> It is also about the language providing you the tools to express certain invariants that the compiler can use.
Compiled code is useless if it is never called. The moment it's called, whether the caller is known during compilation or not, the caller can pass down any additional information it has, whether for that call or to help the callee optimize future calls. Heck, if it really wants to, the compiler could simply let the "compiled" code be the AST or source code itself (plus a JITter), and make the program generate code on the fly when the caller calls it, optimized using whatever information the caller is willing to provide. Again, nothing fundamentally in a language preventing this. It a tradeoff between implementation difficulty, execution latency, program size, and the like.