XMLStarlet with using a variable in XPath expression

Question

I have this command:

find . -iname '*.xml' | xargs xmlstarlet sel -N z="http://abc.com/article/1.0/" \\
--var zgtag="SuperTag" -t -m "/z:profile-extension" -i "//z:tag='$zgtag'" \\
-n -f -o "," -v "count(//z:zgContextsAnonymTotal[z:tag='$zgtag'])" -o "," \\
-v "//z:zgContextsAnonymTotal/z:tag[.='$zgtag']/@z:timestamp" -o "," \\ 
-v  "count(//z:zgContextsREGTotal[z:tag='$zgtag'])" -o "," \\
-v "//z:zgContextsREGTotal/z:tag[.='$zgtag']/@z:timestamp" -o "," \\
-v "count(//z:zgContextsPremiumTotal[z:tag='$zgtag'])" -o "," \\
-v "//z:zgContextsPremiumTotal/z:tag[.='$zgtag']/@z:timestamp" | grep -v ^$

It does not work because the command version without variable (same value repeated instead of using $zgtag show results. What is wrong?

My investigation so far pointed out:

• It has to be --var zgtag="'SuperTag'" because without single quotes the compiled XSLT by xmlstarlet would produce the following variable declaration: <xsl:variable select="SuperTag" name="zgtag"/> which is obviously wrong, the single quotes are missing!
• In the XSLT compiled by xmlstarlet I can see that the variable is NOT inserted correctely, e.g. I see this:

<xsl:choose>
  <xsl:when test="//z:tag=''">

Then I used this version: --var zgtag="'SuperTag'" -m "/z:profile-extension" -i '//z:tag=$zgtag', now I get a correctly compiled XSLT debugging output of xmlstarlet:

  <xsl:template match="/">
    <xsl:variable select="'SuperTag'" name="zgtag"/>
    <xsl:for-each select="/z:profile-extension">
      <xsl:choose>
        <xsl:when test="//z:tag=$zgtag">

However the output still is empty, the same script without the variable (SuperTag directly inserted into the script multiple times) works just fine. So there must be still a problem with this var-styled command.

So any help welcome how to insert and use the defined variable correctly. The definition is not the problem, it's the usage of the variable in the xmlstarlet command.

Answer 1

Let's take a simplified part of your expression (without the namespace, etc.)

xmlstarlet sel --var zgtag="SuperTag" -t -m "/profile-extension" -i "//tag='$zgtag'"

You've declared an XPath variable zgtag, referenced as $zgtag. However, you've put the expression //tag='$zgtag' inside double-quotes, so the shell can parse it. In doing so it evaluates variables, and it knows that as far as it is concerned, $zgtag is unset. So you get this:

xmlstarlet sel --var zgtag="SuperTag" -t -m "/profile-extension" -i "//tag=''"

The command is then executed - and obviously it doesn't do what you want.

The solution, as ever, is to use single quotes to protect the expression from the shell. (And remove the internal single quotes at the point of use, which are not expected by XPath)

xmlstarlet sel --var zgtag='"SuperTag"' -t -m "/profile-extension" -i '//tag=$zgtag'

In general, use single quotes for strings that do not need to be parsed by the shell, and double quotes otherwise. See What is the difference between the "...", '...', $'...', and $"..." quotes in the shell?

---

Worked example. I cannot test an application to your own command because I have neither valid input nor expected output for validating results. Instead I offer you this:

cat >xml <<'EOF'
<?xml version="1.0"?>
<outer>
  <inner x="a">Element A text</inner>
  <inner x="b">Element B text</inner>
</outer>
EOF
xmlstarlet sel -t --var e='"b"' -v '//inner[@x=$e]' -n <xml
Element B text