The product sides of the Rogers--Ramanujan identities and alike often appear to be "transparently modular" (functions). The old work by Rogers (1894) and recent work by Rosengren make use (somewhat implicitly) of this fact for proving the identities with the help of underlying modular equations$-$the main challenge is verifying the latter for the sum sides. Here we speculate on the potentials of using the asymptotics of such $q$-identities or their finite versions for proving them.