# Created By: Virgil Dupras # Created On: 2006/11/11 # Copyright 2013 Hardcoded Software (http://www.hardcoded.net) # # This software is licensed under the "BSD" License as described in the "LICENSE" file, # which should be included with this package. The terms are also available at # http://www.hardcoded.net/licenses/bsd_license import os import os.path as op import logging import subprocess import re import time import shutil from send2trash import send2trash from hscommon.reg import RegistrableApplication from hscommon.notify import Broadcaster from hscommon.path import Path from hscommon.conflict import smart_move, smart_copy from hscommon.util import (delete_if_empty, first, escape, nonone, format_time_decimal, allsame, rem_file_ext) from hscommon.trans import tr from . import directories, results, scanner, export, fs from .gui.deletion_options import DeletionOptions from .gui.details_panel import DetailsPanel from .gui.directory_tree import DirectoryTree from .gui.ignore_list_dialog import IgnoreListDialog from .gui.problem_dialog import ProblemDialog from .gui.stats_label import StatsLabel HAD_FIRST_LAUNCH_PREFERENCE = 'HadFirstLaunch' DEBUG_MODE_PREFERENCE = 'DebugMode' MSG_NO_MARKED_DUPES = tr("There are no marked duplicates. Nothing has been done.") MSG_NO_SELECTED_DUPES = tr("There are no selected duplicates. Nothing has been done.") MSG_MANY_FILES_TO_OPEN = tr("You're about to open many files at once. Depending on what those " "files are opened with, doing so can create quite a mess. Continue?") class DestType: Direct = 0 Relative = 1 Absolute = 2 class JobType: Scan = 'job_scan' Load = 'job_load' Move = 'job_move' Copy = 'job_copy' Delete = 'job_delete' def format_timestamp(t, delta): if delta: return format_time_decimal(t) else: if t > 0: return time.strftime('%Y/%m/%d %H:%M:%S', time.localtime(t)) else: return '---' def format_words(w): def do_format(w): if isinstance(w, list): return '(%s)' % ', '.join(do_format(item) for item in w) else: return w.replace('\n', ' ') return ', '.join(do_format(item) for item in w) def format_perc(p): return "%0.0f" % p def format_dupe_count(c): return str(c) if c else '---' def cmp_value(dupe, attrname): if attrname == 'name': value = rem_file_ext(dupe.name) else: value = getattr(dupe, attrname, '') return value.lower() if isinstance(value, str) else value class DupeGuru(RegistrableApplication, Broadcaster): #--- View interface # open_path(path) # reveal_path(path) # start_job(jobid, func, args=()) ( func(j, *args) ) # ask_yes_no(prompt) --> bool # show_results_window() # show_problem_dialog() # select_dest_folder(prompt: str) --> str # select_dest_file(prompt: str, ext: str) --> str # in fairware prompts, we don't mention the edition, it's too long. PROMPT_NAME = "dupeGuru" DEMO_LIMITATION = tr("will only be able to delete, move or copy 10 duplicates at once") def __init__(self, view, appdata): if view.get_default(DEBUG_MODE_PREFERENCE): logging.getLogger().setLevel(logging.DEBUG) logging.debug("Debug mode enabled") RegistrableApplication.__init__(self, view, appid=1) Broadcaster.__init__(self) self.appdata = appdata if not op.exists(self.appdata): os.makedirs(self.appdata) self.directories = directories.Directories() self.results = results.Results(self) self.scanner = scanner.Scanner() self.options = { 'escape_filter_regexp': True, 'clean_empty_dirs': False, 'ignore_hardlink_matches': False, 'copymove_dest_type': DestType.Relative, } self.selected_dupes = [] self.details_panel = DetailsPanel(self) self.directory_tree = DirectoryTree(self) self.problem_dialog = ProblemDialog(self) self.ignore_list_dialog = IgnoreListDialog(self) self.stats_label = StatsLabel(self) self.result_table = self._create_result_table() self.deletion_options = DeletionOptions() children = [self.result_table, self.directory_tree, self.stats_label, self.details_panel] for child in children: child.connect() #--- Virtual def _get_display_info(self, dupe, group, delta): raise NotImplementedError() def _prioritization_categories(self): raise NotImplementedError() def _create_result_table(self): raise NotImplementedError() #--- Private def _get_dupe_sort_key(self, dupe, get_group, key, delta): if key == 'percentage': m = get_group().get_match_of(dupe) return m.percentage if key == 'dupe_count': return 0 if key == 'marked': return self.results.is_marked(dupe) r = cmp_value(dupe, key) if delta and (key in self.result_table.DELTA_COLUMNS): r -= cmp_value(get_group().ref, key) return r def _get_group_sort_key(self, group, key): if key == 'percentage': return group.percentage if key == 'dupe_count': return len(group) if key == 'marked': return len([dupe for dupe in group.dupes if self.results.is_marked(dupe)]) return cmp_value(group.ref, key) def _do_delete(self, j, link_deleted, use_hardlinks, direct_deletion): def op(dupe): j.add_progress() return self._do_delete_dupe(dupe, link_deleted, use_hardlinks, direct_deletion) j.start_job(self.results.mark_count) self.results.perform_on_marked(op, True) def _do_delete_dupe(self, dupe, link_deleted, use_hardlinks, direct_deletion): if not dupe.path.exists(): return logging.debug("Sending '%s' to trash", dupe.path) str_path = str(dupe.path) if direct_deletion: if op.isdir(str_path): shutil.rmtree(str_path) else: os.remove(str_path) else: send2trash(str_path) # Raises OSError when there's a problem if link_deleted: group = self.results.get_group_of_duplicate(dupe) ref = group.ref linkfunc = os.link if use_hardlinks else os.symlink linkfunc(str(ref.path), str_path) self.clean_empty_dirs(dupe.path[:-1]) def _create_file(self, path): # We add fs.Folder to fileclasses in case the file we're loading contains folder paths. return fs.get_file(path, self.directories.fileclasses + [fs.Folder]) def _get_file(self, str_path): path = Path(str_path) f = self._create_file(path) if f is None: return None try: f._read_all_info(attrnames=self.METADATA_TO_READ) return f except EnvironmentError: return None def _get_export_data(self): columns = [col for col in self.result_table.columns.ordered_columns if col.visible and col.name != 'marked'] colnames = [col.display for col in columns] rows = [] for group_id, group in enumerate(self.results.groups): for dupe in group: data = self.get_display_info(dupe, group) row = [data[col.name] for col in columns] row.insert(0, group_id) rows.append(row) return colnames, rows def _results_changed(self): self.selected_dupes = [d for d in self.selected_dupes if self.results.get_group_of_duplicate(d) is not None] self.notify('results_changed') def _job_completed(self, jobid, exc): # Must be called by subclasses when they detect that an async job is completed. If an # exception was raised during the job, `exc` will be set. Return True when the error was # handled. If we return False when exc is set, a the exception will be re-raised. if exc is not None: return False # We don't handle any exception in here if jobid == JobType.Scan: self._results_changed() if not self.results.groups: self.view.show_message(tr("No duplicates found.")) else: self.view.show_results_window() if jobid in {JobType.Load, JobType.Move, JobType.Delete}: self._results_changed() if jobid == JobType.Load: self.view.show_results_window() if jobid in {JobType.Copy, JobType.Move, JobType.Delete}: if self.results.problems: self.problem_dialog.refresh() self.view.show_problem_dialog() else: msg = { JobType.Copy: tr("All marked files were copied successfully."), JobType.Move: tr("All marked files were moved successfully."), JobType.Delete: tr("All marked files were successfully sent to Trash."), }[jobid] self.view.show_message(msg) @staticmethod def _remove_hardlink_dupes(files): seen_inodes = set() result = [] for file in files: try: inode = file.path.stat().st_ino except OSError: # The file was probably deleted or something continue if inode not in seen_inodes: seen_inodes.add(inode) result.append(file) return result def _select_dupes(self, dupes): if dupes == self.selected_dupes: return self.selected_dupes = dupes self.notify('dupes_selected') def _check_demo(self): if self.should_apply_demo_limitation and self.results.mark_count > 10: msg = tr("You cannot delete, move or copy more than 10 duplicates at once in demo mode.") self.view.show_message(msg) return False return True #--- Public def add_directory(self, d): try: self.directories.add_path(Path(d)) self.notify('directories_changed') except directories.AlreadyThereError: self.view.show_message(tr("'{}' already is in the list.").format(d)) except directories.InvalidPathError: self.view.show_message(tr("'{}' does not exist.").format(d)) def add_selected_to_ignore_list(self): dupes = self.without_ref(self.selected_dupes) if not dupes: self.view.show_message(MSG_NO_SELECTED_DUPES) return msg = tr("All selected %d matches are going to be ignored in all subsequent scans. Continue?") if not self.view.ask_yes_no(msg % len(dupes)): return for dupe in dupes: g = self.results.get_group_of_duplicate(dupe) for other in g: if other is not dupe: self.scanner.ignore_list.Ignore(str(other.path), str(dupe.path)) self.remove_duplicates(dupes) self.ignore_list_dialog.refresh() def apply_filter(self, filter): self.results.apply_filter(None) if self.options['escape_filter_regexp']: filter = escape(filter, set('()[]\\.|+?^')) filter = escape(filter, '*', '.') self.results.apply_filter(filter) self._results_changed() def clean_empty_dirs(self, path): if self.options['clean_empty_dirs']: while delete_if_empty(path, ['.DS_Store']): path = path[:-1] def copy_or_move(self, dupe, copy: bool, destination: str, dest_type: DestType): source_path = dupe.path location_path = first(p for p in self.directories if dupe.path in p) dest_path = Path(destination) if dest_type in {DestType.Relative, DestType.Absolute}: # no filename, no windows drive letter source_base = source_path.remove_drive_letter()[:-1] if dest_type == DestType.Relative: source_base = source_base[location_path:] dest_path = dest_path + source_base if not dest_path.exists(): dest_path.makedirs() # Add filename to dest_path. For file move/copy, it's not required, but for folders, yes. dest_path = dest_path + source_path[-1] logging.debug("Copy/Move operation from '%s' to '%s'", source_path, dest_path) # Raises an EnvironmentError if there's a problem if copy: smart_copy(source_path, dest_path) else: smart_move(source_path, dest_path) self.clean_empty_dirs(source_path[:-1]) def copy_or_move_marked(self, copy): def do(j): def op(dupe): j.add_progress() self.copy_or_move(dupe, copy, destination, desttype) j.start_job(self.results.mark_count) self.results.perform_on_marked(op, not copy) if not self._check_demo(): return if not self.results.mark_count: self.view.show_message(MSG_NO_MARKED_DUPES) return opname = tr("copy") if copy else tr("move") prompt = tr("Select a directory to {} marked files to").format(opname) destination = self.view.select_dest_folder(prompt) if destination: desttype = self.options['copymove_dest_type'] jobid = JobType.Copy if copy else JobType.Move self.view.start_job(jobid, do) def delete_marked(self): if not self._check_demo(): return if not self.results.mark_count: self.view.show_message(MSG_NO_MARKED_DUPES) return if not self.deletion_options.show(self.results.mark_count): return args = [self.deletion_options.link_deleted, self.deletion_options.use_hardlinks, self.deletion_options.direct] logging.debug("Starting deletion job with args %r", args) self.view.start_job(JobType.Delete, self._do_delete, args=args) def export_to_xhtml(self): colnames, rows = self._get_export_data() export_path = export.export_to_xhtml(colnames, rows) self.view.open_path(export_path) def export_to_csv(self): dest_file = self.view.select_dest_file(tr("Select a destination for your exported CSV"), 'csv') if dest_file: colnames, rows = self._get_export_data() export.export_to_csv(dest_file, colnames, rows) def get_display_info(self, dupe, group, delta=False): def empty_data(): return {c.name: '---' for c in self.result_table.COLUMNS[1:]} if (dupe is None) or (group is None): return empty_data() try: return self._get_display_info(dupe, group, delta) except Exception as e: logging.warning("Exception on GetDisplayInfo for %s: %s", str(dupe.path), str(e)) return empty_data() def invoke_custom_command(self): """Calls command in 'CustomCommand' pref with %d and %r placeholders replaced. Using the current selection, %d is replaced with the currently selected dupe and %r is replaced with that dupe's ref file. If there's no selection, the command is not invoked. If the dupe is a ref, %d and %r will be the same. """ cmd = self.view.get_default('CustomCommand') if not cmd: msg = tr("You have no custom command set up. Set it up in your preferences.") self.view.show_message(msg) return if not self.selected_dupes: return dupe = self.selected_dupes[0] group = self.results.get_group_of_duplicate(dupe) ref = group.ref cmd = cmd.replace('%d', str(dupe.path)) cmd = cmd.replace('%r', str(ref.path)) match = re.match(r'"([^"]+)"(.*)', cmd) if match is not None: # This code here is because subprocess. Popen doesn't seem to accept, under Windows, # executable paths with spaces in it, *even* when they're enclosed in "". So this is # a workaround to make the damn thing work. exepath, args = match.groups() path, exename = op.split(exepath) subprocess.Popen(exename + args, shell=True, cwd=path) else: subprocess.Popen(cmd, shell=True) def load(self): self.directories.load_from_file(op.join(self.appdata, 'last_directories.xml')) self.notify('directories_changed') p = op.join(self.appdata, 'ignore_list.xml') self.scanner.ignore_list.load_from_xml(p) self.ignore_list_dialog.refresh() def load_from(self, filename): def do(j): self.results.load_from_xml(filename, self._get_file, j) self.view.start_job(JobType.Load, do) def make_selected_reference(self): dupes = self.without_ref(self.selected_dupes) changed_groups = set() for dupe in dupes: g = self.results.get_group_of_duplicate(dupe) if g not in changed_groups: self.results.make_ref(dupe) changed_groups.add(g) # It's not always obvious to users what this action does, so to make it a bit clearer, # we change our selection to the ref of all changed groups. However, we also want to keep # the files that were ref before and weren't changed by the action. In effect, what this # does is that we keep our old selection, but remove all non-ref dupes from it. self.selected_dupes = [d for d in self.selected_dupes if self.results.get_group_of_duplicate(d).ref is d] self.notify('results_changed') def mark_all(self): self.results.mark_all() self.notify('marking_changed') def mark_none(self): self.results.mark_none() self.notify('marking_changed') def mark_invert(self): self.results.mark_invert() self.notify('marking_changed') def mark_dupe(self, dupe, marked): if marked: self.results.mark(dupe) else: self.results.unmark(dupe) self.notify('marking_changed') def open_selected(self): if len(self.selected_dupes) > 10: if not self.view.ask_yes_no(MSG_MANY_FILES_TO_OPEN): return for dupe in self.selected_dupes: self.view.open_path(dupe.path) def purge_ignore_list(self): self.scanner.ignore_list.Filter(lambda f,s:op.exists(f) and op.exists(s)) self.ignore_list_dialog.refresh() def remove_directories(self, indexes): try: indexes = sorted(indexes, reverse=True) for index in indexes: del self.directories[index] self.notify('directories_changed') except IndexError: pass def remove_duplicates(self, duplicates): self.results.remove_duplicates(self.without_ref(duplicates)) self.notify('results_changed_but_keep_selection') def remove_marked(self): if not self.results.mark_count: self.view.show_message(MSG_NO_MARKED_DUPES) return msg = tr("You are about to remove %d files from results. Continue?") if not self.view.ask_yes_no(msg % self.results.mark_count): return self.results.perform_on_marked(lambda x:None, True) self._results_changed() def remove_selected(self): dupes = self.without_ref(self.selected_dupes) if not dupes: self.view.show_message(MSG_NO_SELECTED_DUPES) return msg = tr("You are about to remove %d files from results. Continue?") if not self.view.ask_yes_no(msg % len(dupes)): return self.remove_duplicates(dupes) def rename_selected(self, newname): try: d = self.selected_dupes[0] d.rename(newname) return True except (IndexError, fs.FSError) as e: logging.warning("dupeGuru Warning: %s" % str(e)) return False def reprioritize_groups(self, sort_key): count = 0 for group in self.results.groups: if group.prioritize(key_func=sort_key): count += 1 self._results_changed() msg = tr("{} duplicate groups were changed by the re-prioritization.").format(count) self.view.show_message(msg) def reveal_selected(self): if self.selected_dupes: self.view.reveal_path(self.selected_dupes[0].path) def save(self): if not op.exists(self.appdata): os.makedirs(self.appdata) self.directories.save_to_file(op.join(self.appdata, 'last_directories.xml')) p = op.join(self.appdata, 'ignore_list.xml') self.scanner.ignore_list.save_to_xml(p) self.notify('save_session') def save_as(self, filename): self.results.save_to_xml(filename) def start_scanning(self): def do(j): j.set_progress(0, tr("Collecting files to scan")) if self.scanner.scan_type == scanner.ScanType.Folders: files = list(self.directories.get_folders(j)) else: files = list(self.directories.get_files(j)) if self.options['ignore_hardlink_matches']: files = self._remove_hardlink_dupes(files) logging.info('Scanning %d files' % len(files)) self.results.groups = self.scanner.get_dupe_groups(files, j) if not self.directories.has_any_file(): self.view.show_message(tr("The selected directories contain no scannable file.")) return self.results.groups = [] self._results_changed() self.view.start_job(JobType.Scan, do) def toggle_selected_mark_state(self): selected = self.without_ref(self.selected_dupes) if not selected: return if allsame(self.results.is_marked(d) for d in selected): markfunc = self.results.mark_toggle else: markfunc = self.results.mark for dupe in selected: markfunc(dupe) self.notify('marking_changed') def without_ref(self, dupes): return [dupe for dupe in dupes if self.results.get_group_of_duplicate(dupe).ref is not dupe] def get_default(self, key, fallback_value=None): result = nonone(self.view.get_default(key), fallback_value) if fallback_value is not None and not isinstance(result, type(fallback_value)): # we don't want to end up with garbage values from the prefs try: result = type(fallback_value)(result) except Exception: result = fallback_value return result def set_default(self, key, value): self.view.set_default(key, value) #--- Properties @property def stat_line(self): result = self.results.stat_line if self.scanner.discarded_file_count: result = tr("%s (%d discarded)") % (result, self.scanner.discarded_file_count) return result