> I still think it would be worthwhile adding close() calls to
> the examples in the documentation (Doc/library/html.parser.rst).

If I add context manager support to HTMLParser I can update the examples to use it, but otherwise I don't think it's worth changing them now.

> BTW I haven’t tested this, and maybe it is not a concern, but even with
> this patch it looks like the parser will buffer unlimited data and
> output nothing until close() if each string it is fed ends with an 
> ampersand (and otherwise contains only plain text, no tags etc).

This is true, but I don't think it's a realistic case.
For this to be a problem you would need:
1) Someone feeding the parser with arbitrary chunks.  Text files are usually fed to the parser whole, or line by line -- arbitrary chunks are uncommon.
2) A file that contains lot of entities.  In most documents charrefs are not very common, and so the chances that a chunk will split one in the middle is low.  Chances that several consecutive charrefs are split in the middle is even lower.
3) A file that is very big.  Even if all the file is buffered until a call to close(), it shouldn't be a concern, since most files have relatively small size.  It is true that this has a quadratic complexity, but I would expect the parsing to complete in a reasonable time for average sizes.