The fun of writing an interpreter is that you can extend it with new features easily. With `eval $1` you're stuck with whatever it does (unless you want to do

The fun of writing an interpreter is that you can extend it with new features easily. With `eval $1` you're stuck with whatever it does (unless you want to do string manipulation of code -- it's only usually a bad idea :-) ) The book Structure and Interpretation of Computer Programs is organized by types of abstractions. The first three sections are about procedures, data, and modularity, but then the fourth is about abstracting language itself (called "metalinguistic abstraction"). This was a principle used often in classic AI research, where you built specific languages to implement different planning systems -- incidentally this is why it's called "scheme". By building a lisp in a lisp, you could make use of all the facilities of the host language and add new useful features. The book demonstrates how to add lazy evaluation, nondeterministic operators, and even Prolog-style logic programming to the language. These sorts of interpreters depend on having a compatible host language. The last section is about how you can transcend limitations by essentially simulating an appropriate CPU in the host language (i.e., a virtual machine). This is how the original Scheme was implemented in its host Lisp, which was necessary because Scheme has tail recursion and Lisps haven't traditionally had that.