"Expected: :" error in ExtendScript

[fusepilot]

I'm trying to use Uglify to compress and obscure Adobe ExtendScript ( ES3 dialect of JS ) code. Uglify seems to strip parentheses in nested ternary expressions. ExtendScript can't understand this and throws an error: Expected: :.

Input code:

// input.js

function toInteger(value) {
  var result = parseFloat(value);
  var remainder = result % 1;

  return result === result ? (remainder ? result - remainder : result) : 0;
}

toInteger(2.1); // Result: 2

Output code from uglifyjs -b -c -o output.js -- input.js:

// output.js

function toInteger(value) {
    var result = parseFloat(value), remainder = result % 1;
    return result === result ? remainder ? result - remainder : result : 0; // Error: Expected: :
}

toInteger(2.1);

I'm aware how edge case this is, but it would be great if there was a compression option to keep the parentheses in for this scenario.

[kzc]

There's no such compression option. But if you want to add parens to the consequent of a conditional here's the UglifyJS2 patch:

--- a/lib/output.js
+++ b/lib/output.js
@@ -639,7 +639,7 @@ function OutputStream(options) {
         if (p instanceof AST_Call && p.expression === this)
             return true;
         // (a = foo) ? bar : baz
-        if (p instanceof AST_Conditional && p.condition === this)
+        if (p instanceof AST_Conditional && (p.condition === this || p.consequent === this))
             return true;
         // (a = foo)["prop"] —or— (a = foo).prop
         if (p instanceof AST_PropAccess && p.expression === this)

If you have the same issue with the ternary alternative you'd incorporate a similar change with p.alternative.

Without patch:

$ echo 'function foo(x,y,z) {return x ? y ? z-1 : x-z : z;}' | uglifyjs -c
function foo(x,y,z){return x?y?z-1:x-z:z}

With the patch:

$ echo 'function foo(x,y,z) {return x ? y ? z-1 : x-z : z;}' | bin/uglifyjs -c
function foo(x,y,z){return x?(y?z-1:x-z):z}

Since this patch makes the output larger it will not be incorporated into uglify. You should probably report this ExtendScript grammar bug to Adobe as it's not ECMAScript compliant.

[rvanvelzen]

UglifyJS targets ECMAScript compliant engines. I'm sorry, this is not going to happen in the upstream, not even with a flag.

Thanks for the report!

[avdg]

I think this is a valid es 5.1 bug (at the side of the adobe interpreter), but I'll have to document the valid parser waterfall before looking for a fix (edit: which is fix their software).

Edit: done

ConditionalExpression: LogicExpression '?' AssignmentExpression ':' AssignmentExpression

Eventually, very deep in the parse tree, an assignementExpression accepts a primaryExpression, which accepts '(' expression ')', which acccepts another conditionalExpression.

[avdg]

Parentheses aren't a compress option, they are read once, converted to ast and inserted again at the output phase if needed (to avoid confusion). Compression is done in AST trees and since the relations there are clear, there is no need for semicolons in the trees themselves.

[kzc]

Major JS engines and browsers accept the code without parens. ExtendScript has a bug.

[avdg]

Yeah, they should fix their engines and meh I care too much about even that piece of closed software.

[rvanvelzen]

I would be willing to re-open this if ExtendScript were a major engine currently in use. I've never of it except for 2 reports here.

Does anyone know whether ExtendScript has significant (enough) market share to consider this?

[vespakoen]

I use ExtendScript as well and ran into the same problem.
I have avoided ternary expressions all together, and am wondering if I can stop UglifyJS optimising if/else's into the ternary expressions all together, I have tried these already without success:

{
  if_return: false,
  conditionals: false,
  booleans: false,
  comparisons: false
}

I also suggest that we "fix" this by adding an option to UglifyJS.

Making Adobe change something in ExtendScript and waiting for everyone to update is going to take many, many years.