Unexpected behaviour with HTMLParser...

Diez B. Roggisch deets at nospam.web.de
Wed Oct 10 08:09:40 CEST 2007

Just Another Victim of the Ambient Morality schrieb:
> "Diez B. Roggisch" <deets at nospam.web.de> wrote in message 
> news:5n2avjFfh6h8U1 at mid.uni-berlin.de...
>> Just Another Victim of the Ambient Morality schrieb:
>>>     HTMLParser is behaving in, what I find to be, strange ways and I 
>>> would like to better understand what it is doing and why.
>>>     First, it doesn't appear to translate HTML escape characters.  I 
>>> don't know the actual terminology but things like &amp; don't get 
>>> translated into & as one would like.  Furthermore, not only does 
>>> HTMLParser not translate it properly, it seems to omit it altogether! 
>>> This prevents me from even doing the translation myself, so I can't even 
>>> working around the issue.
>>>     Why is it doing this?  Is there some mode I need to set?  Can anyone 
>>> else duplicate this behaviour?  Is it a bug?
>> Without code, that's hard to determine. But you are aware of e.g.
>> handle_entityref(name)
>> handle_charref(ref)
>> ?
>     Actually, I am not aware of these methods but I will certainly look into 
> them!
>     I was hoping that the issue would be known or simple before I commited 
> to posting code, something that is, to my chagrin, not easily done with my 
> news client...
>>>     Secondly, HTMLParser often calls handle_data() consecutively, without 
>>> any calls to handle_starttag() in between.  I did not expect this.  In 
>>> HTML, you either have text or you have tags.  Why split up my text into 
>>> successive handle_data() calls?  This makes no sense to me.  At the very 
>>> least, it does this in response to text with &amp; like escape sequences 
>>> (or whatever they're called), so that it may successively avoid those 
>>> translations.
>> That's the way XML/HTML is defined - there is no guarantee that you get 
>> text as whole. If you must, you can collect the snippets yourself, and on 
>> the next end-tag deliver them as whole.
>     I think there's some miscommunication, here.
>     You can't mean "That's the way XML/HTML is defined" because those format 
> specifications say nothing about how the format must be parsed.  As far as I 
> can tell, you either meant to say that that's the way HTMLParser is 
> specified or you're referring to how text in XML/HTML can be broken up by 
> tags, in which case I've already addressed that in my post.  I expected to 
> see handle_starttag() calls in between calls to handle_data().
>     Unless I'm missing something, it simply makes no sense to break up 
> contiguous text into multiple handle_data() calls...

I meant that's the way XML/HTML-parsing is defined, yes.

>>>     Again, why is it doing this?  Is there some mode I need to set?  Can 
>>> anyone else duplicate this behaviour?  Is it a bug?
>> No. It's the way it is, because it would require buffering with unlimited 
>> capacity to ensure this property.
>     It depends on what you mean by "unlimited capacity."  Is it so bad to 
> buffer with as much memory as you have? ...or, at least, have a setting for 
> such operation?  Moreover, you know that you'll never have to buffer more 
> than there is HTML, so you hardly need "unlimited capacity..."  For 
> instance, I believe Xerces does this translation for you 'cause, really, why 
> wouldn't you want it to?

I've been dealing with XML-files that are several gigbytes of size and 
never fit into physical memory. So buffering would severely impact the 
whole system if it was the default of the parser.

And you are wrong - xerces (the SAX-parser, which is the equivalent to 
HTMLParser) explicitly does not do that. It is not guaranteed that the 
character-data is passed in one chunk.

DOM is an etirely different subject, it _has_ to be fully parsed. But 
then, it's often problematic because of that.

>>>     These are serious problems for me and I would greatly appreciate a 
>>> deeper understanding of these issues.
>> HTH, and read the docs.
>     This does help, thank you.  I have obviously read the docs, since I can 
> use HTMLParser enough to find this behaviour.  I don't find the docs to be 
> very explanatory (perhaps I'm reading the wrong docs) and I think they 
> assume you already know a lot about HTML and parsing, which may be necessary 
> assumptions but are not necessarily true...

Well, you at least overlooked the methods I mentioned.


More information about the Python-list mailing list