It certainly leads to code for a different purpose. In an interpreter you need to store enough information so you can choose the right subroutine to call at run

It certainly leads to code for a different purpose. In an interpreter you need to store enough information so you can choose the right subroutine to call at runtime. This must also include enough information as inputs to let the subroutine produce the right results. Generally I think of this as a reference to code and references to data, and a references to where to store the results. For a compiler you need all that information to decide what code to emit but you don’t need to keep it around when the emitted code is used, unless you want to debug or replicate decisions at runtime that were already resolved at compile time. For example, if you had an index into a block of memory that you could prove was only large enough to address elements in the block, there is no need to check at runtime that it won’t be used to create out of range references for a compiler. If the same operation was done by an interpreter you might still check because the code that does the interpretation may not be solely used for that instance of the indexing/addressing operation, as you can guarantee with code produced by a compiler.