Regression in bs4 4.4.0 when manipulating html document

[Jozef Mlich]

Please read more details in red hat buzilla.
https://bugzilla.redhat.com/show_bug.cgi?id=1242599

[Fake Name (lemuix)]

I can confirm something that looks a /lot/ like this same issue:

Traceback (most recent call last):
  File "/media/Storage/Scripts/ReadableWebProxy/WebMirror/processor/HtmlProcessor.py", line 290, in decomposeItems
    have = soup.find_all(True, attrs=key)
  File "/media/Storage/Scripts/ReadableWebProxy/flask/lib/python3.4/site-packages/bs4/element.py", line 1255, in find_all
    return self._find_all(name, attrs, text, limit, generator, **kwargs)
  File "/media/Storage/Scripts/ReadableWebProxy/flask/lib/python3.4/site-packages/bs4/element.py", line 533, in _find_all
    found = strainer.search(i)
  File "/media/Storage/Scripts/ReadableWebProxy/flask/lib/python3.4/site-packages/bs4/element.py", line 1642, in search
    found = self.search_tag(markup)
  File "/media/Storage/Scripts/ReadableWebProxy/flask/lib/python3.4/site-packages/bs4/element.py", line 1613, in search_tag
    attr_value = markup_attr_map.get(attr)
  File "/media/Storage/Scripts/ReadableWebProxy/flask/lib/python3.4/site-packages/bs4/element.py", line 943, in get
    return self.attrs.get(key, default)
AttributeError: 'NoneType' object has no attribute 'get'

In this case, I'm doing some heavy permutation of the HTML tree, decomposing a number of elements. I can confirm the tree is non-None, and the same code works fine when I roll back to 4.3.2, so it looks like this is indeed a regression.

I can't pull out a nice test-case for the moment (too many dependencies), but I can see if I can get something together next weekend.