The Doxygen XML output is formatted so that indent="no" but, in combine.xslt, indent="yes".

If you were to use a tool that can interpret indent="yes" such as Saxon9 (Note: xsltproc automatically assumes indent="no"), the combined XML output is altered from the original Doxygen-generated XML output.

For example, I have the following sample code preformatted in the original Doxygen-generated XML output:

<preformatted>
 // Descriptive text.</preformatted></para><para><preformatted> <ref refid="interfacesample_1_1apiversion12_1_1NewConfig" kindref="compound">NewConfig</ref> nc;</preformatted></para><para><preformatted> nc = conn.prepareConfig();
 nc.run();</preformatted></para><para><preformatted> nc = conn.prepareConfig(sample.txt);
 nc.run();
 </preformatted>

If indent="no", the Doxygen-formatted text is preserved as-is.  If indent="yes" (and you don't use xsltproc), the text is altered to the following format in the combined XML file:

        <preformatted>
 // Descriptive text.</preformatted>
    </para>
    <para>
        <preformatted> 
            <ref refid="interfacesample_1_1apiversion12_1_1NewConfig" 
                kindref="compound">NewConfig</ref> nc;</preformatted>
    </para>
    <para>
        <preformatted> nc = conn.prepareConfig();
nc.run();</preformatted>
    </para>
    <para>
        <preformatted> nc = conn.prepareConfig(sample.txt);
nc.run();</preformatted>

I recently switched from xsltproc to Saxon9.  After running xsl transformations on the combined output, extra white-space and hard breaks were found in the final doc output.  I'm currently overwriting the Doxygen-created combine.xslt prior to running Saxon9 on it; I now get the same results as I did with xsltproc.