Re: [xml] Relaxed entities encoding for html output

[Daniel Veillard <veillard redhat com>]

On Tue, Sep 04, 2012 at 06:36:12PM +0200, rbondue ext orange com wrote:
> Hello,
> I am working on a project where we are using either libxslt or xalan for xslt transformations.
> We have internally deprecated xalan because libxslt is considerably faster, and all other xml processing is 
> performed by libxml2.
> We now would like to drop xalan completely, but there is one important case where both libraries are 
> producing a different output, which prevents us from doing so.
>
> Consider whatever xml file and the following style sheet :
>
>
> <xsl:stylesheet xmlns:xsl="http://www.w3.org/1999/XSL/Transform"; version='1.0'>
> <xsl:output method="html"/>
> <xsl:variable name="apache">&lt;!--apache-stuff--></xsl:variable>
> <xsl:variable name="script">&amp;{My script};</xsl:variable>
>
> <xsl:template match="/">
>     <a href="{$apache}/page.html" onMouseUp="{$script}">link</a>
> </xsl:template>
>
> </xsl:stylesheet>
>
>
>
> libxml2/libxslt currently produce the following file from the transformation:
>
> <a href="&lt;!--apache-stuff--&gt;/page.html" onMouseUp="&amp;{My script};">link</a>
>
>
> And Xerces/Xalan are producing:
>
> <a href="<!--apache-stuff-->/page.html" onMouseUp="&{My script};">link</a>
>
>
> The <!--apache-stuff--> part is supposed to be replaced by the web server for load balancing purpose, but 
> this is not happening when using libxslt because of the escaping (&lt; &gt;),
> And that is the issue we're running into.
>
> I have tracked it down, and the problem lies within libxml2, not libxslt (hence why I am posting on this 
> list!), when the node tree is serialized to text. The enclosed patches are fixing this, and are also 
> implementing a TODO that you had in the code:
>
> The html output method should not escape a & character occurring in an attribute value immediately followed 
> by a { character (see Section B.7.1 of the HTML 4.0 Recommendation).
>
> This is illustrated by the &{My script} part in the example above.
>
> To get back to my issue however, I am not completely sure which behavior is actually correct, as I could 
> not find if '<' and '>' are allowed in attribute values in html (I know '<' is forbidden in xml).
> I run the regression tests, but they added to my confusion:
> Some html tests are now failing in the test suite (runtest), but if I run:
> ./testHTML test/HTML/lt.html
> Then the  output is a lot closer to the input file test/HTML/lt.html, which was not the case before, so 
> this may mean an improvement.
> If this is indeed correct, I'm of course open to any suggestion or comment you may have about the patches, 
> they should apply cleanly to the git trunk.

  Your approach is way too heavy, instead of changing < and & in all
case detecting the full construct first and then special processing
those case is really less disruptive. With that approach no other
test case in libxml2 or libxslt fails. So I commited that restricted
approach but which should handle the cases you raise.

  http://git.gnome.org/browse/libxml2/commit/?id=7d4c529a334845621e2f805c8ed0e154b3350cec

thinkpad:~/XSLT -> xsltproc/xsltproc orange.xsl orange.xsl
<a href="<!--apache-stuff-->/page.html" onMouseUp="&{My script};">link</a>
thinkpad:~/XSLT -> 

Daniel

-- 
Daniel Veillard      | libxml Gnome XML XSLT toolkit  http://xmlsoft.org/
daniel veillard com  | Rpmfind RPM search engine http://rpmfind.net/
http://veillard.com/ | virtualization library  http://libvirt.org/