I'm curious about the following argument in Morrey ("Multiple integrals in the calculus of variations", Lemma 2.3.1). Suppose $f\in L^1[0,1]$ and $$\int_0^1 fg\,dx=0$$ for every test function $g\in C^\infty_c[0,1]$. Then, "by approximations," the same is true if $g$ is merely bounded and measurable, in which case we may take $g=\mathrm{sign}(f)f,$ which implies $f=0$ almost everywhere.
How exactly is one supposed to approximate? Certainly not in the $L^\infty$ norm because the closure of $C^\infty_c[0,1]$ in this norm is not $L^\infty$ itself. Furthermore, $\mathrm{sign}(f)f$ need not be in $L^\infty$, so how does this argument work?