Closed

[ghost]

Closed

[elacuesta]

The explanation for this behaviour is that ItemLoader.item is an attribute of the Loader class (technically a property as seen in https://github.com/scrapy/scrapy/blob/1.5/scrapy/loader/__init__.py#L46-L51), and it's used internally to store data from the add_* methods. Since you are overriding it with an instance of MyItem, every MyLoader instance holds a reference to the same MyItem object:

In [1]: import scrapy 
   ...: from scrapy.loader import ItemLoader 
   ...:  
   ...: class MyItem(scrapy.Item): 
   ...:     a = scrapy.Field() 
   ...:  
   ...: class MyLoader(ItemLoader): 
   ...:     item = MyItem() 
   ...:  
   ...: l1 = MyLoader() 
   ...: l2 = MyLoader() 
   ...: id(l1.item) == id(l2.item)                                                                                                                                                                                                                                              
Out[1]: True

I believe what you are looking for is ItemLoader.default_item_class:

In [4]: import scrapy 
   ...: from scrapy.loader import ItemLoader 
   ...:  
   ...: class MyItem(scrapy.Item): 
   ...:     a = scrapy.Field() 
   ...:  
   ...: class MyLoader(ItemLoader): 
   ...:     default_item_class = MyItem 
   ...:  
   ...: l1 = MyLoader() 
   ...: l2 = MyLoader() 
   ...: l1.add_value("a", "b") 
   ...: l2.add_value("a", []) 
   ...: print(l1.load_item()) 
   ...: print(l2.load_item())                                                                                                                                                                                                                                                   
{'a': ['b']}
{}

[Gallaecio]

@hibpm Do you think the documentation changes proposed at #4099 would have prevented you from redefining ItemLoader.item in the first place? Do you remember what in the existing documentation made you define item in your item loader subclass?