TYPO3 Core

It seems to be a core bug, since the CatXmlImportManager uses

GeneralUtility::xml2tree

to transform the XML code.

That again uses xmlRecompileFromStructValArray to implode array structures back into XML.

If there is no value, this method automatically closes any opening tag into a self closing tag instead of using a closing tag for tags that are not allowed to be self closing.
If the type is detected as "cdata" the whole value is just added as is, which is why CDATA works.

    /**
     * This implodes an array of XML parts (made with xml_parse_into_struct()) into XML again.
     *
     * @param array $vals An array of XML parts, see xml2tree
     * @return string Re-compiled XML data.
     */
    public static function xmlRecompileFromStructValArray(array $vals)
    {
        $XMLcontent = '';
        foreach ($vals as $val) {
            $type = $val['type'];
            // Open tag:
            if ($type === 'open' || $type === 'complete') {
                $XMLcontent .= '<' . $val['tag'];
                if (isset($val['attributes'])) {
                    foreach ($val['attributes'] as $k => $v) {
                        $XMLcontent .= ' ' . $k . '="' . htmlspecialchars($v) . '"';
                    }
                }
                if ($type === 'complete') {
                    if (isset($val['value'])) {
                        $XMLcontent .= '>' . htmlspecialchars($val['value']) . '</' . $val['tag'] . '>';
                    } else {
                        $XMLcontent .= '/>';
                    }
                } else {
                    $XMLcontent .= '>';
                }
                if ($type === 'open' && isset($val['value'])) {
                    $XMLcontent .= htmlspecialchars($val['value']);
                }
            }
            // Finish tag:
            if ($type === 'close') {
                $XMLcontent .= '</' . $val['tag'] . '>';
            }
            // Cdata
            if ($type === 'cdata') {
                $XMLcontent .= htmlspecialchars($val['value']);
            }
        }
        return $XMLcontent;
    }

To get a better understanding of the types "open", "complete" and "close", see
http://php.net/manual/de/function.xml-parse-into-struct.php

Since the method is quite old, I guess this happens with CMS 6 and 7 as well.

According to HTML5 specs, the following tags are allowed to be "void" and therefor self closing.

area, base, br, col, embed, hr, img, input, keygen, link, meta, param, source, track, wbr

Coders.Care Extension Team wrote:

Could you please check your file again to find out if there is any <![CDATA[]]> surrounding your HTML code?
Actually this seems to make a difference at least on our testing system, since exactly your HTML-Code will be imported unchanged with CDATA but as a closed tag without it.

Hi,
Just checked - no, block with script tag is not wrapped with CDATA. Weird, I have 20 translated bodytext fields in file, but only 4 of them are wrapped in CDATA.

Coders.Care Extension Team wrote:

As a workaround you can check the checkbox "Do not check XML" - this automatically wraps content with CDATA to make sure it does not break the XML parser. Still this is just a workaround for the L10nmgr but the bug in xmlRecompileFromStructValArray should be fixed anyway.

It's already checked. If I don't check it - export fails.

Can somebody summarize what the core bug here is and what needs fixing exactly?