If the input to the expression parser contains boolean, null or undefined literals, the parsed expression is built using SymbolNodes instead of ConstantNodes. However, the ConstantNode explicitly contains code that would handle these types, so I believe this is a bug.
On the other hand, string and number literals are parsed to ConstantNodes.
Example:
'use strict';
/* jshint node: true */
var math = require('mathjs');
var parsed = math.parse('a = null; b=true; c=undefined; d = "abc"; e=123');
function print (node, level) {
console.log(" ".repeat(level),"type: ",node.type,",",node.toString());
node.forEach(function (sub) {print(sub,level+1);});
}
print(parsed,0);
Output:
type: BlockNode , a = null;
b = true;
c = undefined;
d = "abc";
e = 123
type: AssignmentNode , a = null
type: SymbolNode , a
type: SymbolNode , null
type: AssignmentNode , b = true
type: SymbolNode , b
type: SymbolNode , true
type: AssignmentNode , c = undefined
type: SymbolNode , c
type: SymbolNode , undefined
type: AssignmentNode , d = "abc"
type: SymbolNode , d
type: ConstantNode , "abc"
type: AssignmentNode , e = 123
type: SymbolNode , e
type: ConstantNode , 123
As can be seen from the output, the first three literals create SymbolNodes, but the last two ConstantNodes.
The result is that the "symbols": "true", "false", "undefined" and "null" can be overridden from the scope when evaluating such expressions, but probably this is undesirable anyway and adds unnecessary code to the compiled expression:
(this is the compiled result of a=null):
scope["a"] = ("null" in scope ? getSafeProperty(scope, "null") : getSafeProperty(math, "null"))
Which is a lot of overhead if such literals occur frequently.
Thanks for bringing this up Pa谩l. From the parsers point of view symbols like true and pi are all the same and are only evaluated at runtime, which works just fine. I agree that from a user point of view (looking at a parsed expression tree) it would be more logic to get a ConstantNode rather than a SymbolNode. Makes sense to me to change this behavior, let's schedule this for the first next breaking release (v4).
Besides true, false, null, and undefined, we should also implement NaN, Infinity, and uninitialized this way. Here, NaN and Infinity have a different value depending on whether you have configured to use numbers or bignumbers.
I've dropped constant uninitialized. This was mainly used for keeping new entries undefined when resizing matrices. Instead you can use null to indicate entries that aren't explicitly set.
This has been adressed in v4
Most helpful comment
Thanks for bringing this up Pa谩l. From the parsers point of view symbols like
trueandpiare all the same and are only evaluated at runtime, which works just fine. I agree that from a user point of view (looking at a parsed expression tree) it would be more logic to get aConstantNoderather than aSymbolNode. Makes sense to me to change this behavior, let's schedule this for the first next breaking release (v4).Besides
true,false,null, andundefined, we should also implementNaN,Infinity, anduninitializedthis way. Here,NaNandInfinityhave a different value depending on whether you have configured to use numbers or bignumbers.