Scrapy: Define items dynamically

Just use a single Field as an arbitrary data placeholder. And then when you want to get the data out, instead of saying for field in item, you say for field in item['row']. You don't need pipelines or loaders to accomplish this task, but they are both used extensively for good reason: they are worth learning.

spider:

from scrapy.item import Item, Field
from scrapy.spider import BaseSpider

class TableItem(Item):
    row = Field()

class TestSider(BaseSpider):
    name = "tabletest"
    start_urls = ('http://scrapy.org?finger', 'http://example.com/toe')

    def parse(self, response):
        item = TableItem()

        row = dict(
            foo='bar',
            baz=[123, 'test'],
        )
        row['url'] = response.url

        if 'finger' in response.url:
            row['digit'] = 'my finger'
            row['appendage'] = 'hand'
        else:
            row['foot'] = 'might be my toe'

        item['row'] = row

        return item

outptut:

stav@maia:/srv/stav/scrapie/oneoff$ scrapy crawl tabletest
2013-03-14 06:55:52-0600 [scrapy] INFO: Scrapy 0.17.0 started (bot: oneoff)
2013-03-14 06:55:52-0600 [scrapy] DEBUG: Overridden settings: {'NEWSPIDER_MODULE': 'oneoff.spiders', 'SPIDER_MODULES': ['oneoff.spiders'], 'USER_AGENT': 'Chromium OneOff 24.0.1312.56 Ubuntu 12.04 (24.0.1312.56-0ubuntu0.12.04.1)', 'BOT_NAME': 'oneoff'}
2013-03-14 06:55:53-0600 [scrapy] DEBUG: Enabled extensions: LogStats, TelnetConsole, CloseSpider, WebService, CoreStats, SpiderState
2013-03-14 06:55:53-0600 [scrapy] DEBUG: Enabled downloader middlewares: HttpAuthMiddleware, DownloadTimeoutMiddleware, UserAgentMiddleware, RetryMiddleware, DefaultHeadersMiddleware, MetaRefreshMiddleware, HttpCompressionMiddleware, RedirectMiddleware, CookiesMiddleware, ChunkedTransferMiddleware, DownloaderStats
2013-03-14 06:55:53-0600 [scrapy] DEBUG: Enabled spider middlewares: HttpErrorMiddleware, OffsiteMiddleware, RefererMiddleware, UrlLengthMiddleware, DepthMiddleware
2013-03-14 06:55:53-0600 [scrapy] DEBUG: Enabled item pipelines:
2013-03-14 06:55:53-0600 [tabletest] INFO: Spider opened
2013-03-14 06:55:53-0600 [tabletest] INFO: Crawled 0 pages (at 0 pages/min), scraped 0 items (at 0 items/min)
2013-03-14 06:55:53-0600 [scrapy] DEBUG: Telnet console listening on 0.0.0.0:6023
2013-03-14 06:55:53-0600 [scrapy] DEBUG: Web service listening on 0.0.0.0:6080
2013-03-14 06:55:53-0600 [tabletest] DEBUG: Crawled (200) <GET http://scrapy.org?finger> (referer: None)
2013-03-14 06:55:53-0600 [tabletest] DEBUG: Scraped from <200 http://scrapy.org?finger>
    {'row': {'appendage': 'hand',
             'baz': [123, 'test'],
             'digit': 'my finger',
             'foo': 'bar',
             'url': 'http://scrapy.org?finger'}}
2013-03-14 06:55:53-0600 [tabletest] DEBUG: Redirecting (302) to <GET http://www.iana.org/domains/example/> from <GET http://example.com/toe>
2013-03-14 06:55:53-0600 [tabletest] DEBUG: Redirecting (302) to <GET http://www.iana.org/domains/example> from <GET http://www.iana.org/domains/example/>
2013-03-14 06:55:53-0600 [tabletest] DEBUG: Crawled (200) <GET http://www.iana.org/domains/example> (referer: None)
2013-03-14 06:55:53-0600 [tabletest] DEBUG: Scraped from <200 http://www.iana.org/domains/example>
    {'row': {'baz': [123, 'test'],
             'foo': 'bar',
             'foot': 'might be my toe',
             'url': 'http://www.iana.org/domains/example'}}
2013-03-14 06:55:53-0600 [tabletest] INFO: Closing spider (finished)
2013-03-14 06:55:53-0600 [tabletest] INFO: Dumping Scrapy stats:
    {'downloader/request_bytes': 1066,
     'downloader/request_count': 4,
     'downloader/request_method_count/GET': 4,
     'downloader/response_bytes': 3833,
     'downloader/response_count': 4,
     'downloader/response_status_count/200': 2,
     'downloader/response_status_count/302': 2,
     'finish_reason': 'finished',
     'finish_time': datetime.datetime(2013, 3, 14, 12, 55, 53, 848735),
     'item_scraped_count': 2,
     'log_count/DEBUG': 13,
     'log_count/INFO': 4,
     'response_received_count': 2,
     'scheduler/dequeued': 4,
     'scheduler/dequeued/memory': 4,
     'scheduler/enqueued': 4,
     'scheduler/enqueued/memory': 4,
     'start_time': datetime.datetime(2013, 3, 14, 12, 55, 53, 99635)}
2013-03-14 06:55:53-0600 [tabletest] INFO: Spider closed (finished)

The custom __setitem__ solution didn't work for me when using item loaders in Scrapy 1.0.3 because the item loader accesses the fields attribute directly:

value = self.item.fields[field_name].get(key, default)

The custom __setitem__ is only called for item-level accesses like item['new field']. Since fields is just a dict, I realized I could simply create an Item subclass that uses a defaultdict to gracefully handle these situations.

In the end, just two extra lines of code:

from collections import defaultdict


class FlexItem(scrapy.Item):
    """An Item that creates fields dynamically"""
    fields = defaultdict(scrapy.Field)

In Scrapy 1.0+ the better way could be to yield Python dicts instead of Item instances if you don't have a well-defined schema. Check e.g. an example on http://scrapy.org/ front page - there is no Item defined.

Use this class:

class Arbitrary(Item):
    def __setitem__(self, key, value):
        self._values[key] = value
        self.fields[key] = {}