xmllint from libxml2 checks that XML is well-formed, validates it against a DTD or XSD schema, pretty-prints it and extracts data with XPath queries. For XML it's what jq is for JSON.
Syntax
xmllint [options] file.xml…
Instead of a file name you can give -, and the XML is read from stdin. Without options xmllint parses the document and prints it back — this way parse errors are visible.
Checking and validation
| Option | What it does |
|---|
--noout | don't print the document — only report errors |
--valid | validate against the DTD declared in the document itself (<!DOCTYPE …>) |
--dtdvalid file.dtd | validate against an external DTD |
--schema file.xsd | validate against an XML Schema (XSD) |
--relaxng file.rng | validate against a RELAX NG schema |
--recover | on parse errors, output whatever could be parsed |
--html | parse as HTML (forgives unclosed tags) |
--nonet | don't fetch DTDs and entities from the network |
--huge | lift internal limits on document size |
Output and transformation
| Option | What it does |
|---|
--format | output with indentation; the indent width is set by the XMLLINT_INDENT variable (two spaces by default) |
--noblanks | remove insignificant spaces and newlines — XML on one line |
--output file (-o) | write the result to a file instead of stdout |
--encode ENCODING | output in another encoding |
--c14n | canonical form: attributes, quotes and empty tags in a uniform style; together with --noblanks it's handy for comparing documents with diff |
--dropdtd | remove <!DOCTYPE …> |
--noent | substitute entity values |
--xinclude | process <xi:include> |
--nsclean | remove redundant namespace declarations |
Queries
| Option | What it does |
|---|
--xpath 'expression' | evaluate XPath and print the result (the document itself isn't printed) |
--shell | interactive shell: cat //title, ls, cd, exit |
XPath: paths
| Expression | What it selects |
|---|
/catalog/book | book elements directly inside the root catalog |
//title | all titles at any depth |
*, @* | any element / any attribute |
//book/@id | id attributes |
//title/text() | the text inside elements, without tags |
.. | parent: //title[../@id="3"] |
//book[@lang="de"] | filter by attribute |
//book[price > 8] | filter by a child element's value |
//book[1], //book[last()], [position() < 3] | by position within its parent (from 1), the last, the first two |
(//title)[2] | the second title in the whole document |
a | b | union of two node sets |
parent::, ancestor::, following-sibling::, preceding-sibling:: | axes: parent, ancestors, siblings after / before |
XPath: functions and operators
| Function | What it does |
|---|
string(…) | the text value of the first node found |
count(…), sum(…) | number of nodes, sum of values |
contains(a, b), starts-with(a, b) | substring, prefix |
normalize-space(…) | trim spaces at the edges and collapse inner ones |
concat(a, b, …), string-length(…), translate(s, abc, ABC) | concatenation, length, character-by-character replacement |
name(…), local-name(…) | element name with / without the namespace prefix |
not(…), boolean(…) | negation, "was anything found" |
and, or, =, !=, <, > | logic and comparisons in […] conditions |
Exit codes
0 — all fine, 1 — parse error (the XML is broken) or file not found, 3 / 4 — the document failed schema / DTD validation, 10 — XPath found nothing or the expression has an error.
Examples
cat > books.xml <<'EOF'
<catalog><book id="1" lang="en"><title>Dune</title><price>9.99</price></book><book id="2" lang="de"><title>Solaris</title><price>7.50</price></book></catalog>
EOF
# check the syntax: silent if all is well
xmllint --noout books.xml && echo OK
# pretty-print
xmllint --format books.xml
# whole elements, with tags
xmllint --xpath '//title' books.xml
# text only — one line per node
xmllint --xpath '//title/text()' books.xml
# a single value: the title of the book with id=2
xmllint --xpath 'string(//book[@id="2"]/title)' books.xml
# an attribute value without the id="…" wrapper
xmllint --xpath 'string(//book[1]/@lang)' books.xml
# how many books and the sum of the prices
xmllint --xpath 'count(//book)' books.xml
xmllint --xpath 'sum(//price)' books.xml
# search by part of the text
xmllint --xpath '//book[contains(title, "un")]/@id' books.xml
# validate the document against an XSD schema
xmllint --noout --schema catalog.xsd books.xml
# XML from stdin
echo '<a><b>1</b></a>' | xmllint --format -
# compare two documents ignoring formatting
diff <(xmllint --noblanks --c14n a.xml) <(xmllint --noblanks --c14n b.xml)
Pitfalls
string() takes only the first node. string(//title) returns Dune, not all the titles. For all values use //title/text(), for the N-th — string((//title)[N]).
//book[2] isn't "the second book in the document", but every book that is the second child of its parent. For document-wide numbering you need parentheses: (//book)[2].
//book/@id prints id="1" together with the attribute name. To get just the value, use string(…/@id).
- An empty result is an error. If XPath found nothing,
xmllint prints XPath set is empty and exits with code 10. But string() of an empty set is just an empty string and code 0.
- Namespaces. If the root has
xmlns="…", the query //v finds nothing, and //p:v gives Undefined namespace prefix. The workaround: //*[local-name()="v"].
- Whitespace in text is preserved.
<title> Dune </title> returns the value with spaces — an = comparison won't match. Use normalize-space().
- Quotes. Put the expression in single quotes, and strings inside the XPath in double quotes:
'//book[@lang="de"]'.
- In old libxml2 versions,
text() results are glued together without a newline. If the values run together, iterate over the nodes by index.
See also
jq · yq · grep · sed · iconv