Empty comments are not detected

[tbayer]

Example (updated and fixed, see comments):

>>> wikicode = mwparserfromhell.parse('Person<ref>Woman<!--Man</ref>Camera TV\n')

>>> wikicode.strip_code()
'PersonWoman<!--ManCamera TV'

Expected instead: 'PersonWomanCamera TV', consistent with MediaWiki's behavior

>>> wikicode.filter_comments()
[]

Expected instead: ['<!--Man']

(Encountered in this real-life example, in the form <ref> ...<!--->]</ref>.)

[lahwaacz]

@tbayer <!---> is not even a valid HTML, there is one missing dash. The correct empty comment is with 4 dashes: <!---->. I think you should fix the source wikicode.

[tbayer]

@lahwaacz There is ambiguity on whether <!---> is valid HTML. But you're right, MediaWiki at least doesn't recognize it as such, and the reason for the inconsistency in the given example is a different one (I abstracted a bit too much here) - MediaWiki ends the comment at the closing tag, but mwparserfromhell doesn't. I'll update the example.

[tbayer]

I have updated the example. The bug apparently has to do with the different choices about recognizing and ending a run-on comment at a closing ref tag. Thus it might or might not be a special case of the preceding issue 255.

[tbayer]

@lahwaacz, to your note about fixing the source wikicode: This seems a bit beside the point, as the issue is about having mwparserfromhell and MediaWiki yield the same result.

(For background: I encountered this problem while using mwp to search for certain kinds of malformed comments so that they can be fixed, in response to a request by editors on English Wikipedia, who had noticed cases where such typos had caused substantial parts of article text to be hidden unintentionally. If mwp doesn't detect such comments even though MW does, then that kind of search won't give the expected results.)

[lahwaacz]

The problems with overlapping HTML nodes are tracked in the epic issue #40.