The title says it all; try the minimal example.

I'd suggest to better verify the input and return such strings unchanged.

Leaving the input unchanged does seem to be what browsers do.  (Issue 7626 has some info on browser behaviour with invalid entity refs.)

Rather than pre-validating the input, I think the exception can be caught and the putative entity returned unchanged, similar to how a keyerror is handled for a named entity.

Would you have any interest in proposing a patch and tests?

Agreed. Here's a patch for HTMLParser. That was easy enough.

With regard to tests, there seems to be already one called test_malformatted_charref in test_htmlparser.py. However, the test tests the whole parser and not only HTMLParser.unescape().

At the same time, HTMLParser.unescape() has the following comment:
"# Internal -- helper to remove special character quoting"

It appears the syntax check is done in line 168 already, but since the unescape function is publicly visible, I'd say that it should be capable of handling all kinds of malformed input, despite that comment. Maybe this comment should be removed.

I'm not entirely sure how to write the test properly, since it doesn't fit into the framework provided by test_htmlparser.py; and unfortunately, my time is rather short at the moment.

Ah, as an undocumented internal interface it may in fact not be appropriate to make this change.  Or it may be.  I'll have to look at the code in more detail to figure that out, or perhaps Senthil will.  (It may even be time to document the function, we'll see.)

Thanks for the patch.

Why not simply remove the additional check in line 168 and leave the responsibility to check the validity of its input to the unescape function (be it explicitly or, like now, lazily). That way, the code changes are minimal, the existing test covers the current issue, and the function gets more robust.

By the way, I came across this function via Stackoverflow:
http://stackoverflow.com/questions/2087370

Yes, I too agree that HTMLParser.unescape() should split-out malformed char-ref just as other browsers do.

But, as unescape function has undocumented/unexposed for releases, I am not sure making it exposed is a good idea. HTMLParser is more for event based parsing of tags, and unescape is a just a helper function in that context.

Given that reasoning if you see the malformatted test, you see that event based parsing does return the malformatted data properly For e.g -  ("data", "&#bad;").

Only calling unescape explicitly does not exhibit this behavior.

Martin: I am not sure if changing something in line 168 would solve the issue. In that particular block of code, the else condition is responsible for throwing the malformed charref on an event. If would like to elaborate a bit more on your suggestion, it would be helpful.

However, I do agree that unescape can be changed as per your patch and I have added a simple test to exercise that change. I think, this can go in.

Fixed this in r87542 in (py3k). unescape is undocumented helper method, so no docs are added.

There was already an issue ( Issue6662) on malformed charref handling and it is fixed.

r87544 (release27-maint) and r87545 (release31-maint).