You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
parallel-library/cps/editbooks.py

890 lines
42 KiB
Python

# -*- coding: utf-8 -*-
# This file is part of the Calibre-Web (https://github.com/janeczku/calibre-web)
# Copyright (C) 2018-2019 OzzieIsaacs, cervinko, jkrehm, bodybybuddha, ok11,
# andy29485, idalin, Kyosfonica, wuqi, Kennyl, lemmsh,
# falgh1, grunjol, csitko, ytils, xybydy, trasba, vrabe,
# ruben-herold, marblepebble, JackED42, SiphonSquirrel,
# apetresc, nanu-c, mutschler
#
# This program 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.
#
# This program 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 this program. If not, see <http://www.gnu.org/licenses/>.
from __future__ import division, print_function, unicode_literals
import os
from datetime import datetime
import json
from shutil import copyfile
from uuid import uuid4
from flask import Blueprint, request, flash, redirect, url_for, abort, Markup, Response
from flask_babel import gettext as _
from flask_login import current_user, login_required
from sqlalchemy.exc import OperationalError
from . import constants, logger, isoLanguages, gdriveutils, uploader, helper
from . import config, get_locale, ub, worker, db
from . import calibre_db
from .web import login_required_if_no_ano, render_title_template, edit_required, upload_required
editbook = Blueprint('editbook', __name__)
log = logger.create()
# Modifies different Database objects, first check if elements have to be added to database, than check
# if elements have to be deleted, because they are no longer used
def modify_database_object(input_elements, db_book_object, db_object, db_session, db_type):
# passing input_elements not as a list may lead to undesired results
if not isinstance(input_elements, list):
raise TypeError(str(input_elements) + " should be passed as a list")
changed = False
input_elements = [x for x in input_elements if x != '']
# we have all input element (authors, series, tags) names now
# 1. search for elements to remove
del_elements = []
for c_elements in db_book_object:
found = False
if db_type == 'languages':
type_elements = c_elements.lang_code
elif db_type == 'custom':
type_elements = c_elements.value
else:
type_elements = c_elements.name
for inp_element in input_elements:
if inp_element.lower() == type_elements.lower():
# if inp_element == type_elements:
found = True
break
# if the element was not found in the new list, add it to remove list
if not found:
del_elements.append(c_elements)
# 2. search for elements that need to be added
add_elements = []
for inp_element in input_elements:
found = False
for c_elements in db_book_object:
if db_type == 'languages':
type_elements = c_elements.lang_code
elif db_type == 'custom':
type_elements = c_elements.value
else:
type_elements = c_elements.name
if inp_element == type_elements:
found = True
break
if not found:
add_elements.append(inp_element)
# if there are elements to remove, we remove them now
if len(del_elements) > 0:
for del_element in del_elements:
db_book_object.remove(del_element)
changed = True
if len(del_element.books) == 0:
db_session.delete(del_element)
# if there are elements to add, we add them now!
if len(add_elements) > 0:
if db_type == 'languages':
db_filter = db_object.lang_code
elif db_type == 'custom':
db_filter = db_object.value
else:
db_filter = db_object.name
for add_element in add_elements:
# check if a element with that name exists
db_element = db_session.query(db_object).filter(db_filter == add_element).first()
# if no element is found add it
# if new_element is None:
if db_type == 'author':
new_element = db_object(add_element, helper.get_sorted_author(add_element.replace('|', ',')), "")
elif db_type == 'series':
new_element = db_object(add_element, add_element)
elif db_type == 'custom':
new_element = db_object(value=add_element)
elif db_type == 'publisher':
new_element = db_object(add_element, None)
else: # db_type should be tag or language
new_element = db_object(add_element)
if db_element is None:
changed = True
db_session.add(new_element)
db_book_object.append(new_element)
else:
if db_type == 'custom':
if db_element.value != add_element:
new_element.value = add_element
elif db_type == 'languages':
if db_element.lang_code != add_element:
db_element.lang_code = add_element
elif db_type == 'series':
if db_element.name != add_element:
db_element.name = add_element
db_element.sort = add_element
elif db_type == 'author':
if db_element.name != add_element:
db_element.name = add_element
db_element.sort = add_element.replace('|', ',')
elif db_type == 'publisher':
if db_element.name != add_element:
db_element.name = add_element
db_element.sort = None
elif db_element.name != add_element:
db_element.name = add_element
# add element to book
changed = True
db_book_object.append(db_element)
return changed
def modify_identifiers(input_identifiers, db_identifiers, db_session):
"""Modify Identifiers to match input information.
input_identifiers is a list of read-to-persist Identifiers objects.
db_identifiers is a list of already persisted list of Identifiers objects."""
changed = False
input_dict = dict([ (identifier.type.lower(), identifier) for identifier in input_identifiers ])
db_dict = dict([ (identifier.type.lower(), identifier) for identifier in db_identifiers ])
# delete db identifiers not present in input or modify them with input val
for identifier_type, identifier in db_dict.items():
if identifier_type not in input_dict.keys():
db_session.delete(identifier)
changed = True
else:
input_identifier = input_dict[identifier_type]
identifier.type = input_identifier.type
identifier.val = input_identifier.val
# add input identifiers not present in db
for identifier_type, identifier in input_dict.items():
if identifier_type not in db_dict.keys():
db_session.add(identifier)
changed = True
return changed
@editbook.route("/delete/<int:book_id>/", defaults={'book_format': ""})
@editbook.route("/delete/<int:book_id>/<string:book_format>/")
@login_required
def delete_book(book_id, book_format):
if current_user.role_delete_books():
book = calibre_db.get_book(book_id)
if book:
try:
result, error = helper.delete_book(book, config.config_calibre_dir, book_format=book_format.upper())
if not result:
flash(error, category="error")
return redirect(url_for('editbook.edit_book', book_id=book_id))
if error:
flash(error, category="warning")
if not book_format:
# delete book from Shelfs, Downloads, Read list
ub.session.query(ub.BookShelf).filter(ub.BookShelf.book_id == book_id).delete()
ub.session.query(ub.ReadBook).filter(ub.ReadBook.book_id == book_id).delete()
ub.delete_download(book_id)
ub.session.commit()
# check if only this book links to:
# author, language, series, tags, custom columns
modify_database_object([u''], book.authors, db.Authors, calibre_db.session, 'author')
modify_database_object([u''], book.tags, db.Tags, calibre_db.session, 'tags')
modify_database_object([u''], book.series, db.Series, calibre_db.session, 'series')
modify_database_object([u''], book.languages, db.Languages, calibre_db.session, 'languages')
modify_database_object([u''], book.publishers, db.Publishers, calibre_db.session, 'publishers')
cc = calibre_db.session.query(db.Custom_Columns).\
filter(db.Custom_Columns.datatype.notin_(db.cc_exceptions)).all()
for c in cc:
cc_string = "custom_column_" + str(c.id)
if not c.is_multiple:
if len(getattr(book, cc_string)) > 0:
if c.datatype == 'bool' or c.datatype == 'integer' or c.datatype == 'float':
del_cc = getattr(book, cc_string)[0]
getattr(book, cc_string).remove(del_cc)
log.debug('remove ' + str(c.id))
calibre_db.session.delete(del_cc)
calibre_db.session.commit()
elif c.datatype == 'rating':
del_cc = getattr(book, cc_string)[0]
getattr(book, cc_string).remove(del_cc)
if len(del_cc.books) == 0:
log.debug('remove ' + str(c.id))
calibre_db.session.delete(del_cc)
calibre_db.session.commit()
else:
del_cc = getattr(book, cc_string)[0]
getattr(book, cc_string).remove(del_cc)
log.debug('remove ' + str(c.id))
calibre_db.session.delete(del_cc)
calibre_db.session.commit()
else:
modify_database_object([u''], getattr(book, cc_string), db.cc_classes[c.id],
calibre_db.session, 'custom')
calibre_db.session.query(db.Books).filter(db.Books.id == book_id).delete()
else:
calibre_db.session.query(db.Data).filter(db.Data.book == book.id).\
filter(db.Data.format == book_format).delete()
calibre_db.session.commit()
except Exception as e:
log.debug(e)
calibre_db.session.rollback()
else:
# book not found
log.error('Book with id "%s" could not be deleted: not found', book_id)
if book_format:
flash(_('Book Format Successfully Deleted'), category="success")
return redirect(url_for('editbook.edit_book', book_id=book_id))
else:
flash(_('Book Successfully Deleted'), category="success")
return redirect(url_for('web.index'))
def render_edit_book(book_id):
calibre_db.update_title_sort(config)
cc = calibre_db.session.query(db.Custom_Columns).filter(db.Custom_Columns.datatype.notin_(db.cc_exceptions)).all()
book = calibre_db.get_filtered_book(book_id)
if not book:
flash(_(u"Error opening eBook. File does not exist or file is not accessible"), category="error")
return redirect(url_for("web.index"))
for lang in book.languages:
lang.language_name = isoLanguages.get_language_name(get_locale(), lang.lang_code)
book = calibre_db.order_authors(book)
author_names = []
for authr in book.authors:
author_names.append(authr.name.replace('|', ','))
# Option for showing convertbook button
valid_source_formats=list()
allowed_conversion_formats = list()
kepub_possible=None
if config.config_converterpath:
for file in book.data:
if file.format.lower() in constants.EXTENSIONS_CONVERT:
valid_source_formats.append(file.format.lower())
if config.config_kepubifypath and 'epub' in [file.format.lower() for file in book.data]:
kepub_possible = True
if not config.config_converterpath:
valid_source_formats.append('epub')
# Determine what formats don't already exist
if config.config_converterpath:
allowed_conversion_formats = constants.EXTENSIONS_CONVERT[:]
for file in book.data:
if file.format.lower() in allowed_conversion_formats:
allowed_conversion_formats.remove(file.format.lower())
if kepub_possible:
allowed_conversion_formats.append('kepub')
return render_title_template('book_edit.html', book=book, authors=author_names, cc=cc,
title=_(u"edit metadata"), page="editbook",
conversion_formats=allowed_conversion_formats,
config=config,
source_formats=valid_source_formats)
def edit_book_ratings(to_save, book):
changed = False
if to_save["rating"].strip():
old_rating = False
if len(book.ratings) > 0:
old_rating = book.ratings[0].rating
ratingx2 = int(float(to_save["rating"]) * 2)
if ratingx2 != old_rating:
changed = True
is_rating = calibre_db.session.query(db.Ratings).filter(db.Ratings.rating == ratingx2).first()
if is_rating:
book.ratings.append(is_rating)
else:
new_rating = db.Ratings(rating=ratingx2)
book.ratings.append(new_rating)
if old_rating:
book.ratings.remove(book.ratings[0])
else:
if len(book.ratings) > 0:
book.ratings.remove(book.ratings[0])
changed = True
return changed
def edit_book_tags(tags, book):
input_tags = tags.split(',')
input_tags = list(map(lambda it: it.strip(), input_tags))
# if input_tags[0] !="": ??
return modify_database_object(input_tags, book.tags, db.Tags, calibre_db.session, 'tags')
def edit_book_series(series, book):
input_series = [series.strip()]
input_series = [x for x in input_series if x != '']
return modify_database_object(input_series, book.series, db.Series, calibre_db.session, 'series')
def edit_book_series_index(series_index, book):
# Add default series_index to book
modif_date = False
series_index = series_index or '1'
#if series_index == '':
# series_index = '1'
if book.series_index != series_index:
book.series_index = series_index
modif_date = True
return modif_date
# Handle book comments/description
def edit_book_comments(comments, book):
modif_date = False
if len(book.comments):
if book.comments[0].text != comments:
book.comments[0].text = comments
modif_date = True
else:
if comments:
book.comments.append(db.Comments(text=comments, book=book.id))
modif_date = True
return modif_date
def edit_book_languages(languages, book, upload=False):
input_languages = languages.split(',')
unknown_languages = []
input_l = isoLanguages.get_language_codes(get_locale(), input_languages, unknown_languages)
for l in unknown_languages:
log.error('%s is not a valid language', l)
flash(_(u"%(langname)s is not a valid language", langname=l), category="warning")
# ToDo: Not working correct
if upload and len(input_l) == 1:
# If the language of the file is excluded from the users view, it's not imported, to allow the user to view
# the book it's language is set to the filter language
if input_l[0] != current_user.filter_language() and current_user.filter_language() != "all":
input_l[0] = calibre_db.session.query(db.Languages). \
filter(db.Languages.lang_code == current_user.filter_language()).first()
return modify_database_object(input_l, book.languages, db.Languages, calibre_db.session, 'languages')
def edit_book_publisher(to_save, book):
changed = False
if to_save["publisher"]:
publisher = to_save["publisher"].rstrip().strip()
if len(book.publishers) == 0 or (len(book.publishers) > 0 and publisher != book.publishers[0].name):
changed |= modify_database_object([publisher], book.publishers, db.Publishers, calibre_db.session, 'publisher')
elif len(book.publishers):
changed |= modify_database_object([], book.publishers, db.Publishers, calibre_db.session, 'publisher')
return changed
def edit_cc_data(book_id, book, to_save):
changed = False
cc = calibre_db.session.query(db.Custom_Columns).filter(db.Custom_Columns.datatype.notin_(db.cc_exceptions)).all()
for c in cc:
cc_string = "custom_column_" + str(c.id)
if not c.is_multiple:
if len(getattr(book, cc_string)) > 0:
cc_db_value = getattr(book, cc_string)[0].value
else:
cc_db_value = None
if to_save[cc_string].strip():
if c.datatype == 'int' or c.datatype == 'bool' or c.datatype == 'float':
if to_save[cc_string] == 'None':
to_save[cc_string] = None
elif c.datatype == 'bool':
to_save[cc_string] = 1 if to_save[cc_string] == 'True' else 0
if to_save[cc_string] != cc_db_value:
if cc_db_value is not None:
if to_save[cc_string] is not None:
setattr(getattr(book, cc_string)[0], 'value', to_save[cc_string])
changed = True
else:
del_cc = getattr(book, cc_string)[0]
getattr(book, cc_string).remove(del_cc)
calibre_db.session.delete(del_cc)
changed = True
else:
cc_class = db.cc_classes[c.id]
new_cc = cc_class(value=to_save[cc_string], book=book_id)
calibre_db.session.add(new_cc)
changed = True
else:
if c.datatype == 'rating':
to_save[cc_string] = str(int(float(to_save[cc_string]) * 2))
if to_save[cc_string].strip() != cc_db_value:
if cc_db_value is not None:
# remove old cc_val
del_cc = getattr(book, cc_string)[0]
getattr(book, cc_string).remove(del_cc)
if len(del_cc.books) == 0:
calibre_db.session.delete(del_cc)
changed = True
cc_class = db.cc_classes[c.id]
new_cc = calibre_db.session.query(cc_class).filter(
cc_class.value == to_save[cc_string].strip()).first()
# if no cc val is found add it
if new_cc is None:
new_cc = cc_class(value=to_save[cc_string].strip())
calibre_db.session.add(new_cc)
changed = True
calibre_db.session.flush()
new_cc = calibre_db.session.query(cc_class).filter(
cc_class.value == to_save[cc_string].strip()).first()
# add cc value to book
getattr(book, cc_string).append(new_cc)
else:
if cc_db_value is not None:
# remove old cc_val
del_cc = getattr(book, cc_string)[0]
getattr(book, cc_string).remove(del_cc)
if not del_cc.books or len(del_cc.books) == 0:
calibre_db.session.delete(del_cc)
changed = True
else:
input_tags = to_save[cc_string].split(',')
input_tags = list(map(lambda it: it.strip(), input_tags))
changed |= modify_database_object(input_tags,
getattr(book, cc_string),
db.cc_classes[c.id],
calibre_db.session,
'custom')
return changed
def upload_single_file(request, book, book_id):
# Check and handle Uploaded file
if 'btn-upload-format' in request.files:
requested_file = request.files['btn-upload-format']
# check for empty request
if requested_file.filename != '':
if '.' in requested_file.filename:
file_ext = requested_file.filename.rsplit('.', 1)[-1].lower()
if file_ext not in constants.EXTENSIONS_UPLOAD:
flash(_("File extension '%(ext)s' is not allowed to be uploaded to this server", ext=file_ext),
category="error")
return redirect(url_for('web.show_book', book_id=book.id))
else:
flash(_('File to be uploaded must have an extension'), category="error")
return redirect(url_for('web.show_book', book_id=book.id))
file_name = book.path.rsplit('/', 1)[-1]
filepath = os.path.normpath(os.path.join(config.config_calibre_dir, book.path))
saved_filename = os.path.join(filepath, file_name + '.' + file_ext)
# check if file path exists, otherwise create it, copy file to calibre path and delete temp file
if not os.path.exists(filepath):
try:
os.makedirs(filepath)
except OSError:
flash(_(u"Failed to create path %(path)s (Permission denied).", path=filepath), category="error")
return redirect(url_for('web.show_book', book_id=book.id))
try:
requested_file.save(saved_filename)
except OSError:
flash(_(u"Failed to store file %(file)s.", file=saved_filename), category="error")
return redirect(url_for('web.show_book', book_id=book.id))
file_size = os.path.getsize(saved_filename)
is_format = calibre_db.get_book_format(book_id, file_ext.upper())
# Format entry already exists, no need to update the database
if is_format:
log.warning('Book format %s already existing', file_ext.upper())
else:
try:
db_format = db.Data(book_id, file_ext.upper(), file_size, file_name)
calibre_db.session.add(db_format)
calibre_db.session.commit()
calibre_db.update_title_sort(config)
except OperationalError as e:
calibre_db.session.rollback()
log.error('Database error: %s', e)
flash(_(u"Database error: %(error)s.", error=e), category="error")
return redirect(url_for('web.show_book', book_id=book.id))
# Queue uploader info
uploadText=_(u"File format %(ext)s added to %(book)s", ext=file_ext.upper(), book=book.title)
worker.add_upload(current_user.nickname,
"<a href=\"" + url_for('web.show_book', book_id=book.id) + "\">" + uploadText + "</a>")
return uploader.process(
saved_filename, *os.path.splitext(requested_file.filename),
rarExecutable=config.config_rarfile_location)
def upload_cover(request, book):
if 'btn-upload-cover' in request.files:
requested_file = request.files['btn-upload-cover']
# check for empty request
if requested_file.filename != '':
ret, message = helper.save_cover(requested_file, book.path)
if ret is True:
return True
else:
flash(message, category="error")
return False
return None
@editbook.route("/admin/book/<int:book_id>", methods=['GET', 'POST'])
@login_required_if_no_ano
@edit_required
def edit_book(book_id):
modif_date = False
# Show form
if request.method != 'POST':
return render_edit_book(book_id)
# create the function for sorting...
calibre_db.update_title_sort(config)
book = calibre_db.get_filtered_book(book_id)
# Book not found
if not book:
flash(_(u"Error opening eBook. File does not exist or file is not accessible"), category="error")
return redirect(url_for("web.index"))
meta = upload_single_file(request, book, book_id)
if upload_cover(request, book) is True:
book.has_cover = 1
modif_date = True
try:
to_save = request.form.to_dict()
merge_metadata(to_save, meta)
# Update book
edited_books_id = None
#handle book title
if book.title != to_save["book_title"].rstrip().strip():
if to_save["book_title"] == '':
to_save["book_title"] = _(u'Unknown')
book.title = to_save["book_title"].rstrip().strip()
edited_books_id = book.id
modif_date = True
# handle author(s)
input_authors = to_save["author_name"].split('&')
input_authors = list(map(lambda it: it.strip().replace(',', '|'), input_authors))
# we have all author names now
if input_authors == ['']:
input_authors = [_(u'Unknown')] # prevent empty Author
modif_date |= modify_database_object(input_authors, book.authors, db.Authors, calibre_db.session, 'author')
# Search for each author if author is in database, if not, authorname and sorted authorname is generated new
# everything then is assembled for sorted author field in database
sort_authors_list = list()
for inp in input_authors:
stored_author = calibre_db.session.query(db.Authors).filter(db.Authors.name == inp).first()
if not stored_author:
stored_author = helper.get_sorted_author(inp)
else:
stored_author = stored_author.sort
sort_authors_list.append(helper.get_sorted_author(stored_author))
sort_authors = ' & '.join(sort_authors_list)
if book.author_sort != sort_authors:
edited_books_id = book.id
book.author_sort = sort_authors
modif_date = True
if config.config_use_google_drive:
gdriveutils.updateGdriveCalibreFromLocal()
error = False
if edited_books_id:
error = helper.update_dir_stucture(edited_books_id, config.config_calibre_dir, input_authors[0])
if not error:
if to_save["cover_url"]:
result, error = helper.save_cover_from_url(to_save["cover_url"], book.path)
if result is True:
book.has_cover = 1
modif_date = True
else:
flash(error, category="error")
# Add default series_index to book
modif_date |= edit_book_series_index(to_save["series_index"], book)
# Handle book comments/description
modif_date |= edit_book_comments(to_save["description"], book)
# Handle identifiers
input_identifiers = identifier_list(to_save, book)
modif_date |= modify_identifiers(input_identifiers, book.identifiers, calibre_db.session)
# Handle book tags
modif_date |= edit_book_tags(to_save['tags'], book)
# Handle book series
modif_date |= edit_book_series(to_save["series"], book)
if to_save["pubdate"]:
try:
book.pubdate = datetime.strptime(to_save["pubdate"], "%Y-%m-%d")
except ValueError:
book.pubdate = db.Books.DEFAULT_PUBDATE
else:
book.pubdate = db.Books.DEFAULT_PUBDATE
# handle book publisher
modif_date |= edit_book_publisher(to_save, book)
# handle book languages
modif_date |= edit_book_languages(to_save['languages'], book)
# handle book ratings
modif_date |= edit_book_ratings(to_save, book)
# handle cc data
modif_date |= edit_cc_data(book_id, book, to_save)
if modif_date:
book.last_modified = datetime.utcnow()
calibre_db.session.commit()
if config.config_use_google_drive:
gdriveutils.updateGdriveCalibreFromLocal()
if "detail_view" in to_save:
return redirect(url_for('web.show_book', book_id=book.id))
else:
flash(_("Metadata successfully updated"), category="success")
return render_edit_book(book_id)
else:
calibre_db.session.rollback()
flash(error, category="error")
return render_edit_book(book_id)
except Exception as e:
log.exception(e)
calibre_db.session.rollback()
flash(_("Error editing book, please check logfile for details"), category="error")
return redirect(url_for('web.show_book', book_id=book.id))
def merge_metadata(to_save, meta):
if to_save['author_name'] == _(u'Unknown'):
to_save['author_name'] = ''
if to_save['book_title'] == _(u'Unknown'):
to_save['book_title'] = ''
for s_field, m_field in [
('tags', 'tags'), ('author_name', 'author'), ('series', 'series'),
('series_index', 'series_id'), ('languages', 'languages'),
('book_title', 'title')]:
to_save[s_field] = to_save[s_field] or getattr(meta, m_field, '')
to_save["description"] = to_save["description"] or Markup(
getattr(meta, 'description', '')).unescape()
def identifier_list(to_save, book):
"""Generate a list of Identifiers from form information"""
id_type_prefix = 'identifier-type-'
id_val_prefix = 'identifier-val-'
result = []
for type_key, type_value in to_save.items():
if not type_key.startswith(id_type_prefix):
continue
val_key = id_val_prefix + type_key[len(id_type_prefix):]
if val_key not in to_save.keys():
continue
result.append( db.Identifiers(to_save[val_key], type_value, book.id) )
return result
@editbook.route("/upload", methods=["GET", "POST"])
@login_required_if_no_ano
@upload_required
def upload():
if not config.config_uploading:
abort(404)
if request.method == 'POST' and 'btn-upload' in request.files:
for requested_file in request.files.getlist("btn-upload"):
try:
modif_date = False
# create the function for sorting...
calibre_db.update_title_sort(config)
calibre_db.session.connection().connection.connection.create_function('uuid4', 0, lambda: str(uuid4()))
# check if file extension is correct
if '.' in requested_file.filename:
file_ext = requested_file.filename.rsplit('.', 1)[-1].lower()
if file_ext not in constants.EXTENSIONS_UPLOAD:
flash(
_("File extension '%(ext)s' is not allowed to be uploaded to this server",
ext=file_ext), category="error")
return Response(json.dumps({"location": url_for("web.index")}), mimetype='application/json')
else:
flash(_('File to be uploaded must have an extension'), category="error")
return Response(json.dumps({"location": url_for("web.index")}), mimetype='application/json')
# extract metadata from file
try:
meta = uploader.upload(requested_file, config.config_rarfile_location)
except (IOError, OSError):
log.error("File %s could not saved to temp dir", requested_file.filename)
flash(_(u"File %(filename)s could not saved to temp dir",
filename= requested_file.filename), category="error")
return Response(json.dumps({"location": url_for("web.index")}), mimetype='application/json')
title = meta.title
authr = meta.author
if title != _(u'Unknown') and authr != _(u'Unknown'):
entry = calibre_db.check_exists_book(authr, title)
if entry:
log.info("Uploaded book probably exists in library")
flash(_(u"Uploaded book probably exists in the library, consider to change before upload new: ")
+ Markup(render_title_template('book_exists_flash.html', entry=entry)), category="warning")
# handle authors
input_authors = authr.split('&')
# handle_authors(input_authors)
input_authors = list(map(lambda it: it.strip().replace(',', '|'), input_authors))
# we have all author names now
if input_authors == ['']:
input_authors = [_(u'Unknown')] # prevent empty Author
sort_authors_list=list()
db_author = None
for inp in input_authors:
stored_author = calibre_db.session.query(db.Authors).filter(db.Authors.name == inp).first()
if not stored_author:
if not db_author:
db_author = db.Authors(inp, helper.get_sorted_author(inp), "")
calibre_db.session.add(db_author)
calibre_db.session.commit()
sort_author = helper.get_sorted_author(inp)
else:
if not db_author:
db_author = stored_author
sort_author = stored_author.sort
sort_authors_list.append(sort_author) # helper.get_sorted_author(sort_author))
sort_authors = ' & '.join(sort_authors_list)
title_dir = helper.get_valid_filename(title)
author_dir = helper.get_valid_filename(db_author.name)
filepath = os.path.join(config.config_calibre_dir, author_dir, title_dir)
saved_filename = os.path.join(filepath, title_dir + meta.extension.lower())
# check if file path exists, otherwise create it, copy file to calibre path and delete temp file
if not os.path.exists(filepath):
try:
os.makedirs(filepath)
except OSError:
log.error("Failed to create path %s (Permission denied)", filepath)
flash(_(u"Failed to create path %(path)s (Permission denied).", path=filepath), category="error")
return Response(json.dumps({"location": url_for("web.index")}), mimetype='application/json')
try:
copyfile(meta.file_path, saved_filename)
os.unlink(meta.file_path)
except OSError as e:
log.error("Failed to move file %s: %s", saved_filename, e)
flash(_(u"Failed to Move File %(file)s: %(error)s", file=saved_filename, error=e), category="error")
return Response(json.dumps({"location": url_for("web.index")}), mimetype='application/json')
if meta.cover is None:
has_cover = 0
copyfile(os.path.join(constants.STATIC_DIR, 'generic_cover.jpg'),
os.path.join(filepath, "cover.jpg"))
else:
has_cover = 1
# combine path and normalize path from windows systems
path = os.path.join(author_dir, title_dir).replace('\\', '/')
# Calibre adds books with utc as timezone
db_book = db.Books(title, "", sort_authors, datetime.utcnow(), datetime(101, 1, 1),
'1', datetime.utcnow(), path, has_cover, db_author, [], "")
modif_date |= modify_database_object(input_authors, db_book.authors, db.Authors, calibre_db.session,
'author')
# Add series_index to book
modif_date |= edit_book_series_index(meta.series_id, db_book)
# add languages
modif_date |= edit_book_languages(meta.languages, db_book, upload=True)
# handle tags
modif_date |= edit_book_tags(meta.tags, db_book)
# handle series
modif_date |= edit_book_series(meta.series, db_book)
# Add file to book
file_size = os.path.getsize(saved_filename)
db_data = db.Data(db_book, meta.extension.upper()[1:], file_size, title_dir)
db_book.data.append(db_data)
calibre_db.session.add(db_book)
# flush content, get db_book.id available
calibre_db.session.flush()
# Comments needs book id therfore only possiblw after flush
modif_date |= edit_book_comments(Markup(meta.description).unescape(), db_book)
book_id = db_book.id
title = db_book.title
error = helper.update_dir_stucture(book_id, config.config_calibre_dir, input_authors[0])
# move cover to final directory, including book id
if has_cover:
try:
new_coverpath = os.path.join(config.config_calibre_dir, db_book.path, "cover.jpg")
copyfile(meta.cover, new_coverpath)
os.unlink(meta.cover)
except OSError as e:
log.error("Failed to move cover file %s: %s", new_coverpath, e)
flash(_(u"Failed to Move Cover File %(file)s: %(error)s", file=new_coverpath,
error=e),
category="error")
# save data to database, reread data
calibre_db.session.commit()
#calibre_db.setup_db(config, ub.app_DB_path)
# Reread book. It's important not to filter the result, as it could have language which hide it from
# current users view (tags are not stored/extracted from metadata and could also be limited)
#book = calibre_db.get_book(book_id)
if config.config_use_google_drive:
gdriveutils.updateGdriveCalibreFromLocal()
if error:
flash(error, category="error")
uploadText=_(u"File %(file)s uploaded", file=title)
worker.add_upload(current_user.nickname,
"<a href=\"" + url_for('web.show_book', book_id=book_id) + "\">" + uploadText + "</a>")
if len(request.files.getlist("btn-upload")) < 2:
if current_user.role_edit() or current_user.role_admin():
resp = {"location": url_for('editbook.edit_book', book_id=book_id)}
return Response(json.dumps(resp), mimetype='application/json')
else:
resp = {"location": url_for('web.show_book', book_id=book_id)}
return Response(json.dumps(resp), mimetype='application/json')
except OperationalError as e:
calibre_db.session.rollback()
log.error("Database error: %s", e)
flash(_(u"Database error: %(error)s.", error=e), category="error")
return Response(json.dumps({"location": url_for("web.index")}), mimetype='application/json')
@editbook.route("/admin/book/convert/<int:book_id>", methods=['POST'])
@login_required_if_no_ano
@edit_required
def convert_bookformat(book_id):
# check to see if we have form fields to work with - if not send user back
book_format_from = request.form.get('book_format_from', None)
book_format_to = request.form.get('book_format_to', None)
if (book_format_from is None) or (book_format_to is None):
flash(_(u"Source or destination format for conversion missing"), category="error")
return redirect(url_for('editbook.edit_book', book_id=book_id))
log.info('converting: book id: %s from: %s to: %s', book_id, book_format_from, book_format_to)
rtn = helper.convert_book_format(book_id, config.config_calibre_dir, book_format_from.upper(),
book_format_to.upper(), current_user.nickname)
if rtn is None:
flash(_(u"Book successfully queued for converting to %(book_format)s",
book_format=book_format_to),
category="success")
else:
flash(_(u"There was an error converting this book: %(res)s", res=rtn), category="error")
return redirect(url_for('editbook.edit_book', book_id=book_id))