Description:
------------
It appears that DOM output formatting activated with the DOMDocument's formatOutput property fails as soon as the document's root element ($doc -> documentElement) contains a text node (even besides other nodes).

Reproduce code:
---------------
function pre_ent_dump($var) {
	printf("<pre>\n%s</pre>\n", htmlentities($var));
}
function pre_var_dump($var) {
	echo "<pre>\n"; var_dump($var); echo "</pre>\n";
}	

$doc = new DOMDocument();
$doc -> load('file.xml');
$doc -> formatOutput = TRUE;

pre_ent_dump($doc -> saveXML()); // (1)

$root = $doc -> documentElement;

pre_var_dump($root -> textContent); // (2)

$child = $doc -> createElement('child');
$root -> appendChild($child);

pre_ent_dump($doc -> saveXML()); // (3)

Actual result:
--------------
[Note : this report is only long because of many examples with slight variations]

PHP Version 5.1.3RC2-dev (2006-03-19 15:30)
(rolled out specially to confirm the issue, also occurring in 5.1.2)
Apache2 2.0.55, WinXP SP2

The reproduce code shows what we are doing : simply	 loading the contents of an XML file, dumping those contents, getting the root element, dumping the possible text content of itself and its descendants, adding a child to the root element, and dumping the whole document contents again. Everything depends, therefore, on the contents of file.xml prior to execution.

When file.xml is :

<?xml version="1.0" standalone="yes"?>
<root></root>

The contents output at (1) and (3) are :

<?xml version="1.0" standalone="yes"?>
<root/>

<?xml version="1.0" standalone="yes"?>
<root>
  <child/>
</root>

As expected. But when file.xml is :

<?xml version="1.0" standalone="yes"?>
<root> </root>

Then the contents output at (1) and (3) are :

<?xml version="1.0" standalone="yes"?>
<root> </root>

<?xml version="1.0" standalone="yes"?>
<root> <child/></root>

(1) appears to be as expected, but in (3) the new child is no longer on a new line and indented. That absence of formatting does not occur with just any sort of child to the root element however, as we now demonstrate with the following contents for file.xml :

<?xml version="1.0" standalone="yes"?>
<root><a></a></root>

The contents output at (1) and (3) are :

<?xml version="1.0" standalone="yes"?>
<root>
  <a/>
</root>

<?xml version="1.0" standalone="yes"?>
<root>
  <a/>
  <child/>
</root>

Both as expected. If we add a text node to the root element though (we use whitespace but this seems to apply to any sequence of characters) :

<?xml version="1.0" standalone="yes"?>
<root> <a></a></root>

Then the contents output at (1) and (3) are :

<?xml version="1.0" standalone="yes"?>
<root> <a/></root>

<?xml version="1.0" standalone="yes"?>
<root> <a/><child/></root>

Now we realize that *both* outputs are broken (the <a /> in (1) should be on its own line, like above).
Interestingly, the text nodes only cause trouble when they are at the root. When the contents of file.xml are :

<?xml version="1.0" standalone="yes"?>
<root><a>
</a></root>

Then the contents output at (1) and (3) are :

<?xml version="1.0" standalone="yes"?>
<root>
  <a>
</a>
</root>

<?xml version="1.0" standalone="yes"?>
<root>
  <a>
</a>
  <child/>
</root>

As they should be.

I have tried working around by loading with loadXML(file_get_contents('file.xml')), by going to and fro with SimpleXML, by cancelling and setting formatOutput again at different stages... Obviously the problem is not there.

I do not know if or how this is related to bugs 23726 or 35673 ; it seems to me I have more details here, but I won't be surprised if you tell me the problem is at the libxml2 level...

Thanks is advance.

P.S. Side note -- nothing to do here, but the bug report CAPTCHA image does not seem to work under Firefox 1.5 win32 (works fine in IE), nor does it accept the CAPTCHA if I get the source, view the image in IE, and type it in Firefox.