Let $(X_t)$ be a continuous one dimensional semimartingale and $f: [0,1] \times \mathbb R \rightarrow \mathbb R$ a continuous function such that $\forall x \in \mathbb R$, $f(x,\cdot)$ is $C^2$ and $\forall y \in \mathbb R$, $f(\cdot,y)$ is $C^2$ almost everywhere (and the number of points where $C^2$ fails is finite).
I would like to apply Itô's lemma on $f(t,X_t)$. The problem is that the "usual" statement of the lemma requires that $f$ is $C^2$, which is not the case here. Can I "cut" the integration in Itô's lemma at all points of failure ? For example, suppose that the only point where $C^2$ fail is $\frac 1 2$. Instead applying Itô's lemma on $f(t,X_t) - f(0,X_0)$, I would apply it to $f(t,X_t) - f(\frac 1 2,X_{\frac 1 2})$ and to $f(\frac 1 2,X_{\frac 1 2}) - f(0,X_0)$ and sum the results. Does that break anything ?