`Dom\XMLDocument::C14N()` seems broken compared to `DOMDocument::C14N()`

[aprat84]

Description

The following code:

$xml = <<<EOF
<?xml version="1.0" encoding="UTF-8"?>
<test:root xmlns:test="http://example.com/dummy/ns">
    <test:a/>
    <test:b test:kind="123">abc</test:b>
</test:root>
EOF;

$d = \Dom\XMLDocument::createFromString($xml);
var_dump($d->C14N(true));

Resulted in this output:

<test:root xmlns:test="http://example.com/dummy/ns" xmlns:xmlns="http://www.w3.org/2000/xmlns/" xmlns:test="http://example.com/dummy/ns">
    <test:a></test:a>
    <test:b test:kind="123">abc</test:b>
</test:root>

But I expected this output instead:

<test:root xmlns:test="http://example.com/dummy/ns">
    <test:a></test:a>
    <test:b test:kind="123">abc</test:b>
</test:root>

However, this code works:

$xml = <<<EOF
<?xml version="1.0" encoding="UTF-8"?>
<test:root xmlns:test="http://example.com/dummy/ns">
    <test:a/>
    <test:b test:kind="123">abc</test:b>
</test:root>
EOF;

$d = new DOMDocument();
$d->loadXML($xml);
var_dump($d->C14N(true));

Shouldn't the new \Dom\XMLDocument class give the same result as the "old" \DOMDocument class?

PHP Version

PHP 8.4.14 (cli) (built: Oct 24 2025 19:07:34) (NTS)
Copyright (c) The PHP Group
Built by https://github.com/docker-library/php
Zend Engine v4.4.14, Copyright (c) Zend Technologies
    with Zend OPcache v8.4.14, Copyright (c), by Zend Technologies

Operating System

Alpine Linux v3.22.2 (php:8.4-fpm-alpine docker image)

[ndossche]

Didn't yet find something better than this https://gist.github.com/ndossche/34a44c33f54d253ebe961203ef59a2e6

[aprat84]

The non-exclusive variant also produces weird results, repeating the namespace declaration in every first level node:

$xml = <<<EOF
<?xml version="1.0" encoding="UTF-8"?>
<ns1:root xmlns:ns1="http://example.com/dummy/ns">
    <ns1:a/>
    <ns1:b>
        <ns1:c>123</ns1:c>
    </ns1:b>
</ns1:root>
EOF;

$d = \Dom\XMLDocument::createFromString($xml);
var_dump($d->C14N());

$d = new DOMDocument();
$d->loadXML($xml);
var_dump($d->C14N());

The first one outputs this:

<ns1:root xmlns:ns1="http://example.com/dummy/ns">
    <ns1:a xmlns:ns1="http://example.com/dummy/ns"></ns1:a>
    <ns1:b xmlns:ns1="http://example.com/dummy/ns">
        <ns1:c>123</ns1:c>
    </ns1:b>
</ns1:root>

While the second one outputs this:

<ns1:root xmlns:ns1="http://example.com/dummy/ns">
    <ns1:a></ns1:a>
    <ns1:b>
        <ns1:c>123</ns1:c>
    </ns1:b>
</ns1:root>

Also note that the DOMDocument::C14N() method produces the same results as the xmllint [--c14n|--exc-c14n] <file> command (libxml2-2.13.9-r0).

[ndossche]

Yes I figured that out too when I experimented with the code yesterday evening.
The root cause is that the new DOM code stores namespaces in a different way, and C14N has very strict rules about how namespaces should get stored. So I need to manually insert fake temporary declarations to satisfy the constraints of the C14N engine.
I tried other tricks but the problem is that in the modern DOM specification there are no real namespace declarations nodes (they're proper attributes), so we don't have a fixed insertion point.