Collecting xref targets - very slow?

["Peter Desjardins" <pdesjardins@supplyscape.com>]

I recently separated the publishing step in which I generate xref
targets from the actual document generation.  It is now apparent that
collecting the xref targets accounts for the majority of the processing
time in my publishing process.  I have two questions about processing a
DocBook document to collect xref targets:

1. Is it unusual that it takes about two minutes to process a 100Kb
DocBook document to collect xref targets?  I am using the following
syntax for the Saxon processor.  My computer has 1.6Ghz and 1GB Ram,
Windows XP, Cygwin, and version 1.71.1 stylesheets.

java com.icl.saxon.StyleSheet \
-o dropping~ \
path/to/document/my-document.xml \
xsl-custom-fo.xsl \
collect.xref.targets="only" \
targets.filename="mydoctargets.db"

2. I found that if I add the -o parameter I avoid some odd output in the
terminal.  DocBook XSL, The Complete Guide does not show that parameter
when processing only to generate the targets.  However, if I do not
allow Saxon to create an empty file, I see an XML declaration in the
terminal output:

     Writing mydoctargets.db for book
     <?xml version="1.0" encoding="utf-8"?>

If I do allow Saxon to write the empty file, I see this:

     Writing mydoctargets.db for book

Is the XML declaration in the terminal output harmless or does it
indicate a possible misconfiguration?

Thanks for your help.

Peter Desjardins