I'm using xsltproc from libxslt version 1.0.4 with libxml version 2.4.3
compiled on my Debian box.

When numbering, in some cases, the count start from zero instead of one.
For instance, the example in section 7.7 of the XSLT Specification from 16
November 1999 gives a wrong result in numbering H4's. Given the template:

  <xsl:template match="h4">
    <xsl:copy>
      <xsl:number level="any" from="h1" count="h2"/><xsl:text>.</xsl:text>
      <xsl:number level="any" from="h2" count="h3"/><xsl:text>.</xsl:text>
      <xsl:number level="any" from="h3" count="h4"/><xsl:text> </xsl:text>
      <xsl:apply-templates/>
    </xsl:copy>
  </xsl:template>
This will supposedly give a 3-part label to a h4 element, but the last
number is off by one (e.g. it will give 1.1.0, 1.1.1, 1.1.2... instead of
1.1.1, 1.1.2, etc.) The problem seems to come from level="any": if I try to
number all h2, h3, ... in the same fashion in my document, the last part of
the label is off by one whereas all the previous parts are correct.

The following template is supposed to give a four-part label, but shows the
problem described here:

   <xsl:template match="h2|h3|h4|h5">
    <xsl:copy>
      <xsl:number level="any" from="h1" count="h2"/>
      <xsl:if test="not(name(.)='h2')">
	<xsl:text>.</xsl:text>
	<xsl:number level="any" from="h2" count="h3"/>
	<xsl:if test="not(name(.)='h3')">
	  <xsl:text>.</xsl:text>
	  <xsl:number level="any" from="h3" count="h4"/>
	  <xsl:if test="not(name(.)='h4')">
	    <xsl:text>.</xsl:text>
	    <xsl:number level="any" from="h4" count="h5"/>
	  </xsl:if>
	</xsl:if>
      </xsl:if>
      <xsl:text> </xsl:text>
      <xsl:apply-templates/>
    </xsl:copy>
  </xsl:template>
The first <h2> is numbered 0 instead of 1, the following <h3> is numbered
1.0 instead of 1.1, etc. Setting format="1" does not help; setting
format="a" gives a blank string instead of "a", "a" instead of "b", etc.
Tests with xalan and saxon gave expected results.

My guess is that level="any" is the culprit; if h2's are numbered with
level="single", the numbers are correct; however, this breaks the numbering
pattern (which is normal).

Numbering with level="multiple" (in other situations) gives correct result,
except that the formatting is different than the one in the spec (e.g. when
given the format "1.1 ", a one-part label will be "1.", where the spec says
it should be "1"; a three-part label will be "1.1 1" where one could expect
"1.1.1", which is how xalan and saxon do it), although this is surely
unrelated and maybe not a bug but rather a different interpretation of the
spec (which is not too clear on this point).