[BUGFIX] Survive a null or unknown code language

[davidsteeb]

A stored code_language value can take the whole page down with a 500, in two ways.

A null language

code_language is declared as code_language text in ext_tables.sql, so the column is nullable. The TCA field has default => '', but a record can still end up with NULL, and the branch did not survive it:

if (!($processedData['data']['code_language'] ?? true)) {
    // autodetect
} else {
    $highlighted = $highlight->highlight($processedData['data']['code_language'], …);
}

?? substitutes only for NULL, so it yields true, the falsy check reads that as "a language was chosen", and NULL goes straight to the highlighter. What happens next depends on the PHP version, but the page dies either way:

PHP	result
8.4	DomainException: Unknown language: "" from Highlighter::highlight()
8.5	Using null as an array offset is deprecated in Highlighter::getLanguage(), turned into an exception by TYPO3's error handler

Normalising the value to a string before deciding puts NULL on the autodetect branch, exactly where an empty string already goes.

A language that no longer resolves

highlight() throws DomainException when the name is not registered, and nothing caught it. A record can stop resolving without ever being edited — the language may be dropped from the registered set, or the record may arrive from an installation that had it. Same outcome as above: a stored value ends the request.

That one exception is now caught and the language detected instead. A language that does resolve is still used as it was, so this is not a blanket catch.

Verified

Against a TYPO3 v14 site on PHP 8.5, one content element, only code_language varied:

code_language	before	after
NULL	HTTP 500	HTTP 200, detected
''	HTTP 200	HTTP 200, detected
'klingonisch' (not registered)	HTTP 500	HTTP 200, detected
'bash'	HTTP 200	HTTP 200, still bash

The last row is the one that matters for the catch: a language that resolves is untouched. Elements that already carried a valid language render exactly as before, and the recovered elements produce the regular markup with an hljs class on the detected language.