[ES6] bug: minifying const variable with inline parameter reuses same variable name for const and var, giving 'n is not defined' error

[tung-jin-chew-hp]

Bug report or feature request?

bug

ES5 or ES6+ input?

ES6

Uglify version (uglifyjs -V)

uglify-es 3.3.8

JavaScript input

var test = (function(){
    function inlinedFunction(data) {
        return {
            children: data,
            count: data.reduce(function(a, b) {
                return a + b;
            })
        }
    }

    function testMinify(){
        if(true) {
            const data = inlinedFunction([1,2,3]);

            if(this.someFunction) {
                this.someFunction(data);
            }
        }
    }

    return testMinify();
})();

The uglifyjs CLI command executed or minify() options used.

uglifyjs -b -c -m -- test.js

JavaScript output or error produced.

Chrome gives an 'Uncaught ReferenceError: n is not defined' error on line 5 when the minified output is pasted into the console and run, because it's trying to assign to 'n' as a variable and a constant.

The minified output looks like the following:

var test = function() {
    return function() {
        {
            const n = {
                children: n = [ 1, 2, 3 ],
                count: n.reduce(function(n, t) {
                    return n + t;
                })
            };
            this.someFunction && this.someFunction(n);
        }
        var n;
    }();
}();

Actually encountered it elsewhere, but this is as simple as I can get the test case. Both the inlined function and the containing function seem to need to use their variables twice, otherwise it's all optimized away. The conditional is also necessary, though it doesn't have to be if(true) , just using that as an example.

[kzc]

Regression from uglify-es@3.3.7.

Workaround:

uglifyjs -m -c inline=false

[kzc]

@alexlamsl This is a case where the rename pass in uglify-es@3.3.7 masked the issue.

$ bin/uglifyjs -V
uglify-es 3.3.7
$ 
$ cat t2842.js | bin/uglifyjs -bmc | node
$ 
$ cat t2842.js | bin/uglifyjs -bmc --no-rename | node
[stdin]:5
                children: n = [ 1, 2, 3 ],
                            ^
ReferenceError: n is not defined

[tung-jin-chew-hp]

Thanks, that's useful to know! Both inline=false and inline=1 work, it only gives errors on inline >= 2.

[alexlamsl]

@kzc may be the variable name collision logic of inline not working for AST_SymbolBlockDeclaration inside AST_Block (as opposed to AST_Scope).

[kzc]

Above my pay grade.

Can you suggest a simple fix?

[kzc]

OT - function inlining is a lot trickier than I first thought. So many bloody edge cases.

[alexlamsl]

Let me finish my $day_job in a couple of hours, and I'll see what I can do.

[alexlamsl]

@kzc AST_Node.block_scope is easily lost during compress, so unless we track down and propagate this there is no way inline can safely operate.

So the simplest fix would be to change inline default value to 1 and call it a day.

[alexlamsl]

FWIW, here's a slighly more reduced test case:

(function() {
    function inlinedFunction(data) {
        return data[data[0]];
    }
    function testMinify() {
        if (true) {
            const data = inlinedFunction([1, 2, 3]);
            console.log(data);
        }
    }
    return testMinify();
})();

[kzc]

So the simplest fix would be to change inline default value to 1 and call it a day.

Then we'd lose inline=2 for any ES5 code processed with uglify-es.

There must be way to identify whether symbols within the function to be inlined conflict with any non-var variable names in the caller at the point of the call.

Let me think about it.

[alexlamsl]

Then we'd lose inline=2 for any ES5 code processed with uglify-es.

They should be using uglify-js in the first place.

[kzc]

They should be using uglify-js in the first place.

Most new projects favor uglify-es over uglify-js@3.