It's compatible enough with older SBCL - extracting the expander's arglist is just a gensym, not technically wrong but not as useful as the destructuring lambda list.

While this could be considered a regression, I don't think it needs conditionalizing on SBCL release. The way to do so, in theory, is check whether (:type :lambda-list) is a kind of info by asking for meta-info ... except that it won't actually fix anything because current SBCL reports that (:type :lambda-list) exists, but it never has data.