Current approach consist on preprocessing, meaning expression string is manipulated to replace the magic word by its value,  regardless the expression language used.

This has some limitations in corner cases (nested constant keys with identical json path, impossibility of knowing if the magic word is escaped or not without double parsing...), so it will be better to try a different approach, which tries to work together with the expression parser

This approach consist on delegating to every expression implementation and try to generate a valid expression with minimal string manipulation.