Assume we have a programming language $L \subseteq \Sigma^* $ satisfying the following natural requirements:
- Programs in $L$ always terminate. I.e., for every program $p\in L$, its semantics $[\![p]\!] : \mathbb{N}\rightarrow\mathbb{N}$ is a recursive total function.
- There is an effective enumeration of $L = \{p_0, p_1, \ldots\}$. That is, given $i$ we can compute the program (a source code string) $p_i$. This essentially requires that the syntax is reasonably regular.
- The language can be interpreted by a Turing machine. I.e., the function $r(i,n) = [\![p_i]\!](n)$ is (total) recursive.
The question then becomes: can $L$ express all the deciders? More formally, given any total $0,1$ valued recursive function $f$, can we always have $f = [\![p_i]\!]$ for some natural $i$?
Take the following function:
$$
g(n) = \begin{cases}
0 & \mbox{if } r(n,n) \neq 0 \\
1 & \mbox{otherwise}
\end{cases}
$$
We have that $g$ can be implemented by a Turing machine, since $r$ is total recursive. Since $g$ is total and returns in $\{0,1\}$ said machine is a decider.
Assume by contradiction $L$ can compute $g$, say $g = [\![p_i]\!]$ for some $i$. Then, by definition of $g$ and $r$ we have
$$
[\![p_i]\!](n) = \begin{cases}
0 & \mbox{if } [\![p_n]\!](n) \neq 0 \\
1 & \mbox{otherwise}
\end{cases}
$$
Now, what happens when $n$ is equal to $i$ ?
$$
[\![p_i]\!](i) = \begin{cases}
0 & \mbox{if } [\![p_i]\!](i) \neq 0 \\
1 & \mbox{otherwise}
\end{cases}
$$
So, $[\![p_i]\!](i)$ is zero iff it is nonzero -- contradiction.
We conclude that $g$ can not be implemented in $L$.