In (Curtright, Fairlie, Zachos 2014), the authors mention (Eq. (14) in this online version) the following relation, known as "Bopp shifts": $$f(x,p)\star g(x,p)=f\left(x+\frac{i}{2}\partial_p,p-\frac{i}{2}\partial_x\right) g(x,p),\tag1$$ where the $\star$-product is defined as $$\star\equiv\exp\left[ \frac{i}{2}(\partial_x^L\partial_p^R - \partial_p^L \partial_x^R)\right],\tag2$$ and I'm denoting with $\partial_i^L,\partial_i^R$ the partial derivative $\partial_i$ applied to the left or right, respectively.
I'm trying to get a better understanding of where this comes from. As far as I understand, $f$ and $g$ are regular functions here (usually Wigner functions I suppose), so $f\star g$ should produce another "regular" function. If this is so, what do the derivatives in the argument of $f$ mean exactly? If I were to simply apply (2) to $f\star g$, I would naively get the following expression: $$f\star g = \sum_{s=0}^\infty \frac{(i/2)^s}{s!} \sum_{k=0}^s (-1)^k (\partial_x^{s-k}\partial_p^k f) (\partial_x^k \partial_p^{s-k}g). \tag3 $$ How is this compatible with (1)?
In fairness, if I were to very handwavily apply (2) to (1) without being too careful, I could think of $\partial_p^R$ and $\partial_x^R$ in the exponential as $c$ numbers, and the $\star$ operator as only acting on $f$, so that $\frac{i}{2}\partial_x^L\partial_p^R$ would be the operator enacting the translation $x\mapsto x+\frac{i}{2}\partial_p$, and similarly for the other term in the exponential. At a purely formal level this seems to make sense, but more concretely I'm not sure what the expression (1) is even supposed to represent, and how it is consistent with (3).