Saxon

It works for me both from the IDE (IntelliJ) and from the command line, both with 9.5.0.2 and 9.5.1.1, using exactly the command line you showed. What JVM are you using?

It fails if I change the XSLT code from match="str/@type" to match="str". The template rule as written wasn't matching anything in your input.

I can confirm the excessively-deep recursion is within the regex engine. As a first step, I'll try and improve the diagnostics.

I think my jvm is 1.7 b3

The basic problem here is that with the regular expression

\|((\|[^|]+\|)+)\|

on encountering a vertical bar immediately after another vertical bar, there are two paths that the regex can take (it can loop round the outer loop, or it can take the exit path). It therefore has to be prepared to backtrack, and in order to backtrack, it maintains its state on the stack. So each character that is consumed ends up using another stack frame.

I may not have your logic exactly right, but I believe you can achieve the required effect something like this:

<xsl:template match="str">
  <xsl:copy>
    <xsl:variable name="tokens" select="tokenize(., '\|\|')"/>
    <xsl:value-of select="$tokens[1]"/>
    <xsl:for-each select="subsequence($tokens,2)">     
        <xsl:analyze-string select="." regex="(\w+):(.+)">
          <xsl:matching-substring>
            <xsl:element name="{regex-group(1)}">
              <xsl:value-of select="regex-group(2)"/>
            </xsl:element>
          </xsl:matching-substring>
          <xsl:non-matching-substring>
             <xsl:value-of select="."/>
          </xsl:non-matching-substring>
        </xsl:analyze-string>
    </xsl:for-each>
  </xsl:copy>
</xsl:template>

There are of course optimizations possible in the regex engine that could improve the situation here (the main one being to move to a non-backtracking algorithm); but most regex engines will have problems with this case.

That mostly works. Unfortunately, it doesn't handle the background info correctly.

It only takes the first line of the background.

@<p dir="ltr"><span

... rest of the background data

@