mirror of
https://github.com/mikf/gallery-dl.git
synced 2024-11-23 11:12:40 +01:00
591 lines
19 KiB
Python
591 lines
19 KiB
Python
# -*- coding: utf-8 -*-
|
|
|
|
# Copyright 2015-2020 Mike Fährmann
|
|
#
|
|
# This program is free software; you can redistribute it and/or modify
|
|
# it under the terms of the GNU General Public License version 2 as
|
|
# published by the Free Software Foundation.
|
|
|
|
import sys
|
|
import time
|
|
import errno
|
|
import logging
|
|
from . import extractor, downloader, postprocessor
|
|
from . import config, text, util, output, exception
|
|
from .extractor.message import Message
|
|
|
|
|
|
class Job():
|
|
"""Base class for Job-types"""
|
|
ulog = None
|
|
|
|
def __init__(self, extr, parent=None):
|
|
if isinstance(extr, str):
|
|
extr = extractor.find(extr)
|
|
if not extr:
|
|
raise exception.NoExtractorError()
|
|
self.extractor = extr
|
|
self.pathfmt = None
|
|
|
|
self._logger_extra = {
|
|
"job" : self,
|
|
"extractor": extr,
|
|
"path" : output.PathfmtProxy(self),
|
|
"keywords" : output.KwdictProxy(self),
|
|
}
|
|
extr.log = self._wrap_logger(extr.log)
|
|
extr.log.debug("Using %s for '%s'", extr.__class__.__name__, extr.url)
|
|
|
|
self.status = 0
|
|
self.pred_url = self._prepare_predicates("image", True)
|
|
self.pred_queue = self._prepare_predicates("chapter", False)
|
|
|
|
if parent:
|
|
pextr = parent.extractor
|
|
|
|
# transfer (sub)category
|
|
if pextr.config("category-transfer", pextr.categorytransfer):
|
|
extr.category = pextr.category
|
|
extr.subcategory = pextr.subcategory
|
|
|
|
# transfer parent directory
|
|
extr._parentdir = pextr._parentdir
|
|
|
|
# reuse connection adapters
|
|
extr.session.adapters = pextr.session.adapters
|
|
|
|
# user-supplied metadata
|
|
self.userkwds = self.extractor.config("keywords")
|
|
|
|
def run(self):
|
|
"""Execute or run the job"""
|
|
try:
|
|
log = self.extractor.log
|
|
for msg in self.extractor:
|
|
self.dispatch(msg)
|
|
except exception.StopExtraction as exc:
|
|
if exc.message:
|
|
log.error(exc.message)
|
|
self.status |= exc.code
|
|
except exception.GalleryDLException as exc:
|
|
log.error("%s: %s", exc.__class__.__name__, exc)
|
|
self.status |= exc.code
|
|
except OSError as exc:
|
|
log.error("Unable to download data: %s: %s",
|
|
exc.__class__.__name__, exc)
|
|
log.debug("", exc_info=True)
|
|
self.status |= 128
|
|
except Exception as exc:
|
|
log.error(("An unexpected error occurred: %s - %s. "
|
|
"Please run gallery-dl again with the --verbose flag, "
|
|
"copy its output and report this issue on "
|
|
"https://github.com/mikf/gallery-dl/issues ."),
|
|
exc.__class__.__name__, exc)
|
|
log.debug("", exc_info=True)
|
|
self.status |= 1
|
|
except BaseException:
|
|
self.status |= 1
|
|
raise
|
|
finally:
|
|
self.handle_finalize()
|
|
return self.status
|
|
|
|
def dispatch(self, msg):
|
|
"""Call the appropriate message handler"""
|
|
if msg[0] == Message.Url:
|
|
_, url, kwds = msg
|
|
if self.pred_url(url, kwds):
|
|
self.update_kwdict(kwds)
|
|
self.handle_url(url, kwds)
|
|
|
|
elif msg[0] == Message.Directory:
|
|
self.update_kwdict(msg[1])
|
|
self.handle_directory(msg[1])
|
|
|
|
elif msg[0] == Message.Queue:
|
|
_, url, kwds = msg
|
|
if self.pred_queue(url, kwds):
|
|
self.handle_queue(url, kwds)
|
|
|
|
elif msg[0] == Message.Urllist:
|
|
_, urls, kwds = msg
|
|
if self.pred_url(urls[0], kwds):
|
|
self.update_kwdict(kwds)
|
|
self.handle_urllist(urls, kwds)
|
|
|
|
elif msg[0] == Message.Metadata:
|
|
self.update_kwdict(msg[1])
|
|
self.handle_metadata(msg[1])
|
|
|
|
elif msg[0] == Message.Version:
|
|
if msg[1] != 1:
|
|
raise "unsupported message-version ({}, {})".format(
|
|
self.extractor.category, msg[1]
|
|
)
|
|
# TODO: support for multiple message versions
|
|
|
|
def handle_url(self, url, kwdict):
|
|
"""Handle Message.Url"""
|
|
|
|
def handle_urllist(self, urls, kwdict):
|
|
"""Handle Message.Urllist"""
|
|
self.handle_url(urls[0], kwdict)
|
|
|
|
def handle_directory(self, kwdict):
|
|
"""Handle Message.Directory"""
|
|
|
|
def handle_metadata(self, kwdict):
|
|
"""Handle Message.Metadata"""
|
|
|
|
def handle_queue(self, url, kwdict):
|
|
"""Handle Message.Queue"""
|
|
|
|
def handle_finalize(self):
|
|
"""Handle job finalization"""
|
|
|
|
def update_kwdict(self, kwdict):
|
|
"""Update 'kwdict' with additional metadata"""
|
|
extr = self.extractor
|
|
kwdict["category"] = extr.category
|
|
kwdict["subcategory"] = extr.subcategory
|
|
if self.userkwds:
|
|
kwdict.update(self.userkwds)
|
|
|
|
def _prepare_predicates(self, target, skip=True):
|
|
predicates = []
|
|
|
|
if self.extractor.config(target + "-unique"):
|
|
predicates.append(util.UniquePredicate())
|
|
|
|
pfilter = self.extractor.config(target + "-filter")
|
|
if pfilter:
|
|
try:
|
|
pred = util.FilterPredicate(pfilter, target)
|
|
except (SyntaxError, ValueError, TypeError) as exc:
|
|
self.extractor.log.warning(exc)
|
|
else:
|
|
predicates.append(pred)
|
|
|
|
prange = self.extractor.config(target + "-range")
|
|
if prange:
|
|
try:
|
|
pred = util.RangePredicate(prange)
|
|
except ValueError as exc:
|
|
self.extractor.log.warning(
|
|
"invalid %s range: %s", target, exc)
|
|
else:
|
|
if skip and pred.lower > 1 and not pfilter:
|
|
pred.index += self.extractor.skip(pred.lower - 1)
|
|
predicates.append(pred)
|
|
|
|
return util.build_predicate(predicates)
|
|
|
|
def get_logger(self, name):
|
|
return self._wrap_logger(logging.getLogger(name))
|
|
|
|
def _wrap_logger(self, logger):
|
|
return output.LoggerAdapter(logger, self._logger_extra)
|
|
|
|
def _write_unsupported(self, url):
|
|
if self.ulog:
|
|
self.ulog.info(url)
|
|
|
|
|
|
class DownloadJob(Job):
|
|
"""Download images into appropriate directory/filename locations"""
|
|
|
|
def __init__(self, url, parent=None):
|
|
Job.__init__(self, url, parent)
|
|
self.log = self.get_logger("download")
|
|
self.archive = None
|
|
self.sleep = None
|
|
self.downloaders = {}
|
|
self.postprocessors = None
|
|
self.out = output.select()
|
|
|
|
if parent:
|
|
self.visited = parent.visited
|
|
pfmt = parent.pathfmt
|
|
if pfmt and parent.extractor.config("parent-directory"):
|
|
self.extractor._parentdir = pfmt.directory
|
|
else:
|
|
self.visited = set()
|
|
|
|
def handle_url(self, url, kwdict, fallback=None):
|
|
"""Download the resource specified in 'url'"""
|
|
postprocessors = self.postprocessors
|
|
pathfmt = self.pathfmt
|
|
archive = self.archive
|
|
|
|
# prepare download
|
|
pathfmt.set_filename(kwdict)
|
|
|
|
if postprocessors:
|
|
for pp in postprocessors:
|
|
pp.prepare(pathfmt)
|
|
|
|
if pathfmt.exists(archive):
|
|
self.handle_skip()
|
|
return
|
|
|
|
if self.sleep:
|
|
time.sleep(self.sleep)
|
|
|
|
# download from URL
|
|
if not self.download(url):
|
|
|
|
# use fallback URLs if available
|
|
for num, url in enumerate(fallback or (), 1):
|
|
util.remove_file(pathfmt.temppath)
|
|
self.log.info("Trying fallback URL #%d", num)
|
|
if self.download(url):
|
|
break
|
|
else:
|
|
# download failed
|
|
self.status |= 4
|
|
self.log.error("Failed to download %s",
|
|
pathfmt.filename or url)
|
|
return
|
|
|
|
if not pathfmt.temppath:
|
|
self.handle_skip()
|
|
return
|
|
|
|
# run post processors
|
|
if postprocessors:
|
|
for pp in postprocessors:
|
|
pp.run(pathfmt)
|
|
|
|
# download succeeded
|
|
pathfmt.finalize()
|
|
self.out.success(pathfmt.path, 0)
|
|
if archive:
|
|
archive.add(kwdict)
|
|
if postprocessors:
|
|
for pp in postprocessors:
|
|
pp.run_after(pathfmt)
|
|
self._skipcnt = 0
|
|
|
|
def handle_urllist(self, urls, kwdict):
|
|
"""Download the resource specified in 'url'"""
|
|
fallback = iter(urls)
|
|
url = next(fallback)
|
|
self.handle_url(url, kwdict, fallback)
|
|
|
|
def handle_directory(self, kwdict):
|
|
"""Set and create the target directory for downloads"""
|
|
if not self.pathfmt:
|
|
self.initialize(kwdict)
|
|
else:
|
|
self.pathfmt.set_directory(kwdict)
|
|
|
|
def handle_metadata(self, kwdict):
|
|
"""Run postprocessors with metadata from 'kwdict'"""
|
|
postprocessors = self.postprocessors
|
|
|
|
if postprocessors:
|
|
kwdict["extension"] = "metadata"
|
|
pathfmt = self.pathfmt
|
|
pathfmt.set_filename(kwdict)
|
|
for pp in postprocessors:
|
|
pp.run_metadata(pathfmt)
|
|
|
|
def handle_queue(self, url, kwdict):
|
|
if url in self.visited:
|
|
return
|
|
self.visited.add(url)
|
|
|
|
if "_extractor" in kwdict:
|
|
extr = kwdict["_extractor"].from_url(url)
|
|
else:
|
|
extr = extractor.find(url)
|
|
if extr:
|
|
self.status |= self.__class__(extr, self).run()
|
|
else:
|
|
self._write_unsupported(url)
|
|
|
|
def handle_finalize(self):
|
|
pathfmt = self.pathfmt
|
|
if self.archive:
|
|
self.archive.close()
|
|
if pathfmt:
|
|
self.extractor._store_cookies()
|
|
if self.postprocessors:
|
|
status = self.status
|
|
for pp in self.postprocessors:
|
|
pp.run_final(pathfmt, status)
|
|
|
|
def handle_skip(self):
|
|
self.out.skip(self.pathfmt.path)
|
|
if self._skipexc:
|
|
self._skipcnt += 1
|
|
if self._skipcnt >= self._skipmax:
|
|
raise self._skipexc()
|
|
|
|
def download(self, url):
|
|
"""Download 'url'"""
|
|
scheme = url.partition(":")[0]
|
|
downloader = self.get_downloader(scheme)
|
|
if downloader:
|
|
try:
|
|
return downloader.download(url, self.pathfmt)
|
|
except OSError as exc:
|
|
if exc.errno == errno.ENOSPC:
|
|
raise
|
|
self.log.warning("%s: %s", exc.__class__.__name__, exc)
|
|
return False
|
|
self._write_unsupported(url)
|
|
return False
|
|
|
|
def get_downloader(self, scheme):
|
|
"""Return a downloader suitable for 'scheme'"""
|
|
try:
|
|
return self.downloaders[scheme]
|
|
except KeyError:
|
|
pass
|
|
|
|
cls = downloader.find(scheme)
|
|
if cls and config.get(("downloader", cls.scheme), "enabled", True):
|
|
instance = cls(self)
|
|
else:
|
|
instance = None
|
|
self.log.error("'%s:' URLs are not supported/enabled", scheme)
|
|
|
|
if cls and cls.scheme == "http":
|
|
self.downloaders["http"] = self.downloaders["https"] = instance
|
|
else:
|
|
self.downloaders[scheme] = instance
|
|
return instance
|
|
|
|
def initialize(self, kwdict=None):
|
|
"""Delayed initialization of PathFormat, etc."""
|
|
config = self.extractor.config
|
|
pathfmt = self.pathfmt = util.PathFormat(self.extractor)
|
|
if kwdict:
|
|
pathfmt.set_directory(kwdict)
|
|
|
|
self.sleep = config("sleep")
|
|
if not config("download", True):
|
|
self.download = pathfmt.fix_extension
|
|
|
|
skip = config("skip", True)
|
|
if skip:
|
|
self._skipexc = None
|
|
if skip == "enumerate":
|
|
pathfmt.check_file = pathfmt._enum_file
|
|
elif isinstance(skip, str):
|
|
skip, _, smax = skip.partition(":")
|
|
if skip == "abort":
|
|
self._skipexc = exception.StopExtraction
|
|
elif skip == "exit":
|
|
self._skipexc = sys.exit
|
|
self._skipcnt = 0
|
|
self._skipmax = text.parse_int(smax)
|
|
else:
|
|
pathfmt.exists = lambda x=None: False
|
|
|
|
archive = config("archive")
|
|
if archive:
|
|
path = util.expand_path(archive)
|
|
try:
|
|
self.archive = util.DownloadArchive(path, self.extractor)
|
|
except Exception as exc:
|
|
self.extractor.log.warning(
|
|
"Failed to open download archive at '%s' ('%s: %s')",
|
|
path, exc.__class__.__name__, exc)
|
|
else:
|
|
self.extractor.log.debug("Using download archive '%s'", path)
|
|
|
|
postprocessors = config("postprocessors")
|
|
if postprocessors:
|
|
pp_log = self.get_logger("postprocessor")
|
|
pp_list = []
|
|
category = self.extractor.category
|
|
|
|
for pp_dict in postprocessors:
|
|
whitelist = pp_dict.get("whitelist")
|
|
blacklist = pp_dict.get("blacklist")
|
|
if (whitelist and category not in whitelist or
|
|
blacklist and category in blacklist):
|
|
continue
|
|
name = pp_dict.get("name")
|
|
pp_cls = postprocessor.find(name)
|
|
if not pp_cls:
|
|
pp_log.warning("module '%s' not found", name)
|
|
continue
|
|
try:
|
|
pp_obj = pp_cls(self, pp_dict)
|
|
except Exception as exc:
|
|
pp_log.error("'%s' initialization failed: %s: %s",
|
|
name, exc.__class__.__name__, exc)
|
|
else:
|
|
pp_list.append(pp_obj)
|
|
|
|
if pp_list:
|
|
self.postprocessors = pp_list
|
|
self.extractor.log.debug(
|
|
"Active postprocessor modules: %s", pp_list)
|
|
|
|
|
|
class SimulationJob(DownloadJob):
|
|
"""Simulate the extraction process without downloading anything"""
|
|
|
|
def handle_url(self, url, kwdict, fallback=None):
|
|
self.pathfmt.set_filename(kwdict)
|
|
self.out.skip(self.pathfmt.path)
|
|
if self.sleep:
|
|
time.sleep(self.sleep)
|
|
if self.archive:
|
|
self.archive.add(kwdict)
|
|
|
|
def handle_directory(self, kwdict):
|
|
if not self.pathfmt:
|
|
self.initialize()
|
|
|
|
|
|
class KeywordJob(Job):
|
|
"""Print available keywords"""
|
|
|
|
def handle_url(self, url, kwdict):
|
|
print("\nKeywords for filenames and --filter:")
|
|
print("------------------------------------")
|
|
self.print_kwdict(kwdict)
|
|
raise exception.StopExtraction()
|
|
|
|
def handle_directory(self, kwdict):
|
|
print("Keywords for directory names:")
|
|
print("-----------------------------")
|
|
self.print_kwdict(kwdict)
|
|
|
|
def handle_queue(self, url, kwdict):
|
|
extr = None
|
|
if "_extractor" in kwdict:
|
|
extr = kwdict["_extractor"].from_url(url)
|
|
|
|
if not util.filter_dict(kwdict):
|
|
self.extractor.log.info(
|
|
"This extractor only spawns other extractors "
|
|
"and does not provide any metadata on its own.")
|
|
|
|
if extr:
|
|
self.extractor.log.info(
|
|
"Showing results for '%s' instead:\n", url)
|
|
KeywordJob(extr, self).run()
|
|
else:
|
|
self.extractor.log.info(
|
|
"Try 'gallery-dl -K \"%s\"' instead.", url)
|
|
else:
|
|
print("Keywords for --chapter-filter:")
|
|
print("------------------------------")
|
|
self.print_kwdict(kwdict)
|
|
if extr or self.extractor.categorytransfer:
|
|
print()
|
|
KeywordJob(extr or url, self).run()
|
|
raise exception.StopExtraction()
|
|
|
|
@staticmethod
|
|
def print_kwdict(kwdict, prefix=""):
|
|
"""Print key-value pairs in 'kwdict' with formatting"""
|
|
suffix = "]" if prefix else ""
|
|
for key, value in sorted(kwdict.items()):
|
|
if key[0] == "_":
|
|
continue
|
|
key = prefix + key + suffix
|
|
|
|
if isinstance(value, dict):
|
|
KeywordJob.print_kwdict(value, key + "[")
|
|
|
|
elif isinstance(value, list):
|
|
if value and isinstance(value[0], dict):
|
|
KeywordJob.print_kwdict(value[0], key + "[][")
|
|
else:
|
|
print(key, "[]", sep="")
|
|
for val in value:
|
|
print(" -", val)
|
|
|
|
else:
|
|
# string or number
|
|
print(key, "\n ", value, sep="")
|
|
|
|
|
|
class UrlJob(Job):
|
|
"""Print download urls"""
|
|
maxdepth = 1
|
|
|
|
def __init__(self, url, parent=None, depth=1):
|
|
Job.__init__(self, url, parent)
|
|
self.depth = depth
|
|
if depth >= self.maxdepth:
|
|
self.handle_queue = self.handle_url
|
|
|
|
@staticmethod
|
|
def handle_url(url, _):
|
|
print(url)
|
|
|
|
@staticmethod
|
|
def handle_urllist(urls, _):
|
|
prefix = ""
|
|
for url in urls:
|
|
print(prefix, url, sep="")
|
|
prefix = "| "
|
|
|
|
def handle_queue(self, url, _):
|
|
try:
|
|
UrlJob(url, self, self.depth + 1).run()
|
|
except exception.NoExtractorError:
|
|
self._write_unsupported(url)
|
|
|
|
|
|
class DataJob(Job):
|
|
"""Collect extractor results and dump them"""
|
|
|
|
def __init__(self, url, parent=None, file=sys.stdout, ensure_ascii=True):
|
|
Job.__init__(self, url, parent)
|
|
self.file = file
|
|
self.data = []
|
|
self.ascii = config.get(("output",), "ascii", ensure_ascii)
|
|
|
|
private = config.get(("output",), "private")
|
|
self.filter = (lambda x: x) if private else util.filter_dict
|
|
|
|
def run(self):
|
|
# collect data
|
|
try:
|
|
for msg in self.extractor:
|
|
self.dispatch(msg)
|
|
except exception.StopExtraction:
|
|
pass
|
|
except Exception as exc:
|
|
self.data.append((exc.__class__.__name__, str(exc)))
|
|
except BaseException:
|
|
pass
|
|
|
|
# convert numbers to string
|
|
if config.get(("output",), "num-to-str", False):
|
|
for msg in self.data:
|
|
util.transform_dict(msg[-1], util.number_to_string)
|
|
|
|
# dump to 'file'
|
|
try:
|
|
util.dump_json(self.data, self.file, self.ascii, 2)
|
|
self.file.flush()
|
|
except Exception:
|
|
pass
|
|
|
|
return 0
|
|
|
|
def handle_url(self, url, kwdict):
|
|
self.data.append((Message.Url, url, self.filter(kwdict)))
|
|
|
|
def handle_urllist(self, urls, kwdict):
|
|
self.data.append((Message.Urllist, list(urls), self.filter(kwdict)))
|
|
|
|
def handle_directory(self, kwdict):
|
|
self.data.append((Message.Directory, self.filter(kwdict)))
|
|
|
|
def handle_metadata(self, kwdict):
|
|
self.data.append((Message.Metadata, self.filter(kwdict)))
|
|
|
|
def handle_queue(self, url, kwdict):
|
|
self.data.append((Message.Queue, url, self.filter(kwdict)))
|