Skip to content

Latest commit

 

History

History
130 lines (100 loc) · 4.65 KB

evaluation.md

File metadata and controls

130 lines (100 loc) · 4.65 KB

Evaluation in Cell

An evaluator takes in tree structures that come from the parser and works out what they mean.

The parser does not look up symbols to find out the function or value you are talking about - it just recognises known shapes and builds them into a structure.

The evaluator is responsible for looking up names, and then actually running the code and producing a result. The evaluator knows that 3+7=10, and it knows how to call functions.

You can find Cell's evaluator at: https://github.com/andybalaam/cell/blob/master/pycell/eval_.py and the tests for it at: https://github.com/andybalaam/cell/blob/master/tests/eval_tests.py

For example, if we have a piece of code like this:

3 + 4;

which has been parsed into a tree like this:

("operation",
    "+",
    ("number", "3"),
    ("number", "4")
)

then the evaluator will do the calculation, and return a value like this:

("number", 7)

Notice that the 7 is no longer in quotes - the evaluator first evaluated the two tree nodes and converted them from unprocessed text ("3" and "4") into real numeric values, before adding them up and returning the numeric value 7.

The evaluator keeps hold of the names of things, and it must make sure names can only be seen from the right place.

For example:

x = "World!";
myfn = {
    x = "Hello, ";
    print( x );
};
myfn();
print( x );

should print "Hello, " and then "World!". If the evaluator messes up the scoping rules, it will print "Hello, " twice, because the changed value of x will leak out.

Note: in some programming languages some leaking like this is allowed and expected, but in Cell there is no leakage.

More complicated than that is the idea of closures, where the scope of a function can follow it around, so that this program:

outerfn = {
    x = 12;
    innerfn = {
        print(x);
    };
    innerfn;
};

thing = outerfn();
thing();

should print "12", because the x defined inside outerfn has been carried around with innerfn. Here, outerfn is a function that returns another function from inside it (innerfn).

The second-last line of the program calls outerfn and stores the result (which is innerfn) inside a new name called thing. So now thing refers to innerfn. So when we call thing on the last line of the program, we are calling innerfn, and innerfn can still see x with a value of 12.

This gets much more complicated when values can change, since every time we call outerfn we get back a new closure, and each could have its own version of x. In fact, we can use this mechanism to build up something like object-oriented programs with objects (functions, here) than can hold state inside them.

[Changing values is done in the set native function, which is described on the Library page.]

To handle all the names, Cell creates an "Environment" for each function, which holds all the names defined in that function (and the values they point to, such as the number 3, a string, or another function). Each environment also points "out" to another environment representing the scope within which it was defined. At the top is the global scope, which is where names that are assigned outside of all functions go. The global scope environment does not point to any scope further "out".

So an environment is really just a name-value mapping, and an optional pointer to another environment. In a language like Python where garbage collection is done for us, its implementation is quite simple. You can find Cell's environment implementation here: https://github.com/andybalaam/cell/blob/master/pycell/env.py

The evaluator understands assignments such as x = 3; or myfn={print("foo");}; to mean it should insert a new name/value pair into the current environment.

Then when we refer to a name as in x * 2; or myfn(); the evaluator looks up the name in the current environment, and keeps looking into further out environments until it find the name, and replaces that part of the tree with the value it has found for that name.

The evaluator continues to replace parts of the tree with values it has looked up, and to combine those values together with rules it knows (such as +) until there is just one value left, which is returned somewhere, or thrown away (presumably after some side-effect like printing has happened).

In order to do useful thing like printing, and in Cell's case, even basic things like if or for, as well as changing values using set, we need some functions. These are defined in the Library.