Rich-Harriscode-red
Experimental toolkit for writing x-to-JavaScript compilers. It is used in Svelte.
API
The code-red package exposes three core functions — b, x and print.
b and x take a template literal and return an ESTree program body, or a single node:
import { b, x } from 'code-red';
const expression = x`i + j`;
assert.equal(expression.type, 'AssignmentExpression');
assert.equal(expression.operator, '+');
assert.equal(expression.left.name, 'i');
assert.equal(expression.right.name, 'j');
const body = b`
const i = 1;
const j = 2;
const k = i + j;
`;
assert.equal(body.length, 3);
assert.equal(body[0].type, 'VariableDeclaration');Expressions in template literals correspond to replacement nodes — so you could express the above like so:
const i = x`i`;
const j = x`j`;
const expression = x`${i} + ${j}`;
const body = b`
const ${i} = 1;
const ${j} = 2;
const k = ${expression};
`;The print function takes a node and turns it into a {code, map} object:
const add = x`
function add(${i}, ${j}) {
return ${expression};
}
`;
print(add).code;
/*
function add(i, j) {
return i + j;
}
*/
i.name = 'foo';
j.name = 'bar';
print(add).code;
/*
function add(foo, bar) {
return foo + bar;
}
*/Prefixes
@-prefixed names (replaceable globals)
So that you can use globals in your code. In Svelte, we use this to insert utility functions.
// input
import { x } from 'code-red';
x`@foo(bar)`
// output
FOO(bar)#-prefixed names (automatically deconflicted names)
So that you can insert variables in your code without worrying if they clash with existing variable names.
bar used in user code and in inserted code gets a $1 suffix:
// input
import { x } from 'code-red';
x`
function foo(#bar) {
return #bar * bar;
}`;
// output
function foo(bar$1) {
return bar$1 * bar;
}Without conflicts, no $1 suffix:
// input
import { b } from 'code-red';
b`const foo = #bar => #bar * 2`;
// output
const foo = bar => bar * 2;Optimiser
TODO add an optimiser that e.g. collapses consecutive identical if blocks
Compiler
TODO add a code-red/compiler module that replaces template literals with the nodes they evaluate to, so that there's nothing to parse at runtime.
Sourcemaps
TODO support source mappings for inserted nodes with location information.