir = os.getenv('NZBNP_DIRECTORY') nzb_filename = os.getenv('NZBNP_FILENAME') nzb_name = os.getenv('NZBNP_NZBNAME') nzb_file_naming = os.getenv('NZBOP_FILENAMING') if nzb_dir is None or nzb_filename is None or nzb_name is None: print('Please run as NZBGet plugin') sys.exit(POSTPROCESS_ERROR) if nzb_file_naming is not None and nzb_file_naming.lower() != 'nzb': print(u'[ERROR] NZBGet setting FileNaming (under Download Queue) ' u'must be set to "Nzb" for this extension to work correctly, exiting.') sys.exit(POSTPROCESS_ERROR) if not os.path.exists(nzb_dir): print('[ERROR] NZB directory doesn\'t exist, exiting') sys.exit(POSTPROCESS_ERROR) if not nzb_filename.lower().endswith('.nzb'): print(u'[ERROR] {} is not a .nzb file.'.format(nzb_filename)) sys.exit(POSTPROCESS_ERROR) nzb = os.path.join(nzb_dir, nzb_filename) if not os.path.exists(nzb): print('[ERROR] {nzb} doesn\'t exist, exiting'.format(nzb=nzb)) sys.exit(POSTPROCESS_ERROR) with open(nzb, mode='rb') as infile: tree = etree.parse(infile) changed = False file_count = 0 totals = set() filenames = set() for f in tree.getiterator('{http://www.newzbin.com/DTD/2003/nzb}file'): subject = f.get('subject') if subject is None: print(u'[DETAIL] No subject in <file>, skipping') continue file_count += 1 result = [re.match(pattern, subject) for pattern in patterns] matched = [m for m in result if m is not None] if len(matched) == 0: print(u'[INFO] No pattern matching subject, exiting.') sys.exit(POSTPROCESS_NONE) elif len(matched) > 1: print(u'[ERROR] Multiple patterns matched, exiting.') sys.exit(POSTPROCESS_ERROR) else: match = matched[0].groupdict() if match['filename'].lower().endswith('.par2'): print(u'[INFO] par2 exists, exiting') sys.exit(POSTPROCESS_NONE) if int(match['segment']) > int(match['total']): print(u'[DETAIL] Segment index is greater then total, skipping') continue # NZBGet subject parsing changes when duplicate filenames are present # prefix duplicates to avoid that if match['filename'] in filenames: match['filename'] = u'{}.{}'.format(file_count, match['filename']) filenames.add(match['filename']) s = u'WtFnZb "{filename}" yEnc ({segment}/{total})'.format( filename = match['filename'], segment = match['segment'], total = match['total']) print(u'[INFO] New subject {subject}'.format(subject=s.encode('ascii', 'ignore'))) f.set('subject', s) changed = True totals.add(int(match['total'])) if not changed: print(u'[WARNING] No subject changed, exiting.') sys.exit(POSTPROCESS_NONE) if len(totals) != 1: print(u'[WARNING] Mixed values for number of total segments, exiting.') sys.exit(POSTPROCESS_NONE) if totals.pop() != file_count: print(u'[WARNING] Listed segment count does not match <file> count, exiting.') sys.exit(POSTPROCESS_NONE) org = u'{}.wtfnzb.original.processed'.format(nzb) exists_counter = 0 while os.path.exists(org): exists_counter += 1 org = u'{}.{}.wtfnzb.original.processed'.format(nzb, exists_counter) print(u'[INFO] Preserving original nzb as {}'.format(org)) os.rename(nzb, org) print(u'[INFO] Writing {}'.format(nzb)) with open(nzb, mode='wb') as outfile: outfile.write(etree.tostring(tree, xml_declaration=True, encoding=tree.docinfo.encoding, doctype=tree.docinfo.doctype)) sys.exit(POSTPROCESS_SUCCESS)