bug: collapse_vars is broken since 3.3.23

[thorn0]

Bug report or feature request?

Bug report

Uglify version (uglifyjs -V)

3.3.23...3.4.9

JavaScript input

// foo.js
function main() {
    var z = { 
        a: 1 
    };
    z = Object.assign({}, z, {
        a: 2
    });
    z.prop = true;
    fn(z, true);
}

function fn(a1) {
    console.log(a1.a === 2 ? 'works as expected' : 'bug');
}

main();

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

> uglifyjs foo.js -c -b -o foo1.js && node foo1
bug

JavaScript output or error produced.

function main() {
    var z = {
        a: 1
    };
    fn(z, (z = Object.assign({}, z, {
        a: 2
    })).prop = !0);
}

function fn(a1) {
    console.log(2 === a1.a ? "works as expected" : "bug");
}

main();

It works fine if I disable collapse_vars:

> uglifyjs foo.js -c collapse_vars=false -b -o foo1.js && node foo1
works as expected

BTW, I don't understand why this specific transformation is triggered by collapse_vars. As per the docs, what collapse_vars does is 'Collapse single-use non-constant variables', however in this case it seems to do something else.

[Maximaximum]

The old issue (for reference): #3096

[Maximaximum]

Any progress here? :(

[ingmarh]

Perhaps related: A fix for a collapse_vars regression has been merged in February, but hasn't been included in a new uglify-es (yet). See the commit.

We since switched to using terser (fork of uglify-es), because uglify-es seems to be no longer maintained (and terser includes the above mentioned fix).

[thorn0]

No, @ingmarh. It was a completely different issue.

[ingmarh]

@thorn0 👌 I see. I mistakenly thought it is an uglify-es issue.