def field_class_to_schematics_field(field: peewee.Field) -> BaseType:
    if isinstance(field, peewee.ForeignKeyField):
        field = field.rel_field

    kwargs = {}

    # 检查是否 require
    if not ((field.default is not None) or field.null or field.sequence or isinstance(field, peewee.AutoField)):
        kwargs['required'] = True

    if field.help_text:
        kwargs['metadata'] = {'description': field.help_text}

    if isinstance(field, peewee.IntegerField):
        return IntType(**kwargs)
    elif isinstance(field, peewee.FloatField):
        return FloatType(**kwargs)
    elif isinstance(field, (PG_JSONField, PG_BinaryJSONField, SQLITE_JSONField)):
        # 注意 SQLITE_JSONField 是一个 _StringField 所以要提前
        return JSONType(**kwargs)
        # HStore 貌似才应该对应 dict，json可以对应任意类型
        # return JSONDictType(StringType, **kwargs)
    elif isinstance(field, peewee.DateTimeField):
        return DateTimeType(**kwargs)
    elif isinstance(field, peewee.DateField):
        return DateType(**kwargs)
    elif isinstance(field, peewee._StringField):
        return StringType(**kwargs)
    elif isinstance(field, peewee.BooleanField):
        return BooleanType(**kwargs)
    elif isinstance(field, peewee.BlobField):
        return BlobType(**kwargs)
    elif isinstance(field, PG_ArrayField):
        field: PG_ArrayField
        return JSONListType(field_class_to_schematics_field(field._ArrayField__field), **kwargs)


# noinspection PyProtectedMember 

def update(self, records: Iterable[DataRecord], values: SQLValuesToWrite, returning=False) -> Union[int, Iterable[DataRecord]]:
        new_vals = {}
        model = self.vcls.model
        db = self.vcls.model._meta.database
        fields = self.vcls._peewee_fields
        cond = self._build_write_condition(records)

        for k, v in values.items():
            if k in fields:
                field = fields[k]
                is_array_field = isinstance(field, ArrayField)

                if is_array_field:
                    if k in values.set_add_fields:
                        # 这里需要加 [v] 的原因是，params需要数组，举例来说为，[v1,v2,v3]
                        # v = SQL('%s || %%s' % field.column_name, [v])
                        v = SQL('(select ARRAY((select unnest(%s)) union (select unnest(%%s))))' % field.column_name, [v])

                    if k in values.set_remove_fields:
                        v = SQL('(select ARRAY((select unnest(%s)) except (select unnest(%%s))))' % field.column_name, [v])

                    # 尚未启用
                    # if k in values.array_append:
                    #     v = SQL('array_append(%s, %%s)' % field.column_name, [v])

                    # if k in values.array_remove:
                    #     v = SQL('array_remove(%s, %%s)' % field.column_name, [v])

                else:
                    if k in values.incr_fields:
                        v = field + v
                    if k in values.decr_fields:
                        v = field - v

                new_vals[k] = v

        with db.atomic(), PeeweeContext(db):
            if isinstance(db, peewee.PostgresqlDatabase):
                q = model.update(**new_vals).where(cond)
                if returning:
                    # cond: peewee.Expression
                    ret = q.returning(*model._meta.fields.values()).execute()
                    to_record = lambda x: PeeweeDataRecord(None, x, view=self.vcls)
                    items = map(to_record, ret)
                    return list(items)
                else:
                    count = q.execute()
                    return count
            else:
                count = model.update(**new_vals).where(cond).execute()
                if not returning: return count

                to_record = lambda x: PeeweeDataRecord(None, x, view=self.vcls)
                return list(map(to_record, model.select().where(cond).execute()))