enhance unused

[alexlamsl]

@kzc this addresses the first two cases in #4855 (comment)

That last one is trickier since it relies on join_vars:

$ echo 'var F={}; F.a=0; F.b=1; F.c=2;' | uglify-js -bc toplevel
 

$ echo 'var F={}; F.a=0; F.b=1; F.c=2;' | uglify-js -bc toplevel,join_vars=0
var F = {};

F.a = 0, F.b = 1, F.c = 2;

which obviously won't work with self references.

I think to make that case work would require extending reduce_vars to cover object mutation in a similar manner to constant / class / function

[kzc]

A one liner - well done!

Handling the object literal self reference would be gravy. I haven't seen it in the wild.

[alexlamsl]

... except there's corner case to be fixed − will take a look after 💤

[kzc]

The third party application tests don't trigger a failure often, but are quite useful when they do.

I think ufuzz is capable of creating property cycles in data structures.

[alexlamsl]

A one liner - well done!

For every problem there is a solution that is simple, neat − and wrong.

[kzc]

As luck would have it, I know of a tool that can transform any JS into a one-liner.
;-)

[kzc]

Do you have a reduced test case for the web-tooling-benchmark.sh failure?

[alexlamsl]

Do you have a reduced test case for the web-tooling-benchmark.sh failure?

TBH I haven't even tried to reproduce the failure locally, since I've already figured out the logical fallacy before I woke up and rewrote the patch.

So if the build testing fails again that's the first thing I'll do 👻

[alexlamsl]

Been busy filtering through all the failure reports from GitHub up until recently.

https://www.githubstatus.com/incidents/cj7gzzj30411

(Yeah, it lasted much longer than claimed, but they've been like that for a while now...)

[alexlamsl]

Hmm... what are you up to?

https://github.com/mishoo/UglifyJS/runs/2404741773?check_suite_focus=true

[kzc]

@alexlamsl Here's another optimization for you to consider...

The alias y below is optimized away correctly with collapse_vars:

$ echo 'var x = Math.random(), y = x; console.log(x/y, y/x, (x+y)/x);' | uglify-js -c toplevel
var x=Math.random();console.log(x/x,x/x,(x+x)/x);

But if the code is altered slightly then the alias y will not be optimized away:

$ echo 'var x, y = x = Math.random(); console.log(x/y, y/x, (x+y)/x);' | uglify-js -c toplevel
var x,y=x=Math.random();console.log(x/y,y/x,(x+y)/x);

If this optimization were to be implemented then the generated ES5 code at the end of this comment:

!function() {
  var e, t, r = {},
    n = {};
  t = n = function(e, t) {
    return t.get ? t.get.call(e) : t.value
  }, n.default = t, n.__esModule = true;
  var u, a = n,
    o = {};
  u = o = function(e, t, r) {
    if (!t.has(e)) throw new TypeError("attempted to " + r + " private field on non-instance");
    return t.get(e)
  }, o.default = u, o.__esModule = true;
  var l = o;
  e = r = function(e, t) {
    var r = l(e, t, "get");
    return a(e, r)
  }, r.default = e, r.__esModule = true
}();

could be completely optimized away with uglify-js -c merge_vars=false,passes=9 aided by #4858.

After pass 1 the unneeded assignments r = {}, n = {} and o = {} would be dropped by reduce_vars leaving all top level declared variables in the example as single assignment.

I suspect that merge_vars may be at odds with this proposed optimization since it would complicate the analysis if the aliased single assigned variables are repurposed in the example above. Or perhaps not.

[alexlamsl]

Taught collapse_vars to handle:

$ echo 'var x, y = x = Math.random(); console.log(x/y, y/x, (x+y)/x);' | uglify-js -bc toplevel
var x;

x = Math.random();

console.log(x / x, x / x, (x + x) / x);

But that still leave your larger example with:

$ uglify-js test.js -bc merge_vars=false,passes=9
!function() {
    function n(e, t) {
        return t.get ? t.get.call(e) : t.value;
    }
    function o(e, t, r) {
        if (!t.has(e)) throw new TypeError("attempted to " + r + " private field on non-instance");
        return t.get(e);
    }
    (n.default = n).__esModule = !0;
    var a = n;
    (o.default = o).__esModule = !0;
    var r, l = o;
    ((r = function(e, t) {
        var r = l(e, t, "get");
        return a(e, r);
    }).default = r).__esModule = !0;
}();

[kzc]

The impasse appears to be the remaining aliases a = n and l = o and the nested assignment of r. If the code is rewritten as the following, it optimizes nicely...

$ cat test.js
!function() {
    function n(e, t) {
        return t.get ? t.get.call(e) : t.value;
    }
    function o(e, t, r) {
        if (!t.has(e)) throw new TypeError("attempted to " + r + " private field on non-instance");
        return t.get(e);
    }
    (n.default = n).__esModule = !0;
    (o.default = o).__esModule = !0;
    var r = function(e, t) {
        var r = o(e, t, "get");
        return n(e, r);
    };
    (r.default = r).__esModule = !0;
}();

$ cat test.js | uglify-js -bc

$ 

Some optimizations prevent other optimizations from occurring. It is a difficult balancing act. It wouldn't help in this case, but there might be heuristics that could be employed such as: If passes > 1 then merge_vars could only run on the final pass, for example.

As far as alias dropping goes, if one variable mirrors another and neither have subsequent assignments then the original variable could be used for the alias. No doubt there will be complications in practise. The devil is in the details.

[alexlamsl]

Some optimizations prevent other optimizations from occurring. It is a difficult balancing act.

My plan is just to identify those and generalise the impeded option, e.g. reduce_vars can now handle multiple assign-read cycles within a single variable, so merge_vars won't interfere with its operations.

The impasse appears to be the remaining aliases a = n and l = o and the nested assignment of r. If the code is rewritten as the following, it optimizes nicely...

Thanks for the analysis − will take a closer look when I re-emerge from my alcove...

[kzc]

This seems to be an obstacle to further optimization...

Compare this code:

$ cat r1.js

    effect();
    var r;
    r = function(e, t) {
        var r = o(e, t, "get");
        return n(e, r);
    };
    (r.default = r).__esModule = !0;

$ cat r1.js | uglify-js -c toplevel,passes=9
var r;effect(),((r=function(e,r){return r=o(e,r,"get"),n(e,r)}).default=r).__esModule=!0;

to:

$ cat r2.js

    effect();
    var r = function(e, t) {
        var r = o(e, t, "get");
        return n(e, r);
    };
    (r.default = r).__esModule = !0;

$ cat r2.js | uglify-js -c toplevel
effect();