notes-computer-programming-REST GET is referentially transparent

While reading an example from http://www.infoq.com/articles/tilkov-rest-doubts about GET being used to implement computation (the example was similar to using the resource http://example.com/sum?addend1=2&addend2=3 to refer to the result of the computation 2+3), I realized that the idea that the GET method is "safe" and "cacheable" is almost the same as referential transparency. "Safe" combined with "stateless" means no side-effects, and "cacheable" means it won't change over time (now, even RESTful GET is a little different from that, because it's okay for the contents of resources to change sometimes -- which is why I say "almost the same").

So, the reason that languages like Haskell are more inherently parallelizable, and the reason that the RESTful web is scalable are similar. In the following, remember than fetching a resource with GET is like calling a referentially transparent function:

• except for when the output of one function becomes the input of another, different servers/CPUs can execute multiple functions at once, because GET has no side-effects (it is "safe" and "stateless" -- that is, it doesn't change state at all)
• because GET has no side-effects, it doesn't matter if two different servers deal with two different GET requests in sequence, even if they both come from the same client. Similarly, each step in a composition of referentially transparent functions could be executed on different CPUs, as long as the output of the former function is transmitted
• cacheing/memoization because referentially transparent functions only need to be called once

Neat!