Burak Arslan schrieb am 16.09.2014 um 10:12:

Void elements are html tags that don't need a closing tag. <br> is a well known example:

...

...

...

from lxml.builder import E html.tostring(E.br())

'<br>'

...

...

...

etree.tostring(E.br())

'<br/>'

...

...

...

<p> is not a void element, so:

...

...

...

html.tostring(E.p())

'<p></p>'

...

...

...

etree.tostring(E.p())

'<p/>'

see the full list:

http://www.w3.org/TR/html5/syntax.html#elements-0

While working on etree.htmlfile, I noticed that the following tags are not treated as void:

embed, keygen, source, track, wbr

see: https://github.com/lxml/lxml/pull/142#issuecomment-55588559

because of this, lxml can produce invalid html.

It's libxml2 serialising them, though. lxml has no word in this.

OTOH, it might be a good idea to make xf.element() aware of self-closing tags in HTML mode and let it a) write them out without closing tag and b) raise an exception when users attempt to write content into them.

You'd then write

with xf.element('meta', ...): pass

which isn't really worse than

xf.write(etree.Element('meta', ...))

That's another inconsistency, because the latter would still serialise some self-closing elements with closing tags. But I still consider it an improvement to provide a way for users to get it right at all.

Stefan

Burak Arslan schrieb am 30.09.2014 um 10:05:

On 09/25/14 19:59, Stefan Behnel wrote:

...

It's libxml2 serialising them, though. lxml has no word in this.

OTOH, it might be a good idea to make xf.element() aware of self-closing tags in HTML mode and let it a) write them out without closing tag and b) raise an exception when users attempt to write content into them.

I disagree. If the user explicitly wants to write something inside, say, an input element, (s)he should be free to do so, just like the user is free to write namespaced elements. This may be for good reason -- e.g. for a fancy javascript framework that expects that kind of input.

Granted.

So, what you're saying is that even

with xf.element('meta', ...): pass

should write an opening and a closing tag?

...

You'd then write

with xf.element('meta', ...): pass

which isn't really worse than

xf.write(etree.Element('meta', ...))

That's another inconsistency, because the latter would still serialise some self-closing elements with closing tags.

Which ones? How come? I'm testing every single one of them here:

https://github.com/lxml/lxml/blob/f8bdb7d48b5cf4403aa6bbbf10f5f8da6a4a510e/s...

Yes, and you put a comment there that said:

""" # FIXME: These don't get serialized as void elements. """

Stefan

Burak Arslan schrieb am 30.09.2014 um 10:32:

On 09/30/14 11:20, Stefan Behnel wrote:

...

So, what you're saying is that even

with xf.element('meta', ...):
    pass

should write an opening and a closing tag?

yes.

I don't really have a preference here, so the current way of working (i.e. the above) is ok with me.

...

...

...

You'd then write

with xf.element('meta', ...): pass

which isn't really worse than

xf.write(etree.Element('meta', ...))

That's another inconsistency, because the latter would still serialise some self-closing elements with closing tags.

Which ones? How come? I'm testing every single one of them here:

https://github.com/lxml/lxml/blob/f8bdb7d48b5cf4403aa6bbbf10f5f8da6a4a510e/s...

Yes, and you put a comment there that said:

""" # FIXME: These don't get serialized as void elements. """

Ah, sure :) So am I supposed to file a bug in https://bugzilla.gnome.org/buglist.cgi?product=libxml2 ?

There are almost 2000 open issues there, that's scary.

Yes. It usually helps to write the patch yourself and send it to the mailing list instead. :)

Should be easy in this case, as the functionality is already there, so it's most likely just about extending a list somewhere in the HTML code.

Stefan