""" 此文件作为 `django.db.models` 的 shortcut 这样做的优点与缺点为: 优点: - 包命名都统一为 `models` - 用户在使用的时候只导入本文件即可 缺点: - 此文件中添加代码的时候,注意不要跟 `django.db.models` 中的命名冲突 """ import uuid from django.db import models from django.db import transaction from django.db.models import F, ExpressionWrapper, CASCADE from django.db.models import QuerySet from django.utils.translation import gettext_lazy as _ from ..const.signals import SKIP_SIGNAL class ChoicesMixin: _value2label_map_: dict @classmethod def get_label(cls, value: (str, int)): return cls._value2label_map_[value] class BaseCreateUpdateModel(models.Model): created_by = models.CharField(max_length=128, null=True, blank=True, verbose_name=_('Created by')) updated_by = models.CharField(max_length=128, null=True, blank=True, verbose_name=_('Updated by')) date_created = models.DateTimeField(auto_now_add=True, null=True, blank=True, verbose_name=_('Date created')) date_updated = models.DateTimeField(auto_now=True, verbose_name=_('Date updated')) comment = models.TextField(default='', blank=True, verbose_name=_('Comment')) class Meta: abstract = True class JMSBaseModel(BaseCreateUpdateModel): id = models.UUIDField(default=uuid.uuid4, primary_key=True) class Meta: abstract = True def __str__(self): return str(self.id) def output_as_string(field_name): return ExpressionWrapper(F(field_name), output_field=models.CharField()) class MultiTableChildQueryset(QuerySet): def bulk_create(self, objs, batch_size=None): assert batch_size is None or batch_size > 0 if not objs: return objs self._for_write = True objs = list(objs) parent_model = self.model._meta.pk.related_model parent_objs = [] for obj in objs: parent_values = {} for field in [f for f in parent_model._meta.fields if hasattr(obj, f.name)]: parent_values[field.name] = getattr(obj, field.name) parent_objs.append(parent_model(**parent_values)) setattr(obj, self.model._meta.pk.attname, obj.id) parent_model.objects.bulk_create(parent_objs, batch_size=batch_size) with transaction.atomic(using=self.db, savepoint=False): self._batched_insert(objs, self.model._meta.local_fields, batch_size) return objs def CASCADE_SIGNAL_SKIP(collector, field, sub_objs, using): # 级联删除时,操作日志标记不保存,以免用户混淆 try: for obj in sub_objs: setattr(obj, SKIP_SIGNAL, True) except: pass CASCADE(collector, field, sub_objs, using)