Defeating Racket’s separate compilation guarantee

Racket stratifies every program into a compile-time part and a run-time part, and it restricts communication between them to limited, well-defined channels (mainly via expanding to code that does something at run-time). After compilation has finished, Racket ensures that compile-time side effects are thrown away, and the run-time code starts over with fresh, untouched state. The full Racket module system does not have only two phases, since macros can also be used at compile-time to implement other macros, effectively creating a separate “compile-time” for the compile-time code.

Source: lexi-lambda.github.io