I used Mercury intensively for about a year a while ago. I think what ultimately turned me off was the incompleteness of the implementation (several of the most

I used Mercury intensively for about a year a while ago. I think what ultimately turned me off was the incompleteness of the implementation (several of the most interesting features were only 80% implemented, and the unimplemented corner cases were undocumented and caused the compiler to crash, so you just had to guess what it was you were doing that was too advanced), and the incomprehensibility of the error messages (an order of magnitude even more incomprehensible than C++ template errors). Notably, the lack of support from the instantiatedness system for partially instantiated structures put Mercury at a big disadvantage to Prolog's expressiveness in my opinion. So now, instead of using Mercury and wishing it better supported partial instantiation and had comprehensible error messages, I use Prolog and wish it had even a basic static analyzer and support for the !Var and @Var syntaxes.