diff options
Diffstat (limited to 'raphodo/thumbnailpara.py')
-rwxr-xr-x | raphodo/thumbnailpara.py | 732 |
1 files changed, 732 insertions, 0 deletions
diff --git a/raphodo/thumbnailpara.py b/raphodo/thumbnailpara.py new file mode 100755 index 0000000..3fedb94 --- /dev/null +++ b/raphodo/thumbnailpara.py @@ -0,0 +1,732 @@ +#!/usr/bin/env python3 + +# Copyright (C) 2011-2017 Damon Lynch <damonlynch@gmail.com> + +# This file is part of Rapid Photo Downloader. +# +# Rapid Photo Downloader is free software: you can redistribute it and/or +# modify it under the terms of the GNU General Public License as published by +# the Free Software Foundation, either version 3 of the License, or +# (at your option) any later version. +# +# Rapid Photo Downloader is distributed in the hope that it will be useful, +# but WITHOUT ANY WARRANTY; without even the implied warranty of +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +# GNU General Public License for more details. +# +# You should have received a copy of the GNU General Public License +# along with Rapid Photo Downloader. If not, +# see <http://www.gnu.org/licenses/>. + +""" +Worker process to get thumbnails from Thumbnail or FDO cache, or +read thumbnail / file from the device being downloaded from. + +For each device, there is one of these workers. + +Sends thumbnail processing tasks to load balancer, which will in turn +send it to extractors. + +By default, will set extractors to get the file's metadata time if +the metadata time is not already found in the rpd_file. +""" + +__author__ = 'Damon Lynch' +__copyright__ = "Copyright 2011-2017, Damon Lynch" + +try: + using_injected = 'profile' in dict(__builtins__) +except: + using_injected = False +finally: + if not using_injected: + # use of line_profiler not detected + def profile(func): + def inner(*args, **kwargs): + return func(*args, **kwargs) + return inner + +import os +import sys +import logging +import pickle +from collections import deque +from operator import attrgetter +from typing import Optional, Tuple, Set + +import zmq +from PyQt5.QtGui import QImage +from PyQt5.QtCore import QSize +import psutil +import gphoto2 as gp +try: + import rawkit + have_rawkit = True +except ImportError: + have_rawkit = False + +from raphodo.rpdfile import RPDFile +from raphodo.interprocess import (WorkerInPublishPullPipeline, + GenerateThumbnailsArguments, + GenerateThumbnailsResults, + ThumbnailExtractorArgument) +from raphodo.constants import (FileType, ThumbnailSize, ThumbnailCacheStatus, + ThumbnailCacheDiskStatus, ExtractionTask, + ExtractionProcessing, orientation_offset, thumbnail_offset, + ThumbnailCacheOrigin, datetime_offset) +from raphodo.camera import Camera, CameraProblemEx +from raphodo.cache import ThumbnailCacheSql, FdoCacheLarge +from raphodo.utilities import (GenerateRandomFileName, create_temp_dir, CacheDirs) +from raphodo.preferences import Preferences +from raphodo.rescan import RescanCamera + + +def split_list(alist: list, wanted_parts=2): + """ + Split list into smaller parts + http://stackoverflow.com/questions/752308/split-list-into-smaller-lists + :param alist: the list + :param wanted_parts: how many lists it should be split into + :return: the split lists + """ + length = len(alist) + return [alist[i * length // wanted_parts: (i + 1) * length // wanted_parts] + for i in range(wanted_parts)] + + +def split_indexes(length: int): + """ + For the length of a list, return a list of indexes into it such + that the indexes start with the middle item, then the middle item + of the remaining two parts of the list, and so forth. + + Perhaps this algorithm could be optimized, as I did it myself. But + hey it works and for now that's the main thing. + + :param length: the length of the list i.e. the number of indexes + to be created + :return: the list of indexes + """ + l = list(range(length)) + n = [] + master = deque([l]) + while master: + l1, l2 = split_list(master.popleft()) + if l2: + n.append(l2[0]) + l2 = l2[1:] + if l1: + master.append(l1) + if l2: + master.append(l2) + return n + + +def get_temporal_gaps_and_sequences(rpd_files, temporal_span): + """ + For a sorted list of rpd_files, identify those rpd_files which are + more than the temporal span away from each other, and those which are + less than the temporal span from each other. + + Does not analyze clusters. + + For instance, you have 1000 photos from a day's photography. You + sort them into a list ordered by time, earliest to latest. You then + get all the photos that were take more than an hour after the + previous photo, and those that were taken within an hour of the + previous photo. + . + :param rpd_files: the sorted list of rpd_files, earliest first + :param temporal_span: the time span that triggers a gap + :return: the rpd_files that signify gaps, and all the rest of the + rpd_files (which are in sequence) + """ + if rpd_files: + prev = rpd_files[0] + gaps = [prev] + sequences = [] + for i, rpd_file in enumerate(rpd_files[1:]): + if rpd_file.modification_time - prev.modification_time > \ + temporal_span: + gaps.append(rpd_file) + else: + sequences.append(rpd_file) + prev = rpd_file + return (gaps, sequences) + return None + + +class GetThumbnailFromCache: + """ + Try to get thumbnail from Rapid Photo Downloader's thumbnail cache + or from the FreeDesktop.org cache. + """ + + def __init__(self, use_thumbnail_cache: bool) -> None: + + if use_thumbnail_cache: + self.thumbnail_cache = ThumbnailCacheSql(create_table_if_not_exists=False) + else: + self.thumbnail_cache = None + + # Access large size Freedesktop.org thumbnail cache + self.fdo_cache_large = FdoCacheLarge() + + self.thumbnail_size_needed = QSize(ThumbnailSize.width, ThumbnailSize.height) + + def image_large_enough(self, size: QSize) -> bool: + """Check if image is equal or bigger than thumbnail size.""" + return (size.width() >= self.thumbnail_size_needed.width() or + size.height() >= self.thumbnail_size_needed.height()) + + def get_from_cache(self, rpd_file: RPDFile, + use_thumbnail_cache: bool=True + ) -> Tuple[ExtractionTask, bytes, str, ThumbnailCacheOrigin]: + """ + Attempt to get a thumbnail for the file from the Rapid Photo Downloader thumbnail cache + or from the FreeDesktop.org 256x256 thumbnail cache. + + :param rpd_file: + :param use_thumbnail_cache: whether to use the + :return: + """ + + task = ExtractionTask.undetermined + thumbnail_bytes = None + full_file_name_to_work_on = '' + origin = None # type: Optional[ThumbnailCacheOrigin] + + # Attempt to get thumbnail from Thumbnail Cache + # (see cache.py for definitions of various caches) + if self.thumbnail_cache is not None and use_thumbnail_cache: + get_thumbnail = self.thumbnail_cache.get_thumbnail_path( + full_file_name=rpd_file.full_file_name, + mtime=rpd_file.modification_time, + size=rpd_file.size, + camera_model=rpd_file.camera_model) + rpd_file.thumbnail_cache_status = get_thumbnail.disk_status + if get_thumbnail.disk_status != ThumbnailCacheDiskStatus.not_found: + origin = ThumbnailCacheOrigin.thumbnail_cache + task = ExtractionTask.bypass + if get_thumbnail.disk_status == ThumbnailCacheDiskStatus.failure: + rpd_file.thumbnail_status = ThumbnailCacheStatus.generation_failed + rpd_file.thumbnail_cache_status = ThumbnailCacheDiskStatus.failure + elif get_thumbnail.disk_status == ThumbnailCacheDiskStatus.found: + rpd_file.thumbnail_cache_status = ThumbnailCacheDiskStatus.found + if get_thumbnail.orientation_unknown: + rpd_file.thumbnail_status = ThumbnailCacheStatus.orientation_unknown + else: + rpd_file.thumbnail_status = ThumbnailCacheStatus.ready + with open(get_thumbnail.path, 'rb') as thumbnail: + thumbnail_bytes = thumbnail.read() + + # Attempt to get thumbnail from large FDO Cache if not found in Thumbnail Cache + # and it's not being downloaded directly from a camera (if it's from a camera, it's + # not going to be in the FDO cache) + + if task == ExtractionTask.undetermined and not rpd_file.from_camera: + get_thumbnail = self.fdo_cache_large.get_thumbnail( + full_file_name=rpd_file.full_file_name, + modification_time=rpd_file.modification_time, + size=rpd_file.size, + camera_model=rpd_file.camera_model) + if get_thumbnail.disk_status == ThumbnailCacheDiskStatus.found: + rpd_file.fdo_thumbnail_256_name = get_thumbnail.path + thumb = get_thumbnail.thumbnail # type: QImage + if thumb is not None: + if self.image_large_enough(thumb.size()): + task = ExtractionTask.load_file_directly + full_file_name_to_work_on = get_thumbnail.path + origin = ThumbnailCacheOrigin.fdo_cache + rpd_file.thumbnail_status = ThumbnailCacheStatus.fdo_256_ready + + return task, thumbnail_bytes, full_file_name_to_work_on, origin + +# How much of the file should be read in from local disk and thus cached +# by they kernel +cached_read = dict( + cr2=260 * 1024, + dng=504 * 1024, + nef=400* 1024 +) + + +def preprocess_thumbnail_from_disk(rpd_file: RPDFile, + processing: Set[ExtractionProcessing]) -> ExtractionTask: + """ + Determine how to get a thumbnail from a photo or video that is not on a camera + (although it may have directly come from there during the download process) + + Does not return the name of the file to be worked on -- that's the responsibility + of the method calling it. + + :param rpd_file: details about file from which to get thumbnail from + :param processing: set that holds processing tasks for the extractors to perform + :return: extraction task required + """ + + if rpd_file.file_type == FileType.photo: + if rpd_file.is_tiff(): + available = psutil.virtual_memory().available + if rpd_file.size <= available: + bytes_to_read = rpd_file.size + if rpd_file.mdatatime: + task = ExtractionTask.load_file_directly + else: + task = ExtractionTask.load_file_and_exif_directly + processing.add(ExtractionProcessing.resize) + else: + # Don't try to extract a thumbnail from + # a file that is larger than available + # memory + task = ExtractionTask.bypass + bytes_to_read = 0 + else: + if rpd_file.is_jpeg() and rpd_file.from_camera and rpd_file.is_mtp_device: + # jpeg photos from smartphones don't have embedded thumbnails + task = ExtractionTask.load_file_and_exif_directly + processing.add(ExtractionProcessing.resize) + else: + task = ExtractionTask.load_from_exif + processing.add(ExtractionProcessing.orient) + bytes_to_read = cached_read.get(rpd_file.extension, 400 * 1024) + + if bytes_to_read: + if not rpd_file.download_full_file_name: + try: + with open(rpd_file.full_file_name, 'rb') as photo: + # Bring the file into the operating system's disk cache + photo.read(bytes_to_read) + except FileNotFoundError: + logging.error( + "The download file %s does not exist", rpd_file.download_full_file_name + ) + else: + # video + if rpd_file.thm_full_name is not None: + if not rpd_file.mdatatime: + task = ExtractionTask.load_file_directly_metadata_from_secondary + # It's the responsibility of the calling code to assign the + # secondary_full_file_name + else: + task = ExtractionTask.load_file_directly + processing.add(ExtractionProcessing.strip_bars_video) + processing.add(ExtractionProcessing.add_film_strip) + else: + if rpd_file.mdatatime: + task = ExtractionTask.extract_from_file + else: + task = ExtractionTask.extract_from_file_and_load_metadata + + return task + + +class GenerateThumbnails(WorkerInPublishPullPipeline): + + def __init__(self) -> None: + self.random_filename = GenerateRandomFileName() + super().__init__('Thumbnails') + + def cache_full_size_file_from_camera(self, rpd_file: RPDFile) -> bool: + """ + Get the file from the camera chunk by chunk and cache it. + + :return: True if operation succeeded, False otherwise + """ + if rpd_file.file_type == FileType.photo: + cache_dir = self.photo_cache_dir + else: + cache_dir = self.video_cache_dir + cache_full_file_name = os.path.join( + cache_dir, '{}.{}'.format(self.random_filename.name(), rpd_file.extension) + ) + try: + self.camera.save_file_by_chunks( + dir_name=rpd_file.path, + file_name=rpd_file.name, + size=rpd_file.size, + dest_full_filename=cache_full_file_name, + progress_callback=None, + check_for_command=self.check_for_controller_directive, + return_file_bytes=False + ) + except CameraProblemEx as e: + #TODO report error + return False + else: + rpd_file.cache_full_file_name = cache_full_file_name + return True + + def cache_file_chunk_from_camera(self, rpd_file: RPDFile, offset: int) -> bool: + if rpd_file.file_type == FileType.photo: + cache_dir = self.photo_cache_dir + else: + cache_dir = self.video_cache_dir + cache_full_file_name = os.path.join( + cache_dir, '{}.{}'.format(self.random_filename.name(), rpd_file.extension) + ) + try: + self.camera.save_file_chunk( + dir_name=rpd_file.path, + file_name=rpd_file.name, + chunk_size_in_bytes=min(offset, rpd_file.size), + dest_full_filename=cache_full_file_name + ) + rpd_file.temp_cache_full_file_chunk = cache_full_file_name + return True + except CameraProblemEx as e: + #TODO problem reporting + return False + + def do_work(self) -> None: + try: + self.generate_thumbnails() + except SystemExit as e: + sys.exit(e) + except: + if hasattr(self, 'device_name'): + logging.error("Exception generating thumbnails for %s", self.device_name) + else: + logging.error("Exception generating thumbnails") + logging.exception("Traceback:") + + def generate_thumbnails(self) -> None: + self.camera = None + arguments = pickle.loads(self.content) # type: GenerateThumbnailsArguments + self.device_name = arguments.name + logging.info("Generating %s thumbnails for %s", len(arguments.rpd_files), arguments.name) + if arguments.log_gphoto2: + gp.use_python_logging() + + self.frontend = self.context.socket(zmq.PUSH) + self.frontend.connect("tcp://localhost:{}".format(arguments.frontend_port)) + + self.prefs = Preferences() + + # If the entire video is required to extract the thumbnail, which is determined + # when extracting sample metadata from a video during the device scan + entire_video_required = arguments.entire_video_required + + # Access and generate Rapid Photo Downloader thumbnail cache + use_thumbnail_cache = self.prefs.use_thumbnail_cache + + thumbnail_caches = GetThumbnailFromCache(use_thumbnail_cache=use_thumbnail_cache) + + photo_cache_dir = video_cache_dir = None + cache_file_from_camera = False + + rpd_files = arguments.rpd_files + + # with open('tests/thumbnail_data_medium_no_tiff', 'wb') as f: + # pickle.dump(rpd_files, f) + + # Must sort files by modification time prior to temporal analysis needed to figure out + # which thumnbnails to prioritize + rpd_files = sorted(rpd_files, key=attrgetter('modification_time')) + + time_span = arguments.proximity_seconds + + rpd_files2 = [] + + if rpd_files: + gaps, sequences = get_temporal_gaps_and_sequences(rpd_files, time_span) + + rpd_files2.extend(gaps) + + indexes = split_indexes(len(sequences)) + rpd_files2.extend([sequences[idx] for idx in indexes]) + + assert len(rpd_files) == len(rpd_files2) + rpd_files = rpd_files2 + + if arguments.camera is not None: + self.camera = Camera( + model=arguments.camera, port=arguments.port, + specific_folders=self.prefs.folders_to_scan + ) + + if not self.camera.camera_initialized: + # There is nothing to do here: exit! + logging.debug( + "Prematurely exiting thumbnail generation due to lack of access to camera %s", + arguments.camera + ) + self.disconnect_logging() + self.send_finished_command() + sys.exit(0) + + must_make_cache_dirs = (not self.camera.can_fetch_thumbnails or cache_file_from_camera) + + if must_make_cache_dirs or arguments.need_video_cache_dir: + # If downloading complete copy of the files to + # generate previews, then may as well cache them to speed up + # the download process + cache_file_from_camera = must_make_cache_dirs + self.photo_cache_dir = create_temp_dir( + folder=arguments.cache_dirs.photo_cache_dir, + prefix='rpd-cache-{}-'.format(self.device_name[:10])) + self.video_cache_dir = create_temp_dir( + folder=arguments.cache_dirs.video_cache_dir, + prefix='rpd-cache-{}-'.format(self.device_name[:10])) + cache_dirs = CacheDirs(self.photo_cache_dir, self.video_cache_dir) + self.content = pickle.dumps(GenerateThumbnailsResults( + scan_id=arguments.scan_id, + cache_dirs=cache_dirs), pickle.HIGHEST_PROTOCOL) + self.send_message_to_sink() + + from_thumb_cache = 0 + from_fdo_cache = 0 + + if self.camera: + rescan = RescanCamera(camera=self.camera, prefs=self.prefs) + rescan.rescan_camera(rpd_files) + rpd_files = rescan.rpd_files + if rescan.missing_rpd_files: + logging.error( + "%s files could not be relocated on %s", + len(rescan.missing_rpd_files), self.camera.display_name + ) + for rpd_file in rescan.missing_rpd_files: # type: RPDFile + self.content = pickle.dumps( + GenerateThumbnailsResults(rpd_file=rpd_file, thumbnail_bytes=None), + pickle.HIGHEST_PROTOCOL + ) + self.send_message_to_sink() + + for rpd_file in rpd_files: # type: RPDFile + # Check to see if the process has received a command + self.check_for_controller_directive() + + exif_buffer = None + file_to_work_on_is_temporary = False + secondary_full_file_name = '' + processing = set() # type: Set[ExtractionProcessing] + + # Attempt to get thumbnail from Thumbnail Cache + # (see cache.py for definitions of various caches) + + cache_search = thumbnail_caches.get_from_cache(rpd_file) + task, thumbnail_bytes, full_file_name_to_work_on, origin = cache_search + if task != ExtractionTask.undetermined: + if origin == ThumbnailCacheOrigin.thumbnail_cache: + from_thumb_cache += 1 + # logging.debug("Thumbnail for %s found in RPD thumbnail cache", + # rpd_file.full_file_name) + else: + assert origin == ThumbnailCacheOrigin.fdo_cache + logging.debug( + "Thumbnail for %s found in large FDO cache", rpd_file.full_file_name + ) + from_fdo_cache += 1 + processing.add(ExtractionProcessing.resize) + if not rpd_file.mdatatime: + # Since we're extracting the thumbnail from the FDO cache, + # need to grab its metadata too. + # Reassign the task + task = ExtractionTask.load_file_directly_metadata_from_secondary + # It's not being downloaded from a camera, so nothing + # special to do except assign the name of the file from which + # to extract the metadata + secondary_full_file_name = rpd_file.full_file_name + logging.debug( + "Although thumbnail found in the cache, tasked to extract " + "file time recorded in metadata from %s", secondary_full_file_name + ) + if task == ExtractionTask.undetermined: + # Thumbnail was not found in any cache: extract it + if self.camera: # type: Camera + if rpd_file.file_type == FileType.photo: + if self.camera.can_fetch_thumbnails: + task = ExtractionTask.load_from_bytes + if rpd_file.is_jpeg_type(): + # gPhoto2 knows how to get jpeg thumbnails + try: + exif_buffer = self.camera.get_exif_extract_from_jpeg( + rpd_file.path, rpd_file.name + ) + except CameraProblemEx as e: + # TODO handle error? + exif_buffer = None + else: + # gPhoto2 does not know how to get RAW thumbnails, so we do that + # part ourselves + if rpd_file.extension == 'crw': + # Could cache this file, since reading its entirety + # But does anyone download a CRW file from the camera these + # days?! + bytes_to_read = rpd_file.size + else: + bytes_to_read = min( + rpd_file.size, orientation_offset.get( + rpd_file.extension, 500 + ) + ) + exif_buffer = self.camera.get_exif_extract( + rpd_file.path, rpd_file.name, bytes_to_read + ) + try: + thumbnail_bytes = self.camera.get_thumbnail( + rpd_file.path, rpd_file.name + ) + except CameraProblemEx as e: + #TODO report error + thumbnail_bytes = None + processing.add(ExtractionProcessing.strip_bars_photo) + processing.add(ExtractionProcessing.orient) + else: + # Many (all?) jpegs from phones don't include jpeg previews, + # so need to render from the entire jpeg itself. Slow! + + # If rawkit is not installed, then extract merely a part of + # phone's raw format, and try to extract the jpeg preview + # from it (which probably doesn't exist!). This is fast. + # If have rawkit, download and render an image from the + # RAW + if not rpd_file.is_jpeg() and not have_rawkit: + bytes_to_read = thumbnail_offset.get(rpd_file.extension) + if bytes_to_read: + exif_buffer = self.camera.get_exif_extract( + rpd_file.path, rpd_file.name, bytes_to_read + ) + task = ExtractionTask.load_from_exif_buffer + processing.add(ExtractionProcessing.orient) + if (task == ExtractionTask.undetermined and + self.cache_full_size_file_from_camera(rpd_file)): + if rpd_file.is_jpeg(): + task = ExtractionTask.load_file_and_exif_directly + processing.add(ExtractionProcessing.resize) + processing.add(ExtractionProcessing.orient) + else: + task = ExtractionTask.load_from_exif + processing.add(ExtractionProcessing.resize) + processing.add(ExtractionProcessing.orient) + full_file_name_to_work_on = rpd_file.cache_full_file_name + else: + # Failed to generate thumbnail + task = ExtractionTask.bypass + else: + # video from camera + if rpd_file.thm_full_name is not None: + # Fortunately, we have a special video thumbnail file + # Still need to get metadata time, however. + + if entire_video_required: + offset = rpd_file.size + else: + offset = datetime_offset.get(rpd_file.extension) + # If there is no offset, there is not point trying to extract the + # metadata time from part of the video. It's not ideal, + # but if this is from a camera on which there were any other files + # we can assume we've got a somewhat accurate date time for it from + # the modification time. + # The only exception is if the video file is not that big, in which + # case it's worth reading in its entirety: + if offset is None and rpd_file.size < 4000000: + offset = rpd_file.size + + if rpd_file.mdatatime or not offset: + task = ExtractionTask.load_from_bytes + elif self.cache_file_chunk_from_camera(rpd_file, offset): + task = ExtractionTask.load_from_bytes_metadata_from_temp_extract + secondary_full_file_name = rpd_file.temp_cache_full_file_chunk + else: + # For some reason was unable to download part of the video file + task = ExtractionTask.load_from_bytes + + try: + thumbnail_bytes = self.camera.get_THM_file(rpd_file.thm_full_name) + except CameraProblemEx as e: + # TODO report error + thumbnail_bytes = None + processing.add(ExtractionProcessing.strip_bars_video) + processing.add(ExtractionProcessing.add_film_strip) + else: + if not entire_video_required: + # For most videos, extract a small part of the video and use + # that to generate thumbnail + offset = thumbnail_offset.get(rpd_file.extension) + if offset: + offset = max(offset, datetime_offset.get(rpd_file.extension)) + if offset and self.cache_file_chunk_from_camera(rpd_file, offset): + task = ExtractionTask.extract_from_file_and_load_metadata + full_file_name_to_work_on = rpd_file.temp_cache_full_file_chunk + file_to_work_on_is_temporary = True + if task == ExtractionTask.undetermined: + if self.cache_full_size_file_from_camera(rpd_file): + task = ExtractionTask.extract_from_file_and_load_metadata + full_file_name_to_work_on = rpd_file.cache_full_file_name + else: + # Failed to generate thumbnail + task = ExtractionTask.bypass + else: + # File is not on a camera + task = preprocess_thumbnail_from_disk(rpd_file=rpd_file, processing=processing) + if task != ExtractionTask.bypass: + if rpd_file.thm_full_name is not None: + full_file_name_to_work_on = rpd_file.thm_full_name + if task == ExtractionTask.load_file_directly_metadata_from_secondary: + secondary_full_file_name = rpd_file.full_file_name + else: + full_file_name_to_work_on = rpd_file.full_file_name + + if task == ExtractionTask.bypass: + self.content = pickle.dumps( + GenerateThumbnailsResults(rpd_file=rpd_file, thumbnail_bytes=thumbnail_bytes), + pickle.HIGHEST_PROTOCOL + ) + self.send_message_to_sink() + + elif task != ExtractionTask.undetermined: + # Send data to load balancer, which will send to one of its + # workers + + self.content = pickle.dumps( + ThumbnailExtractorArgument( + rpd_file=rpd_file, + task=task, + processing=processing, + full_file_name_to_work_on=full_file_name_to_work_on, + secondary_full_file_name=secondary_full_file_name, + exif_buffer=exif_buffer, + thumbnail_bytes = thumbnail_bytes, + use_thumbnail_cache=use_thumbnail_cache, + file_to_work_on_is_temporary=file_to_work_on_is_temporary, + write_fdo_thumbnail=False, + send_thumb_to_main=True + ), + pickle.HIGHEST_PROTOCOL) + self.frontend.send_multipart([b'data', self.content]) + + if arguments.camera: + self.camera.free_camera() + # Delete our temporary cache directories if they are empty + if photo_cache_dir is not None: + if not os.listdir(self.photo_cache_dir): + os.rmdir(self.photo_cache_dir) + if video_cache_dir is not None: + if not os.listdir(self.video_cache_dir): + os.rmdir(self.video_cache_dir) + + logging.debug("Finished phase 1 of thumbnail generation for %s", self.device_name) + if from_thumb_cache: + logging.info( + "{} of {} thumbnails for {} came from thumbnail cache".format( + from_thumb_cache, len(rpd_files), self.device_name + ) + ) + if from_fdo_cache: + logging.info( + "{} of {} thumbnails of for {} came from Free Desktop cache".format( + from_fdo_cache, len(rpd_files), self.device_name + ) + ) + + self.disconnect_logging() + self.send_finished_command() + + def cleanup_pre_stop(self): + if self.camera is not None: + self.camera.free_camera() + +if __name__ == "__main__": + generate_thumbnails = GenerateThumbnails() |