I updated DPL3 for ReDOS validation, if it further broke things, apologies, but it is only meant to disallow dangerous regex patterns that can lead to ReDoS. So if it uses one, then the behavior is expected. Could you provide an example regex that is broken, and a page that shows it broken please? Thanks!

As for RegexFunctions itself, I will be looking to try to re-enable it shortly as well, or if it is still dangerous, a permanent removal of the extension entirely. Security is a foremost concern and we must be sure we only have extensions adhearing to that. I apologise for the inconvenience, but depending here, this may not be possible to fix.

Bit of a long story, but here goes...

• The dictionary entries use {{Decl}}, a tag used to determine declension codes in Tovasala; the relevant detection is {{#dplreplace:{{PAGENAME}}|([eoaiu]{{!}}é{{!}}ê{{!}}ọ{{!}}ạ{{!}}ar{{!}}(?<![eou])at)$||{{PAGENAME}}}}. (Morpheme tables are unaffected output-wise [while entries display "Unknown error"], but the coding residue from the RgxF days is another matter entirely; I'll try to clean that up as I find time.)
• Meanwhile, the entries themselves use {{#dplreplace:{{#arrayprint:word-stem}}|((?<![iu])e{{!}}ē).$|è{{#sub:{{#arrayprint:word-stem}}|-1}}}} in sub-entries for derivative forms of Tovasala headwords. (Note the {{#arrayprint:}} from the Arrays extension.) For an example of the missing stems, see "esil" (an adposition meaning "from").
• Furthermore, {{Definition/Arrays}} is awash in replacement calls.

...How was this a double post? (I only pressed "Submit" just once...)