The authors of the Dhall configuration language claim that their language is not Turing complete. They aren't confused about the meaning, they are using the same definition as computability theory. Then they claim that not being Turing complete makes their language more secure, which is highly debateable, if not false.

This author has latched onto the idea, but is very loose about the meaning of "Turing complete". At one point he claims it means that the language has side effects that can be used to interfere with or take over the host (that's not what it means). At another point he claims that Turing completeness allows Row Hammer attacks (and implies that "total functional programming" somehow prevents writing such attacks). The reality is different. Security vulnerabilities can be present in any kind of interpreter. Rowhammer requires the ability to trick the interpreter into repeatly hammering on the same contiguous array of memory, which doesn't require Turing completeness. A Turing complete language need not be vulnerable to this, if the run time data structures and interpreter are carefully designed.

In a Turing complete language, you can write a program that loops forever, and attempts to allocate an infinite amount of memory. The "total functional programming" model of Dhall prevents you from doing these things, and that's supposed to be a big win. But Dhall does not prevent you from allocating all of the memory in the machine, and it doesn't prevent you from writing a loop that iterates for billions of years. So there's no benefit at all, in terms of protecting you from malicious code.