I found a quite extreme edge case, by which set! warn-on-reflection false, while would be overriden cleanly when analyzing such a ns directly, would have effect when that ns is a transitive dep of some other ns.
No amount of hacking could correct this, is probably it's a defect in t.ana.
As a possible fix, one could abstain from evaling forms detected to be exactly set! warn-on-reflection false, and/or rewrite similar ones to true.
I found a quite extreme edge case, by which
set! warn-on-reflection false, while would be overriden cleanly when analyzing such a ns directly, would have effect when that ns is a transitive dep of some other ns.No amount of hacking could correct this, is probably it's a defect in t.ana.
As a possible fix, one could abstain from evaling forms detected to be exactly
set! warn-on-reflection false, and/or rewrite similar ones totrue.