Question

If I output the string before adding it as a text node to the DOMDocument tree then I can see that the original UTF-8 encoding is preserved. All umlauts are UTF-8 encoded – for sure.

Then I add the string and output the DOM-tree-object through saveXML() and all umlauts have been replaced by their respective numerical entity.

I create the DOMDocument like this: $xmlDoc = new \DOMDocument('1.0', 'UTF-8');

Shouldn’t XML then keep all UTF-8 encoded chars alone as long as they aren’t XML-special chars?

Answer 1

I don’t think that this is a bug. DOMDocument::loadXML() simply seems to override the internal version and encoding settings with the ones detected in the given XML string – actually overriding everything that has been set in the DOMDocument constructor.

$d = new DOMDocument('1.0', 'UTF-8');
$d->loadXML('<root><node/></root>');
var_dump(array(
    'encoding' => $d->encoding, // NULL
    'version'    => $d->xmlVersion // 1.0
));
$d->loadXML('<?xml version="1.0" encoding="ISO-8859-1"?><root><node/></root>');
var_dump(array(
    'encoding' => $d->encoding, // ISO-8859-1
    'version'    => $d->xmlVersion // 1.0
));

So if you’re using DOMDocument::loadXML() you have to ensure that the XML string contains a valid XML declaration.
The constructor arguments are used when you built the document from scratch.