I feel that, at this stage, is only about how a compiler glag, specifically "-release" works. For other configurations, there is no problem: event if the optimizer optimizes based on asserts, the asserts themselves are part of the code: code is there and the assertion will fail before execution enters the optimized path. This is just like any other optimization, nothing special about it.

The problem with "-release" might be formulated in that it optimizes based on a code that is no longer present (the asserts are wiped out). It keeps the optimization, but it dismisses the garde-fous.