Comment made by: rickmoynihan
Thanks for upping my privileges Alex.
I didn't mean to over egg the pudding, but perhaps I ended up doing it all the same! :-)
My reasoning was that if you're using a transducer to map/reduce over values taken from user input, then an attacker could cause a loop to abort earlier than expected; which might then have other consequences. I did't have a specific attack vector in mind, but I'd seen also the sentinel ::halt and had imagined that a user could inject "clojure.core/halt" into a HTTP header. If ring has a middleware to keywordize keys then you have the sentinel value for a transducer injected by a user, and that sentinel could then potentially be used to shortcut the validation/encoding/escaping etc of other fields.
However I've looked a bit more at the implementation of {{halt-when}} and other transducers and it seems the key isn't ever mixed with user data; and that ::none is used to indicate the first pass over by some transducer functions. So it looks like I was mistaken. I doubt now that this is worth resolving.