Update README.md

corywalker · Jan 23, 2017 · 4c17577 · 4c17577
1 parent 62b8c1d
commit 4c17577
Showing 1 changed file with 2 additions and 0 deletions.
diff --git a/README.md b/README.md
@@ -77,6 +77,8 @@ The pattern matching system can be very slow, especially when working with `Orde
 
 I'm interested in trying to apply Golang's concurrency paradigms to the evaluation sequence. Some low hanging fruit would be to have parallel computation of mapping pure functions onto Lists or other expressions (computing the derivative of a list expressions). Similarly, supporting automatic threading of Listable functions would be nice (computing sin(x) of a large array). The evaluation of an expression often starts with evaluating each of the parameters at the beginning. This could potentially be made concurrent as well. A more complicated but very interesting application would be to break down the pattern matching engine into concurrent components. We would have to be very careful about side effects here, so we might need to overhaul our scoping constructs or somehow restrict access to the EvalState. Another option would be to create a function that predicts if another function has side effects (is this feasible?). A true prediction could allow the system to fall back to non-concurrent evaluation.
 
+Since there are at least two other replacement engines that implement the same syntax that I know of ([another one here](https://github.com/jyh1/mmaclone)), it would could be useful to decide on a standard link protocol such that the replacement engine is independent of the rules that run on top of it. Another layer of abstraction is the frontend. Really the hierarchy goes as follows: core <- rules <- frontend. It would be nice to see all of these functions factored out and interchangeable.
+
 Also of interest is to build up some formal theory on the rule definitions. There should be some pre-existing literature on this, as term-rewriting is a studied field. Some interesting questions to answer are:
 
 1. Given a set of rules for a symbol (or the universe of rules?), can we find duplicates or reduce the set of rules to the most fundamental ones? Answering such a question would improve the efficiency and clarity of the system.