"""major_projects_grabber.py Download documents from the NSW DPE Major Projects website. Example usage: # Grab a single project modification using its job id, and save in 'files' major_projects_grabber -i 1746 -o files # Grab all modifications in search results page, and save in current folder major_projects_grabber -u "http://majorprojects.planning.nsw.gov.au/index.pl?action=search&authority_id=547" """ import os import re import sys import shutil import logging import requests import argparse import warnings from lxml import html from tqdm import tqdm from requests.exceptions import ConnectionError, InvalidURL def make_safe(s): """Remove characters that would be invalid in a filepath""" # Remove trailing spaces s_safe = s.strip() # Remove '\', '*', '"', '<', '>' '|' s_safe = re.sub('\\\|\*|"|<|>\|', '', s_safe) # Replace '/' and ':' with '-' s_safe = re.sub(':', ' -', s_safe) s_safe = re.sub('/', '-', s_safe) return s_safe def mod_ids_from_search(search_results_url): """Get modification job IDs from search results URL""" # Get HTML of search results page page = requests.get(search_results_url) # Create HTML tree tree = html.fromstring(page.content) # Find job ids of items in results list mod_ids = [] mods = tree.xpath('//*[@class="vpa_list"]/tr/td[1]/a') for mod in mods: mod_ids.append(re.search('(?<=job_id=)\d+', mod.get('href')).group()) return mod_ids def get_document_list(mod_id, output_dir): """Get list of documents from project modification ID""" # Get html from mod page mod_url = ('http://majorprojects.planning.nsw.gov.au/' 'index.pl?action=view_job&job_id=' + mod_id) mod_page = requests.get(mod_url) mod_tree = html.fromstring(mod_page.content) # Get mod details project_name = mod_tree.xpath( '//*[@class="vpa_header vpa_header_contrast"]/h2')[0].text mod_name = mod_tree.xpath( '//*[@class="vpa_header vpa_header_contrast"]/h1')[0].text # Get list of document folders folders = mod_tree.xpath('//div[@class="folder_row"]') # Remove invalid characters before creating folders project_name = make_safe(project_name) mod_name = make_safe(mod_name) # Create modification folder mod_dir = os.path.join(output_dir, project_name, mod_name) try: os.makedirs(mod_dir, exist_ok=True) except FileNotFoundError: # Fix destination path if longer than 255 characters (Windows only) mod_dir = '\\\\?\\' + os.path.abspath(mod_dir) os.makedirs(mod_dir, exist_ok=True) # Add note if no documents are found on portal if not folders: txt_name = 'No documents on DPE portal for this modification.txt' open(os.path.join(mod_dir, txt_name), 'a').close() # Create link to DPE Major Projects page for current modification text = """ """.format(mod_url) with open(os.path.join(mod_dir, 'DPE-portal-page.html'), 'w') as f: f.write(text) document_data = [] for folder in folders: folder_name = folder.xpath('a[2]')[0].text.strip() # Get documents in current folder documents = folder.xpath('ul/li/a') for document in documents: doc = {} doc['url'] = document.get('href') doc['name'] = document.text doc['document_path'] = os.path.join( output_dir, project_name, mod_name, folder_name, doc['name']) document_data.append(doc) return document_data def download_document(url, document_path): """Download document from given url""" # Check if destination path is too long (Windows filename limitation) try: open(document_path, 'a').close() except FileNotFoundError: document_path = '\\\\?\\' + os.path.abspath(document_path) # Create output directories as required os.makedirs(os.path.dirname(document_path), exist_ok=True) # Check if file exists if os.path.isfile(document_path): pass else: try: # Attempt to download file r = requests.get(url, stream=True) except (ConnectionError, InvalidURL): logging.error( ('Failed to download {4}\n' ' Project: {1}\n' ' Modification: {2}\n' ' Folder: {3}\n').format(*document_path.split(os.sep))) return # Write file to disk with open(document_path, 'wb') as f: shutil.copyfileobj(r.raw, f) def main(): example_text = """examples: # Grab a single project modification using its job id, and save in 'files' major_projects_grabber -i 1746 -o files # Grab all modifications in search results page, and save in current folder major_projects_grabber -u "http://majorprojects.planning.nsw.gov.au/index.pl?action=search&authority_id=547" """ # Set up command line arguments parser = argparse.ArgumentParser( epilog=example_text, formatter_class=argparse.RawDescriptionHelpFormatter) parser.add_argument( '-o', metavar='OUTPUT_DIR', default='.', help='root output directory') parser.add_argument( '-i', metavar='ID', default=[], help='modification job id(s)', nargs='*') parser.add_argument('-u', metavar='URL', help='url of search results page') # Print usage if no arguments are provided if len(sys.argv) == 1: parser.print_help(sys.stderr) sys.exit(1) # Parse arguments args = parser.parse_args() search_results_url = args.u output_dir = args.o mod_ids = args.i # Set up log File os.makedirs(output_dir, exist_ok=True) log_name = os.path.join(output_dir, 'errors.log') logging.basicConfig(filename=log_name, level=logging.ERROR) # Get mod IDs from search results if search_results_url: search_mod_ids = mod_ids_from_search(search_results_url) mod_ids.extend(search_mod_ids) # Get list of documents from given modification IDs mod_pbar = tqdm(mod_ids) for mod_id in mod_pbar: document_data = get_document_list(mod_id, output_dir) # Download documents for current modification doc_pbar = tqdm(document_data) for doc in doc_pbar: # Update progress bars mod_name = doc['document_path'].split(os.sep)[-3] doc_name = doc['document_path'].split(os.sep)[-1] mod_pbar.set_description(mod_name) doc_pbar.set_description(doc_name) # Download document download_document(doc['url'], doc['document_path']) # Tidy up console after tqdm print('\n') # Stop logging logging.shutdown() with open(log_name, 'r') as f: log_data = f.read() # Check contents of log file if log_data: warnings.warn( 'Some files failed to download. See log for details.', stacklevel=2) else: os.remove(log_name) if __name__ == '__main__': main()