import json import os from aformatter import format_all_entries, per_line_entries import traceback import scale_back from scale_back import ScaleConfig out_ids = [] atlas_ids = {} authors = [] with open('../web/all-authors.txt', 'r', encoding='utf-8') as authors_file: authors = authors_file.read().strip().split() with open('../data/read-ids.txt', 'r', encoding='utf-8') as ids_file: out_ids = ids_file.read().strip().split() with open('../web/atlas.json', 'r', encoding='utf-8') as atlas_file: atlas_data = json.loads(atlas_file.read()) last_id = 0 for i, entry in enumerate(atlas_data): atlas_ids[entry['id']] = i id = entry['id'] if id.isnumeric() and int(id) > last_id and int(id) - last_id < 100: last_id = int(id) patches_dir = "../data/patches/" permanent_patch_file = "temp-atlas.json" if not os.path.exists(patches_dir): print("Patches folder not found. Exiting.") exit() base_image_path = os.path.join('..', 'web', '_img', 'canvas', 'place30') ScaleConfig.image1 = os.path.join(base_image_path, '159.png') filenames = os.listdir(patches_dir) filenames.append(permanent_patch_file) for filename in filenames: is_permanent_file = filename == permanent_patch_file if is_permanent_file: f = filename else: f = os.path.join(patches_dir, filename) print(f"{filename}: Processing...") if not os.path.isfile(f) or not f.endswith('json'): continue try: with open(f, 'r', encoding='utf-8') as entry_file: entries = json.loads(entry_file.read()) if not isinstance(entries, list): entries = [entries] format_all_entries(entries) scale_back.swap_source_dest('164', '165', os.path.join(base_image_path, '163_159.png')) scale_back.scale_back_entries(entries) scale_back.swap_source_dest('165', '166', os.path.join(base_image_path, '164_159.png')) scale_back.scale_back_entries(entries) scale_back.swap_source_dest('166', '167', os.path.join(base_image_path, '165_159.png')) scale_back.scale_back_entries(entries) for entry in entries: if '_reddit_id' in entry: reddit_id = entry['_reddit_id'] if reddit_id in out_ids: print(f"{filename}: Submission from {entry['id']} has been included! This will be ignored from the merge.") continue out_ids.append(reddit_id) del entry['_reddit_id'] # This wouldn't work if it is an edit. # If needed, we can add a type to the patch to be more foolproof. # if entry['id'] in out_ids: # print(f"{filename}: Submission from {entry['id']} has been included! This will be ignored from the merge.") # continue if '_author' in entry: author = entry['_author'] if author not in authors: authors.append(author) del entry['_author'] if isinstance(entry['id'], int) and entry['id'] < 1 or entry['id'] == '0': last_id += 1 print(f"{filename}: Entry is new, assigned ID {last_id}") entry['id'] = str(last_id) elif isinstance(entry['id'], int): entry['id'] = str(entry['id']) elif not is_permanent_file and entry['id'] not in out_ids: out_ids.append(entry['id']) if entry['id'] in atlas_ids: index = atlas_ids[entry['id']] print(f"{filename}: Edited {atlas_data[index]['id']}.") atlas_data[index] = entry else: print(f"{filename}: Added {entry['id']}.") atlas_data.append(entry) if not is_permanent_file: os.remove(f) except: print(f"{filename}: Something went wrong; patch couldn't be implemented. Skipping.") traceback.print_exc() print('Writing...') with open('../web/atlas.json', 'w', encoding='utf-8') as atlas_file: per_line_entries(atlas_data, atlas_file) with open('../data/read-ids.txt', 'w', encoding='utf-8') as ids_file: ids_file.write("\n".join(out_ids) + "\n") with open('../web/all-authors.txt', 'w', encoding='utf-8') as authors_file: authors_file.write("\n".join(authors) + "\n") print('All done.')