You can find at least one example of a graph-processing algorithm - looking for cycles - in my XSLT 2.0 Programmer's Reference. It's entirely doable, but you need to become familiar with functional programming using recursion.

A slight complication here is that you want to find a set of paths. That's most naturally modelled as a sequence of sequences of nodes; but unfortunately the XPath 2.0 data model doesn't allow you to manipulate sequences of sequences. One workaround might be to collapse them into a single sequence, using the document node as a marker to indicate where one sequence ends and the next one starts. Another approach is to model each path as a string consisting of the node id values, comma separated. That approach makes it easy to select the paths that pass through B, H, and K using a regular expression applied to the string representation of the path.

To find all the paths that start at B, you can use a recursive function something like this:

<xsl:function name="f:extend-path-set" as="xs:string*"> <xsl:param name="path-set" as="xs:string*"> <xsl:for-each select="$path-set"> <xsl:variable name="path" select="tokenize(., ',')"/> <xsl:sequence select="."/> <xsl:for-each select="$graph//edge[@source=$path[last()]][not($path = @target)]"> <xsl:sequence select="f:extend-path-set(concat(., ',', @target))"/> </xsl:for-each> </xsl:for-each> </xsl:function>

<xsl:variable name="paths-starting-at-B" select="f:extend-path-set('B')"/>

And then you can find all the paths through B, H, K as

<xsl:variable name="paths-through-B-H-K" select="$paths-starting-at-B[matches(concat(',',.,','), ',B,[.*,]*H,[.*,]*K,')]"/>

Not tested.

Note that the predicate [not($path = @target)] is to stop infinite looping if there is a cycle in the data. $graph is a global variable set to the root node of the input document.