263475 - CSS added using xsl:processing-instruction fails to apply to XSLT XML results

Status: VERIFIED INVALID

(Core :: XSLT, defect)

Description

[Greg K.]

In the attached example, Firefox 0.10.1/Win fails to apply the CSS stylesheet
added to the XHTML results generated by the XSL transform.

A static XHTML page demonstrates that the style should apply.

Comment 1

[Greg K.]

Attachment: CSS stylesheet for testcase

Comment 2

[Greg K.]

Attachment: XSLT transform for testcase

Comment 3

[Greg K.]

Attachment: XML testcase

Comment 4

[Greg K.]

Attachment: XHTML example demonstrating validity of CSS logic

Comment 5

[Axel Hecht]

http://lxr.mozilla.org/seamonkey/source/extensions/transformiix/source/xslt/txMozillaXMLOutput.cpp#330

just returns for PIs in HTML output. Seems wrong to me, is that from back in the 
days when html and xml documents were much more different than they are now?
Note, the spec says that PIs should be ended with > instead of ?> for html output,
seems fine to have them.
http://www.w3.org/TR/xslt#section-HTML-Output-Method

Greg, XHTML inside html is not really the best thing since sliced bread, but
that's not really INVALIDating this.

Comment 6

[Greg K.]

Attachment: Corrected XSLT transform for testcase

Comment 7

[Greg K.]

Attachment: Corrected XML testcase

Comment 8

[Greg K.]

Well, I can't make the XML testcase work on Bugzilla. Anyone know why?

Comment 9

[Axel Hecht]

text/xsl is not a mime type, after I fixed that, they work.

Comment 10

[Peter Van der Beken [:peterv]]

Processing instructions don't load stylesheets for regular HTML documents, why
should they work when transforming to an HTML document?

Comment 11

[Greg K.]

Should they not work when transforming to XHTML?

Anyway, it doesn't work when transforming to pure XML, either. I used XHTML in
order to include inline CSS style for easier testing.

Comment 12

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

Greg, the last attachment transforms to HTML since you have method="html" on the
xsl:output element. If you were to transform to XHTML or XML you should get
better results (i.e. by dropping the method attribute).

I'm sort of inclined to say that we're doing the right thing by dropping PIs
when transforming into HTML. Mostly because mozilla in general doesn't seem to
support the idea of PIs in HTML documents. See for example

http://lxr.mozilla.org/mozilla/source/content/html/document/src/nsHTMLDocument.cpp#1324

To be honest though, I don't know how correct mozilla is in doing that.

Comment 13

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

Seems like HTML does in fact have processing instructions:

http://www.w3.org/TR/html4/appendix/notes.html#h-B.3.6

It's questionable if transformiix should support them though considering that
the rest of mozilla doesn't. And it's even more questionable if an
'xml-stylesheet' processing instruction does anything in a HTML document.

I would say that the latter question points to us doing the right thing now.

Comment 14

[Greg K.]

But, the thing is, it doesn't work even in a pure XML context. Transform
<foo>foo</foo> to <p>foo</p> (in, yes, XML mode using method="xml"); try to
apply a style to the p using a CSS file referenced by a PI also generated by the
XSLT, and the style application fails.

Comment 15

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

could you supply a testcase?

Comment 16

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

oh, i think the problem is that your stylesheet is broken. You never actually
output a PI. Look at its location in the xsl file, it's outside all templates.

Comment 17

[Greg K.]

Ah, you're right, of course. With that fix, it works. My bad. Thanks.

Comment 18

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

.