This endeavor, and the kind of difficulties it encounters, reminds me la of endeavor of making a flat map of the Earth (e.g., the Mercator projection), and the kind of difficulties it encounters.
Here are some interesting quotes from the Wikipedia article on SMILES.
“Typically, a number of equally valid SMILES strings can be written for a molecule.”
“the algorithm fails for a number of simple cases (e.g. cuneane, 1,2-dicyclopropylethane) and cannot be considered a correct method for representing a graph canonically. There is currently no systematic comparison across commercial software to test if such flaws exist in those packages.”