Thanks for the report.

HTMLParser.error() was supposed to raise an exception, but the BeautifulSoup project just prints a warning here:

    def error(self, msg):
        warnings.warn(msg)

https://bazaar.launchpad.net/~leonardr/beautifulsoup/bs4/view/head:/bs4/builder/_htmlparser.py#L69

As a result of this, the code doesn't stop executing in the following branch:

else:
    self.error('unknown status keyword %r in marked section' % rawdata[i+3:j])
if not match:
    return -1

https://github.com/python/cpython/blob/3.7/Lib/_markupbase.py#L159

Note that HTMLParser.error() was removed in Python 3.5 (73a4359#diff-1a7486df8279dbac7f20abd487947845L171) and there is an open issue about the status of _markupbase.ParserBase.error(): bpo-31844.

I also think that 73a4359#diff-1a7486df8279dbac7f20abd487947845L171 may have caused a minor regression when it was removed the error() method and its uses from the HTMLParser class. It still calls the parse_marked_section() method of _markupbase.ParserBase() which it then calls the error() method of _markupbase.ParserBase():

elif rawdata[i:i+3] == '<![':
    return self.parse_marked_section(i)

https://github.com/python/cpython/blob/3.7/Lib/html/parser.py#L264

Attaching a test case for this issue since the self.error code path was not covered in the current test suite. A PR is open proposing match variable initialisation with None #17643 .

diff --git Lib/test/test_htmlparser.py Lib/test/test_htmlparser.py
index a2bfb39d16..a9aff11706 100644
--- Lib/test/test_htmlparser.py
+++ Lib/test/test_htmlparser.py
@@ -766,6 +766,18 @@ class AttributesTestCase(TestCaseBase):
                           [("href", "http://www.example.org/\">;")]),
                          ("data", "spam"), ("endtag", "a")])
 
+    def test_invalid_keyword_error_exception(self):
+        class InvalidMarkupException(Exception):
+            pass
+
+        class MyHTMLParser(html.parser.HTMLParser):
+
+            def error(self, message):
+                raise InvalidMarkupException(message)
+
+        parser = MyHTMLParser()
+        with self.assertRaises(InvalidMarkupException):
+            parser.feed('<![invalid>')
 
 if __name__ == "__main__":
     unittest.main()

(Author of PR #17643)

Since the behavior of self.error() is determined by the subclass implementation, an Exception is not guaranteed. How should this be handled? It seems the options are:

• continue execution, in which case 'match' needs to be defined (I proposed initialization to None, which results in returning -1 on the next line)
• return a value
• raise an Exception

Happy to update the PR with @XTreak's test cases.

HTMLParser is supposed to follow the HTML5 standard, and never raise an error.

For the example in the first comment ("<![hi world]>"), the steps should be:

• https://html.spec.whatwg.org/multipage/parsing.html#data-state:tag-open-state
• https://html.spec.whatwg.org/multipage/parsing.html#tag-open-state:markup-declaration-open-state
• https://html.spec.whatwg.org/multipage/parsing.html#markup-declaration-open-state:bogus-comment-state
• https://html.spec.whatwg.org/multipage/parsing.html#bogus-comment-state

I agree that the error should be fixed by setting match to None, and a test case that triggers the UnboundLocalError (before the fix) should be added as well (what provided by Karthikeyan looks good).

However, it also seems wrong that HTMLParser ends up calling self.error() through Lib/_markupbase.py ParserBase after HTMLParser.error() and all the calls to it have been removed. _markupbase.py is internal, so it should be safe to remove ParserBase.error() and the code that calls it as suggested in bpo-31844 (and possibly to merge _markupbase into html.parser too). Even if this is done and the call to self.error() is removed from ParserBase.parse_marked_section(), match still needs to be set to None (either in the else branch or before the if/elif block).

_markupbase.py is internal, so it should be safe to remove
ParserBase.error() and the code that calls it as suggested in bpo-31844

Should I reopen #8562 then?

I think so.
It might be worth double-checking if BeautifulSoup (and possibly other 3rd party libs) use _markupbase.py and/or ParserBase.error(). If they do, giving them a heads up and/or going through a regular deprecation process might be good, otherwise PR 8562 can be reopened and merged after removing the call to self.error() in ParserBase.parse_marked_section().

Here's a simplified real world example that I found if it helps anyone:

import bs4

html = '''<html><head><!fill in here--> <![end--><!-- start: SSI y-xtra-topspace.shtml --><!-- --></head><body></body></html>'''

soup = bs4.BeautifulSoup(html, 'html.parser')

For reference, my Python version is 3.8.2 and my bs4 version is 4.8.1

I updated to bs4 version 4.9.0 and the same issue occurs.

I'm the maintainer of Beautiful Soup. I learned about this issue when one of my users filed a bug for it against Beautiful Soup (https://bugs.launchpad.net/beautifulsoup/+bug/1883264).

BeautifulSoupHTMLParser (my subclass of html.parser.HTMLParser) implements error() only to prevent the NotImplementedError raised by ParserBase.error(). The docstring for my error() implementation says: "this method is called only on very strange markup and our best strategy is to pretend it didn't happen and keep going."

It shouldn't affect Beautiful Soup if, in a future version of Python, HTMLParser.error is never called and ParserBase is removed.

Based on the changelog, it seems that the fix for this issue has only landed in Python 3.10 but it's still affecting versions 3.7, 3.8 and 3.9. There are a few projects such as Apache Beam that have only recently added support for Python 3.9 with no support for 3.10. Beautiful Soup is a popular tool that's affected by this bug and so my question is if backporting this bugfix to at least Python 3.9 which is still an officially supported release is on the table?

Marek, I can merge a fix for 3.9 for this. I don't think we should be removing _markupbase.ParserBase.error() in 3.9 as #52808 is doing. So we'd need a new patch, like #61843. However, that will need a test as well.

Would you be interested in creating the patch?

Alright, the PR is up. This is my first contribution to Python and I had to sign the Contributor Agreement so it may take 1 business day before you're able to accept it. In the meantime, can you please review and let me know if anything needs to change? I wasn't sure where / how to add an entry in the NEWS.d folder. Can you please help me? Thanks!

any question about this issue? if not, i think it is better to close it

This has been fixed in these PRs so it can be closed:

• bpo-31844: Remove _markupbase.ParserBase.error() #8562
• [3.9] bpo-34480: fix bug where match variable is used prior to being defined (GH-17643) #32256