I remember someone in a conlang group trying to develop a language for a species of sentient cuttlefish who communicate by changing colour in specific patterns and arrangements. That makes more sense than a smell-based language, especially for an originally water-based species (smells tend to linger in water, and mingle and overlap in ways that make them almost impossible to differentiate).

Unfortunately that person decided it was too complicated for them, but then they were working with a full spectrum of possible colours arranged onto a 2 dimensional plane; both elements are an order or magnitude more complex than the strictly linear progression of a small number of mouth sounds.