XML Minifier

// Remove whitespace-only text nodes, but only where the element's children
// are all elements. Anything else is mixed content and belongs to the data.
function minifyXml(source) {
  const doc = new DOMParser().parseFromString(source, 'application/xml');
  if (doc.querySelector('parsererror')) {
    throw new Error(doc.querySelector('parsererror').textContent.trim());
  }

  const strip = (el) => {
    const kids = [...el.childNodes];
    const elementOnly =
      kids.some((n) => n.nodeType === 1) &&
      kids.every((n) => n.nodeType !== 3 || !n.nodeValue.trim());
    if (elementOnly) {
      for (const n of kids) if (n.nodeType === 3) el.removeChild(n);
    }
    for (const child of el.children) strip(child);
  };

  strip(doc.documentElement);
  return new XMLSerializer().serializeToString(doc);
}

// XMLSerializer emits <a/> for an empty element, so the output is not
// byte-identical to input that wrote <a></a>. Same document, different bytes.

# lxml's remove_blank_text is the closest equivalent, and it is honest about
# being a guess: with no DTD it keeps blank text whose siblings carry real
# characters, which is what saves mixed content.
from lxml import etree

parser = etree.XMLParser(
    remove_blank_text=True,
    resolve_entities=False,   # no entity expansion
    no_network=True,          # never fetch a DTD or an include
    load_dtd=False,
    huge_tree=False,
)
tree = etree.fromstring(source.encode('utf-8'), parser)
minified = etree.tostring(tree, encoding='unicode')

# Standard library, no dependency, same rule applied by hand:
#
#   from defusedxml.ElementTree import fromstring
#   root = fromstring(source)
#   for el in root.iter():
#       if len(el) and not (el.text or '').strip():
#           el.text = None
#       if not (el.tail or '').strip():
#           el.tail = None

import javax.xml.XMLConstants;
import javax.xml.parsers.DocumentBuilderFactory;
import javax.xml.transform.*;
import javax.xml.transform.dom.DOMSource;
import javax.xml.transform.stream.StreamResult;
import javax.xml.xpath.*;
import java.io.StringWriter;
import org.w3c.dom.*;

DocumentBuilderFactory dbf = DocumentBuilderFactory.newInstance();
dbf.setFeature(XMLConstants.FEATURE_SECURE_PROCESSING, true);
dbf.setFeature("http://apache.org/xml/features/disallow-doctype-decl", true);
dbf.setXIncludeAware(false);
Document doc = dbf.newDocumentBuilder().parse(new java.io.File("in.xml"));

// setIgnoringElementContentWhitespace only works when a DTD or schema tells
// the parser which content is element-only, which is usually not the case.
// So select the blank text nodes explicitly. The second predicate leaves
// mixed content alone: it skips blanks whose siblings hold real characters.
XPath xpath = XPathFactory.newInstance().newXPath();
NodeList blanks = (NodeList) xpath.evaluate(
    "//text()[not(normalize-space())][not(../text()[normalize-space()])]",
    doc, XPathConstants.NODESET);
for (int i = 0; i < blanks.getLength(); i++) {
    Node n = blanks.item(i);
    n.getParentNode().removeChild(n);
}

TransformerFactory tf = TransformerFactory.newInstance();
tf.setFeature(XMLConstants.FEATURE_SECURE_PROCESSING, true);
Transformer t = tf.newTransformer();
t.setOutputProperty(OutputKeys.INDENT, "no");
StringWriter out = new StringWriter();
t.transform(new DOMSource(doc), new StreamResult(out));
System.out.println(out);

using System.Text;
using System.Xml;

// IgnoreWhitespace drops every whitespace-only text node the reader sees.
// Without a schema that includes the space between two inline elements in
// mixed content, so use this on element-only documents and set it false when
// the document carries prose.
var readerSettings = new XmlReaderSettings
{
    IgnoreWhitespace = true,
    DtdProcessing = DtdProcessing.Prohibit,   // no external entities
    XmlResolver = null,
    MaxCharactersFromEntities = 1024 * 1024,
};

var writerSettings = new XmlWriterSettings
{
    Indent = false,
    NewLineHandling = NewLineHandling.None,
};

var output = new StringBuilder();
using (var reader = XmlReader.Create(new StringReader(source), readerSettings))
using (var writer = XmlWriter.Create(output, writerSettings))
{
    writer.WriteNode(reader, defattr: true);
}
Console.WriteLine(output.ToString());

<?php
$doc = new DOMDocument();

// preserveWhiteSpace must be false before the load, not after. libxml2 then
// applies its blank-node heuristic, which keeps blanks inside mixed content.
$doc->preserveWhiteSpace = false;
$doc->formatOutput = false;

libxml_use_internal_errors(true);
if (!$doc->loadXML($source, LIBXML_NONET)) {
    foreach (libxml_get_errors() as $e) {
        fprintf(STDERR, "XML error at line %d, column %d: %s\n",
            $e->line, $e->column, trim($e->message));
    }
    libxml_clear_errors();
    exit(1);
}

// saveXML() still ends the document with a newline; trim it if the byte
// count is what you are optimising.
echo rtrim($doc->saveXML());

# --noblanks is libxml2's minifier. It removes ignorable whitespace only.
xmllint --noblanks --nonet document.xml > minified.xml

# xmllint has no flag for dropping comments, and sed cannot do it correctly
# (a comment may contain a > character). Use an identity transform with no
# template for comment() nodes:
cat > strip-comments.xsl <<'XSL'
<xsl:stylesheet version="1.0" xmlns:xsl="http://www.w3.org/1999/XSL/Transform">
  <xsl:output method="xml" indent="no"/>
  <xsl:template match="@*|node()">
    <xsl:copy><xsl:apply-templates select="@*|node()"/></xsl:copy>
  </xsl:template>
  <xsl:template match="comment()"/>
</xsl:stylesheet>
XSL
xsltproc --nonet strip-comments.xsl document.xml | xmllint --noblanks --nonet -

# Measure the two questions separately:
wc -c document.xml minified.xml
gzip -9 -c document.xml | wc -c
gzip -9 -c minified.xml | wc -c