+import io
import json
+from os import unlink
import os.path
+from urllib.parse import urljoin
from django.db import models
from time import sleep
+from django.contrib.sites.models import Site
from django.utils.functional import cached_property
+from django.utils.timezone import now
from django.utils.translation import gettext_lazy as _
from django_pglocks import advisory_lock
import requests
from archive.constants import status
from archive.settings import FILES_SAVE_PATH, ADVERT, ORGANIZATION, PROJECT
from archive.utils import OverwriteStorage, sha1_file
+from youtube.utils import concat_audio, standardize_audio
class License(models.Model):
name = models.CharField(max_length=128, unique=True, db_index=True, verbose_name="Nazwa")
sponsors = models.TextField(blank=True, null=True, verbose_name="Sponsorzy")
description = models.TextField(blank=True, verbose_name="Opis")
+ private_notes = models.TextField(blank=True, verbose_name="Prywatne notatki")
+ config = models.ForeignKey('Config', models.PROTECT)
+ can_sell = models.BooleanField(default=True, verbose_name="Do sprzedaży")
+ required_license = models.ForeignKey('License', models.PROTECT, blank=True, null=True, verbose_name='Wymagana licencja')
+ youtube = models.ForeignKey('youtube.YouTube', models.PROTECT)
+ icon = models.FileField(upload_to='archive/project', blank=True, null=True)
+ info_flac = models.FileField(upload_to='archive/info_flac', blank=True)
class Meta:
verbose_name = _("project")
" finansowanego przez %s" % self.sponsors if self.sponsors else "",
)
+ def get_icon_url(self):
+ if not self.icon:
+ return ''
+ return urljoin(
+ 'https://' + Site.objects.get_current().domain,
+ self.icon.url
+ )
+
-class Piece(models.Model):
+class Config(models.Model):
name = models.CharField(max_length=255)
- source_file = models.FileField(upload_to='piece')
- is_outro = models.BooleanField(default=False)
- min_audiobook_duration = models.IntegerField(default=0)
+ intro_flac = models.FileField(upload_to='config/intro_flac', blank=True)
+ intro_min_seconds = models.IntegerField()
+ outro_flac = models.FileField(upload_to='config/outro_flac', blank=True)
+ outro_min_seconds = models.IntegerField()
+
+ class Meta:
+ verbose_name = _("Configuration")
+ verbose_name_plural = _("Configurations")
def __str__(self):
return self.name
+ def prepare_audio(self, audiobook):
+ total_duration = audiobook.total_duration
+ files = []
+ if self.intro_flac and total_duration > self.intro_min_seconds and audiobook.is_first:
+ files.append(standardize_audio(self.intro_flac.path))
+ files.append(standardize_audio(audiobook.source_file.path))
+ if self.outro_flac and total_duration > self.outro_min_seconds and audiobook.is_last:
+ files.append(standardize_audio(self.outro_flac.path))
+ output = concat_audio(files)
+ for d in files:
+ unlink(d)
+ return output
+
def source_upload_to(intance, filename):
return os.path.join(FILES_SAVE_PATH, filename) # FIXME: what about really long file names?
source_file = models.FileField(upload_to=source_upload_to, max_length=255,
verbose_name=_('source file'), editable=False)
source_sha1 = models.CharField(max_length=40, editable=False)
+ duration = models.FloatField(null=True, editable=False)
title = models.CharField(max_length=255, verbose_name=_('title'))
part_name = models.CharField(max_length=255, verbose_name=_('part name'), help_text=_('eg. chapter in a novel'),
encoded_by = models.CharField(max_length=255, verbose_name=_('encoded by'))
date = models.CharField(max_length=255, verbose_name=_('date'))
project = models.ForeignKey(Project, models.PROTECT, verbose_name=_('project'))
- slug = models.SlugField(max_length=120, help_text=_('WL catalogue slug of the book.'))
+ slug = models.SlugField(max_length=120, blank=True, help_text=_('WL catalogue slug of the book.'))
translator = models.CharField(max_length=255, null=True, blank=True, verbose_name=_('translator'))
modified = models.DateTimeField(null=True, editable=False)
license = models.ForeignKey(License, models.PROTECT, null=True, blank=True, verbose_name=_('license'))
+ license_secondary = models.ForeignKey(License, models.PROTECT, null=True, blank=True, verbose_name=_('secondary license'), related_name='secondary')
# publishing process
mp3_status = models.SmallIntegerField(null=True, editable=False, choices=status.choices)
- mp3_task = models.CharField(max_length=64, null=True, editable=False)
- mp3_tags = models.TextField(null=True, editable=False)
mp3_file = models.FileField(null=True, upload_to='archive/final', storage=OverwriteStorage(), editable=False)
- mp3_published_tags = models.TextField(null=True, editable=False)
mp3_published = models.DateTimeField(null=True, editable=False)
ogg_status = models.SmallIntegerField(null=True, editable=False, choices=status.choices)
- ogg_task = models.CharField(max_length=64, null=True, editable=False)
- ogg_tags = models.TextField(null=True, editable=False)
ogg_file = models.FileField(null=True, upload_to='archive/final', storage=OverwriteStorage(), editable=False)
- ogg_published_tags = models.TextField(null=True, editable=False)
ogg_published = models.DateTimeField(null=True, editable=False)
youtube_status = models.SmallIntegerField(null=True, editable=False, choices=status.choices)
- youtube_task = models.CharField(max_length=64, null=True, editable=False)
- youtube_tags = models.TextField(null=True, editable=False)
- youtube_published_tags = models.TextField(null=True, editable=False)
youtube_published = models.DateTimeField(null=True, editable=False)
youtube_id = models.CharField(max_length=255, blank=True, default='')
youtube_queued = models.DateTimeField(null=True, blank=True)
def parts_count(self):
return type(self).objects.filter(slug=self.slug).count()
+ @property
+ def total_duration(self):
+ return type(self).objects.filter(slug=self.slug).aggregate(s=models.Sum('duration'))['s']
+
+ @property
+ def is_first(self):
+ return not type(self).objects.filter(slug=self.slug, index__lte=self.index).exclude(pk=self.pk).exists()
+
+ @property
+ def is_last(self):
+ return not type(self).objects.filter(slug=self.slug, index__gte=self.index).exclude(pk=self.pk).exists()
+
@property
def youtube_volume_count(self):
total = 0
return (
not self.youtube_volume
or not type(self)
- .objects.filter(youtube_volume=self.youtube_volume, index__lt=self.index)
+ .objects.filter(slug=self.slug, youtube_volume=self.youtube_volume, index__lt=self.index)
.exists()
)
- def get_mp3_tags(self): return json.loads(self.mp3_tags) if self.mp3_tags else None
- def get_ogg_tags(self): return json.loads(self.ogg_tags) if self.ogg_tags else None
- def get_mp3_published_tags(self): return json.loads(self.mp3_published_tags) if self.mp3_published_tags else None
- def get_ogg_published_tags_tags(self): return json.loads(self.ogg_published_tags) if self.ogg_published_tags else None
- def set_mp3_tags(self, tags): self.mp3_tags = json.dumps(tags)
- def set_ogg_tags(self, tags): self.ogg_tags = json.dumps(tags)
+ def youtube_publish(self):
+ if not self.is_youtube_publishable:
+ return False
+ self.youtube_status = status.QUEUED
+ self.youtube_queued = now()
+ self.save(update_fields=['youtube_status', 'youtube_queued'])
def published(self):
return self.mp3_published and self.ogg_published
+ def publish(self, user, publish=True):
+ from . import tasks
+ self.mp3_status = self.ogg_status = status.WAITING
+ self.save(update_fields=['mp3_status', 'ogg_status'])
+ tasks.Mp3Task.delay(user.id, self.pk, publish=publish).task_id
+ tasks.OggTask.delay(user.id, self.pk, publish=publish).task_id
+
def get_source_sha1(self):
assert self.pk or self.source_sha1
if not self.source_sha1:
if self.project.sponsors:
tags['funded_by'] = self.project.sponsors
- if self.source_sha1:
- tags['flac_sha1'] = self.source_sha1
+ tags['flac_sha1'] = self.get_source_sha1()
+
return tags
+ def prepare_audio(self):
+ return self.project.config.prepare_audio(self)
+
@cached_property
def book(self):
- apidata = requests.get(f'https://wolnelektury.pl/api/books/{self.slug}/').json()
+ if self.slug:
+ apidata = requests.get(f'https://wolnelektury.pl/api/books/{self.slug}/').json()
+ else:
+ return {}
return apidata
+
+ @property
+ def document(self):
+ from librarian.document import WLDocument, parser
+ from lxml import etree
+
+ xml_url = self.book.get('xml', None)
+ if xml_url is None:
+ return None
+
+ return WLDocument(url=xml_url)
+
+ @property
+ def cover(self):
+ from librarian.cover import LogoWLCover
+ return LogoWLCover(self.document.meta).output_file.get_bytes()