I've raised a bug report against the spec on this one, but I would be
interested to know how various processors handle it. Please try it and
report the results. The bug report is at

http://www.w3.org/Bugs/Public/show_bug.cgi?id=5849

Stylesheet:

<?xml version="1.0"?>
<xsl:stylesheet xmlns:xsl="http://www.w3.org/1999/XSL/Transform"
version="2.0">

I would be interested to know what other XSLT 1.0 or 2.0 processors do
with this one.

<?xml version="1.0"?>
<doc nr="0">
  <a mark="true" nr="1"/>
  <a nr="2"/>
  <a nr="3"/>
  <a nr="4"/>
  <a mark="true" nr="1"/>
  <a nr="2"/>
  <a nr="3"/>
  <a nr="4"/>
</doc>

<?xml version="1.0" encoding="UTF-8"?>
<doc nr="0">
  <a mark="true" nr="0"/>
  <a nr="2"/>
  <a nr="3"/>
  <a nr="4"/>
  <a mark="true" nr="0"/>
  <a nr="6"/>
  <a nr="7"/>
  <a nr="8"/>
</doc>

<?xml version="1.0" encoding="UTF-8"?><doc nr="0">
  <a mark="true" nr="0"/>
  <a nr="2"/>
  <a nr="3"/>
  <a nr="4"/>
  <a mark="true" nr="0"/>
  <a nr="6"/>
  <a nr="7"/>
  <a nr="8"/>
</doc>

I think the intuitively correct result is:

<?xml version="1.0" encoding="UTF-8"?>
<doc nr="">
  <a mark="true" nr="1"/>
  <a nr="2"/>
  <a nr="3"/>
  <a nr="4"/>
  <a mark="true" nr="1"/>
  <a nr="2"/>
  <a nr="3"/>
  <a nr="4"/>
</doc>

Do you agree?

LibXSLT's output is better, I'd say, as it generates a "0" on <doc>,
which I find more intuitive as a number than an empty string.