[xslt] xsl:key, key() [+current()]- LibXSLT disagrees with fellow processors [vol.2]

[Jan Pokorný <jpokorny redhat com>]

Hello,

first, there's a references recap I found for this topic:

- original mail I've borrowed the subject from [1]
- bugzilla.gnome.org bug [2]
- thread on xsl-list ML [3], which may explain why Saxon behaves as it
  does (and hence differs from current behavior of libxslt in this
  aspect)

I also have just observed this possible discrepancy with something
that can be distilled to this test case serving both as a template
and the input to proceed:

<xsl:stylesheet version="1.0"
    xmlns:xsl="http://www.w3.org/1999/XSL/Transform";
    xmlns:int="__internal__">

<xsl:output method="text" indent="no"/>
<xsl:key name="revmap-bar-to-foo"
         match="int:foo"
         use="//int:bar[ name = current()/@name]"/>

<int:foo name="bla"/>
<int:foo name="bla"/>
<int:foo name="bla"/>

<int:bar name="bla"/>

<xsl:template match="*">
        <xsl:value-of select="concat(
            'answer1:',
                count(int:foo),
            '&#xA;',
            'answer2:',
                count(key('revmap-bar-to-foo',
                          //int:bar[ name='bla']))
        )"/>
</xsl:template>
</xsl:stylesheet>


with xsltproc:
> answer1:3
> answer2:0

with saxon:
> answer1:3
> answer2:3


So my initial question is if something prevents current() being evaluated
in the context of key's use parameter as per common sense for XSLT and
also as per another XSLT processor, i.e., to have it represent
the particular node that was selected as per key's match parameter
while building the key table up.

Also, I was unable to come up with a general workaround solution (except
for theoretical key-less procedural bruteforce technique rendering key's
XSLT provision rather limited, which is another reason I think disabling
current() there is not in line with the natural intentions).

When this situation is more clear, I can try putting some effort to
patching such as-of-then bug, but maybe it's just a feature :)

Disclaimer:
I haven't studied XSLT spec enough to find any authoritative
{counter-,}argument.


[1] https://mail.gnome.org/archives/xslt/2008-April/msg00027.html
[2] https://bugzilla.gnome.org/show_bug.cgi?id=676335
[3] http://www.mhonarc.org/archive/html/xsl-list/2012-12/msg00129.html

-- 
Jan