Wrong default value of DOMDocument.xmlStandalone

[uuf6429]

The documentation here says that xmlStandalone defaults to false when missing, but the following test fails on PHP 8.1*:

    public function testThatStandaloneDefaultsToFalse(): void
    {
        $doc = new \DOMDocument();

        $doc->loadXML('<?xml version="1.0" encoding="US-ASCII"?><root/>');

        $this->assertFalse($doc->xmlStandalone);
    }

* I haven't tried it on other PHP versions and 3v4l.org sadly misses DOMDocument..

[cmb69]

It doesn't look like this depends on the PHP version; PHP treats that libxml2 int as bool. However, as of libxml2 2.6.31 (released 2004), it is no longer a simple bool, but rather:

    int             standalone; /* standalone document (no external refs) 
				     1 if standalone="yes"
				     0 if standalone="no"
				    -1 if there is no XML declaration
				    -2 if there is an XML declaration, but no
					standalone attribute was specified */

The W3C DOM Level 3 specs declare it as boolean, and state "This is false when unspecified." So the PHP documentation would actually be correct, but not the implementation.

However, the WHATWG DOM specification lists that interface member as removed.

Given that the WHATWG DOM specification supersedes the respective W3C specifications, I don't think that fixing the actual bug makes sense, so we should (a) document the current behavior, and (b) deprecate ::$xmlStandalone. If we don't want to deprecate that property, we should probably fix the bug (i.e. treat negative values as false). @beberlei, what do you think about this?

FWIW, we only require libxml2 ≥ 2.7.6 as of PHP 7.4.0; previously, we required only libxml2 ≥ 2.6.11, so builds with these old libxml2 versions would show a different behavior.

[uuf6429]

The W3C DOM Level 3 specs declare it as boolean, and state "This is false when unspecified." So the PHP documentation would actually be correct, but not the implementation.

Yep, that's kind of what I meant...I started investigating especially after reading the W3C spec contradicting the PHP behaviour.

However, the WHATWG DOM specification lists that interface member as removed.

On the PHP side, apparently we've deprecated version (but not xmlVersion) and standalone (but not xmlStandalone), meanwhile encoding/xmlEncoding is a similar situation but not deprecated.
The bigger problem here is that:

1. we cannot force DOMDocument not to output the XML declaration (because LIBXML_NOXMLDECL is not implemented)
2. since the xmlXX fields are readonly, the only way to change the declaration is by changing the writeable (but deprecated fields)
3. as it is right now, we can't remove the standalone attribute from the declaration either, ie the xml input will not (functionally) match the output

But these are obviously out of the scope of the documentation.

In my specific case, I think I'll change my implementation to save the root xml node and prepend the original xml declaration (instead of saving the entire document).

[cmb69]

On the PHP side, apparently we've deprecated version (but not xmlVersion) and standalone (but not xmlStandalone), meanwhile encoding/xmlEncoding is a similar situation but not deprecated.

Ugh. None of these properties are actually deprecated (E_DEPRECATED); only the docs say they are (and that since 2007 at least). We should probably really deprecate ::$version and ::$standalone, as these don't appear to have ever been specified. Not sure what to do about ::$encoding and friends; W3C DOM Level 3 specifies ::$inputEncoding and ::$xmlEncoding, while WHATWG has the latter removed, and says the former would be a legacy alias of ::$characterSet which we do not yet support at all.

1. we cannot force DOMDocument not to output the XML declaration (because LIBXML_NOXMLDECL is not implemented)

Indeed, another doc bug (at least).

2. since the xmlXX fields are readonly, the only way to change the declaration is by changing the writeable (but deprecated fields)

Only ::$xmlEncoding (and ::$actualEncoding) are supposed to be readonly; ::$xmlStandalone and ::$xmlVersion are not.

[uuf6429]

Not sure what to do about ::$encoding and friends

Only ::$xmlEncoding (and ::$actualEncoding) are supposed to be readonly; ::$xmlStandalone and ::$xmlVersion are not.

That is correct, my bad.

Doesn't that also mean encoding is PHP's way of providing a writable property (since the other two are not as per spec)?

[nielsdos]

I'm going to move this issue to php-src, as this is an implementation bug.