diff --git a/configure.php b/configure.php
index 1a07591639..85c09b77cd 100755
--- a/configure.php
+++ b/configure.php
@@ -652,11 +652,43 @@ function dtd_text_entities()
 }
 checkvalue($ac["GENERATE"]);
 
-function dom_load( DOMDocument $dom , string $filename , string $baseURI = "" ) : bool
+echo "Creating monolithic temp/manual.xml... ";
+$dom = new DOMDocument();
+
+if ( dom_load( $dom , "{$ac['srcdir']}/{$ac["INPUT_FILENAME"]}" ) )
+{
+    dom_saveload( $dom ); // correct file/line/column on error messages
+    echo " done.\n";
+}
+else
+{
+    echo "failed.\n";
+    print_xml_errors();
+    xml_broken_file_check();
+    errors_are_bad(1);
+}
+
+function dom_load( DOMDocument $dom , string $filename , bool $firstLoad = true ) : bool
 {
     $filename = realpath( $filename );
-    $options = LIBXML_NOENT | LIBXML_COMPACT | LIBXML_BIGLINES | LIBXML_PARSEHUGE;
-    return $dom->load( $filename , $options );
+
+    // On the first load we cannot use LIBXML_NSCLEAN, because
+    // libxml drops all namespaces inside DTD entities.
+
+    $options = LIBXML_NOENT
+             | LIBXML_COMPACT
+             | LIBXML_BIGLINES
+             | LIBXML_PARSEHUGE;
+
+    if ( ! $firstLoad )
+        $options |= LIBXML_NSCLEAN;
+
+    $ret = $dom->load( $filename , $options );
+
+    if ( $ret && $firstLoad )
+        xml_trim_first( $dom );
+
+    return $ret;
 }
 
 function dom_saveload( DOMDocument $dom , string $filename = "" ) : string
@@ -666,28 +698,88 @@ function dom_saveload( DOMDocument $dom , string $filename = "" ) : string
 
     libxml_clear_errors();
     $dom->save( $filename );
-    dom_load( $dom , $filename );
+    dom_load( $dom , $filename , false );
 
     return $filename;
 }
 
-echo "Creating monolithic temp/manual.xml... ";
-$dom = new DOMDocument();
-
-if ( dom_load( $dom , "{$ac['srcdir']}/{$ac["INPUT_FILENAME"]}" ) )
+function xml_trim_first( DOMDocument $doc )
 {
-    dom_saveload( $dom ); // correct file/line/column on error messages
-    echo " done.\n";
+    $xpath = new DOMXPath( $doc );
+    $dtdNode = null;
+    $dels = [];
+
+    // Remove DTD Document Type, as all entity references
+    // are already expanded at this point.
+
+    foreach( $doc->childNodes as $node )
+        if ( $node->nodeType == XML_DOCUMENT_TYPE_NODE )
+            $dtdNode = $node;
+    if ( $dtdNode != null )
+        $node->parentNode->removeChild( $dtdNode );
+
+    // Remove all XML comments, as they are distracting,
+    // in reverse order, outside enumerations.
+
+    $comments = $xpath->query( "//comment()" );
+    for ( $idx = $comments->length - 1 ; $idx >= 0 ; $idx-- )
+    {
+        $node = $comments[ $idx ];
+        $node->parentNode->removeChild( $node );
+    }
+
+    // See docbook/docbookwsi.php.
+
+    // These first five elements account for almost 50%
+    // of all insignificant Docbook whitespace in manual.
+
+    $trimList[] = 'refsect1';
+    $trimList[] = 'varlistentry';
+    $trimList[] = 'listitem';
+    $trimList[] = 'row';
+    $trimList[] = 'methodsynopsis';
+
+    // The first twenty elements account for almost 95%
+    // of all insignificant Docbook whitespace in manual.
+
+    $trimList[] = 'refentry';
+    $trimList[] = 'fieldsynopsis';
+    $trimList[] = 'variablelist';
+    $trimList[] = 'example';
+    $trimList[] = 'simplelist';
+    $trimList[] = 'refnamediv';
+    $trimList[] = 'tbody';
+    $trimList[] = 'reference';
+    $trimList[] = 'classsynopsis';
+    $trimList[] = 'section';
+    $trimList[] = 'tgroup';
+    $trimList[] = 'thead';
+    $trimList[] = 'itemizedlist';
+    $trimList[] = 'note';
+    $trimList[] = 'sect2';
+
+    xml_trim_docbook_wsi( $doc->documentElement , $trimList );
 }
-else
+
+function xml_trim_docbook_wsi( DOMNode $parent , array $trimList )
 {
-    echo "failed.\n";
-    print_xml_errors();
-    individual_xml_broken_check();
-    errors_are_bad(1);
+    $nodeList = $parent->childNodes;
+    for ( $idx = $nodeList->length - 1 ; $idx >= 0 ; $idx-- )
+    {
+        $node = $nodeList[ $idx ];
+        $type = $node->nodeType;
+
+        if ( $type == 1 ) // XML_ELEMENT_NODE
+            xml_trim_docbook_wsi( $node , $trimList );
+
+        if ( $type == 3 ) // XML_TEXT_NODE
+            if ( in_array( $parent->nodeName , $trimList ) )
+                if ( trim( $node->nodeValue ) == '' )
+                    $parent->removeChild( $node );
+    }
 }
 
-function individual_xml_broken_check()
+function xml_broken_file_check()
 {
     $cmd = array();
     $cmd[] = $GLOBALS['ac']['PHP'];
diff --git a/docbook/docbookwsi.php b/docbook/docbookwsi.php
new file mode 100644
index 0000000000..58a20ef002
--- /dev/null
+++ b/docbook/docbookwsi.php
@@ -0,0 +1,193 @@
+<?php
+// SPDX-License-Identifier: 0BSD
+// © André L F S Bacci <ae#php.net>
+/*
+This script reads a RelaxNG XML file, and calculates all elements
+that _not_ contain <text/> contents, in all alternatives. That is,
+the list shows all elements that can have all inter-element
+whitespace removed, without affecting the expected parsing of the
+XML document that follows the RelaxNG specification.
+
+If run with a second XML argument, the script will calculate all savings
+that can be done by stripping these insignificant whitespace between
+elements.
+
+See mentions of 'docbookwsi' on source code for parts that need to be
+updated if/when a new version of Docbook is used.                */
+
+$arg0 = array_shift( $argv ) ?? null;
+$rngFile = array_shift( $argv ) ?? null;
+$xmlFile = array_shift( $argv ) ?? null;
+
+if ( $rngFile == null )
+{
+    print "Usage: '$argv0' rngFile [xmlFile]\n\n";
+    return;
+}
+
+$list = generate_element_trim_list( $rngFile );
+
+if ( $xmlFile == null )
+{
+    foreach( $list as $elem => $hasText )
+        if ( ! $hasText )
+            print "$elem\n";
+    exit( 0 );
+}
+else
+    xml_trim_stats( $xmlFile , $list );
+
+exit( 0 );
+
+function generate_element_trim_list( string $rngFilename ) : array
+{
+    $doc = new DOMDocument();
+    if ( ! $doc->load( $rngFilename , LIBXML_NOBLANKS ) )
+        throw new Exception( "XML load failed.\n" );
+
+    // First, we get all elements definitions that directly
+    // mentions <text/>, and also gather all <ref>s they refer.
+
+    $elemText = [];
+    $elemRefs = [];
+
+    $xpath1 = new DOMXpath( $doc );
+    $xpath2 = new DOMXpath( $doc );
+    $xpath1->registerNamespace ( 'rng' , 'http://relaxng.org/ns/structure/1.0' );
+    $xpath2->registerNamespace ( 'rng' , 'http://relaxng.org/ns/structure/1.0' );
+
+    $list = $xpath1->query( '//rng:element' );
+    foreach( $list as $elem )
+    {
+        $name = $elem->getAttribute( 'name' );
+        if ( $name == '' )
+            continue;
+
+        $text = count ( $xpath2->query( './/rng:text' , $elem ) );
+        $refs = $xpath2->query( './/rng:ref' , $elem );
+
+        $elemText[ $name ] = $text;
+        $elemRefs[ $name ] = [];
+
+        foreach( $refs as $ref )
+        {
+            $refName = $ref->getAttribute( 'name' );
+            $elemRefs[ $name ][] = $refName;
+        }
+    }
+
+    unset( $xpath1 );
+    unset( $xpath2 );
+
+    // After all elements are collected, and directly textual elements
+    // are marked, we can remove all <element>s, as they cannot influence
+    // if a parent element is trimmable or not, and so that any <text/>
+    // inside of a <element> cannot be found by XPaths, while exploring
+    // the original <element>'s <ref>erences.
+
+    $xpath3 = new DOMXpath( $doc );
+    $xpath3->registerNamespace ( 'rng' , 'http://relaxng.org/ns/structure/1.0' );
+
+    $todoDels = [];
+    $dels = $xpath3->query( '//rng:element' );
+    foreach( $dels as $del )
+        array_push( $todoDels , $del );
+    foreach( $todoDels as $del )
+        $del->parentNode->removeChild( $del );
+
+    // Then, we explore all references of all elements, for
+    // indirect mentions of <text/>s.
+
+    foreach( $elemText as $name => $text )
+    {
+        $text = element_references_contains_text( $doc , $name , $elemRefs[ $name ] );
+        $elemText[ $name ] |= $text;
+    }
+
+    return $elemText;
+}
+
+function element_references_contains_text( DOMDocument $doc , string $elemName , array $refs ) : bool
+{
+    $ret = false;
+    $doneRefs = [];
+    $todoRefs = array_unique( $refs );
+
+    $xpath = new DOMXpath( $doc );
+    $xpath->registerNamespace ( 'rng' , 'http://relaxng.org/ns/structure/1.0' );
+
+    while ( ( $refName = array_pop( $todoRefs ) ) != null )
+    {
+        $doneRefs[ $refName ] = true;
+
+        $defs = $xpath->query( "//rng:define[@name='$refName']" );
+        if ( $defs->count() != 1 )
+            throw new Exception( "Unique define search failed for '$refName'." );
+        $def = $defs[0];
+
+        $text = count ( $xpath->query( './/rng:text' , $def ) );
+        if ( $text )
+            return true;
+
+        $subRefs = $xpath->query( './/rng:ref' , $def );
+        foreach( $subRefs as $subRef )
+        {
+            $subRefName = $subRef->getAttribute( 'name' );
+            if ( isset( $doneRefs[ $subRefName ] ) )
+                continue;
+            $todoRefs[] = $subRefName;
+        }
+    }
+
+    return false;
+}
+
+function xml_trim_stats( string $xmlFilename , array $elemText )
+{
+    $doc = new DOMDocument();
+    if ( ! $doc->load( $xmlFilename ) )
+        throw new Exception( "XML load failed.\n" );
+
+    $stats = [];
+    xml_trim_stats_enter( $doc->documentElement , $elemText, $stats );
+    arsort( $stats );
+
+    $total = 0;
+    foreach( $stats as $elem => $trimSize )
+    {
+        print "$trimSize $elem\n";
+        $total += $trimSize;
+    }
+    print "\ntotal $total\n";
+}
+
+function xml_trim_stats_enter( DOMNode $node , array $elemText , array & $stats , int $level = 0 )
+{
+    $name = $node->nodeName;
+    $text = $elemText[ $name ] ?? true;
+
+    if ( ! $text )
+    {
+        $size = 0;
+        $dels = [];
+
+        foreach( $node->childNodes as $child )
+            if ( $child->nodeType == XML_TEXT_NODE )
+                if ( trim( $child->nodeValue ) == '' )
+                    $dels[] = $child;
+
+        foreach( $dels as $del )
+        {
+            $size += strlen( $del->nodeValue );
+            $del->parentNode->removeChild( $del );
+        }
+
+        if ( isset( $stats[ $name ] ) )
+            $stats[ $name ] += $size;
+        else
+            $stats[ $name ] = $size;
+    }
+
+    foreach( $node->childNodes as $child )
+        xml_trim_stats_enter( $child , $elemText , $stats , $level + 1 );
+}
\ No newline at end of file