It is a common practice to move the $dx$ around when solving ODE and we take for granted when we integrate both sides. However, I've been rather uncomfortable with this. From an analysis perspective, $dx$ itself doesn't make sense to me. Instead, we always consider $\frac{df(x)}{dx}$, or the operator $\frac{d}{dx}$ by itself.
Now, some people argue that both $dy$ and $dx$ are infinitesimal things that are not equal to zero, and thus we can move them around. But that doesn't convince me neither. Shouldn't we concretely define what $dy$ and $dx$ are before we do anything to them? Is it possible to justify moving $dx$ around by using $\delta-\epsilon$ analysis?
As for integrating both sides, what annoys me is that sometimes extra terms (usually the constant $C$) appear. It's tempting to think of "integrating both sides" as a kind of operations (maybe not binary?) just like add or subtract, but that doesn't seem right because by apply indefinite integral a function we actually get a bunch of functions. What is the right way to think of this process? Should we think about this process under the framework of algebra, maybe?
Hope my questions make sense. Thanks.