diff --git a/DemoPrograms/Browser_START_HERE_Demo_Programs_Browser.py b/DemoPrograms/Browser_START_HERE_Demo_Programs_Browser.py index 9c887971..6fe12e1d 100644 --- a/DemoPrograms/Browser_START_HERE_Demo_Programs_Browser.py +++ b/DemoPrograms/Browser_START_HERE_Demo_Programs_Browser.py @@ -261,26 +261,28 @@ def find_in_file(string, demo_files_dict, regex=False, verbose=False, window=Non matches = re.finditer(br'^' + bytes(".*("+re.escape(string) + ").*$", 'utf-8'), s, re.MULTILINE) if matches: if show_first_match: - file_list.append(file) - num_files += 1 + #file_list.append(file) + #num_files += 1 match_array = [] matched_str = matches.group(0).decode('utf-8') - if ("==" not in matched_str and "b'" not in matched_str): + if not all(x in matched_str for x in ("b'", '=')) and len(matched_str) < 500: # safe to assume this is not a base64 string as it does not contain the proper ending match_array.append(matches.group(0).decode('utf-8')) matched_dict[full_filename] = match_array + file_list.append(file) + num_files += 1 else: # We need to do this because strings are "falsy" in Python, but empty matches still return True... append_file = False match_array = [] for match_ in matches: - match_str = match_.group(0).decode('utf-8') - if match_str: - if len(match_str) < 500 and "==" not in match_str and "b'" not in match_str: - match_array.append(match_str) - if append_file is False: - append_file = True + matched_str = match_.group(0).decode('utf-8') + if matched_str: + if not all(x in matched_str for x in ("b'", '=')) and len(matched_str) < 500: + # if len(match_str) < 500 and "=" not in match_str and "b'" not in match_str: + match_array.append(matched_str) + append_file = True if append_file: file_list.append(file) num_files += 1