enhance pure_getters, reduce_vars & unused

[alexlamsl]

No description provided.

[alexlamsl]

@kzc this takes care of #4858 (comment) & #4858 (comment)

Casting #4858 (comment) into the Void would also require upgrading collapse_vars which is WIP

[kzc]

Cool! Looking forward to seeing what ufuzz comes up with.

Hey, this is strange...

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


$ cat r2.js | uglify-js -bc toplevel | od -c
0000000    e   f   f   e   c   t   (   )   ;  \n  \n  \n                
0000014

Why the extra newlines for r2.js?

[kzc]

Top level AST_EmptyStatement...

$ cat r2.js | uglify-js -bc toplevel -o ast
{
  "_class": "AST_Toplevel",
  "globals": [
    "1000006 effect",
    "1000007 o",
    "1000008 n"
  ],
  "body": [
    {
      "_class": "AST_SimpleStatement",
      "body": {
        "_class": "AST_Call",
        "expression": {
          "_class": "AST_SymbolRef",
          "name": "effect",
          "thedef": "1000006 effect"
        },
        "args": []
      }
    },
    {
      "_class": "AST_EmptyStatement"
    }
  ],
  "enclosed": [
    "1000006 effect",
    "1000007 o",
    "1000008 n",
    "1000001 r"
  ],
  "variables": [
    "1000001 r"
  ]
}

[alexlamsl]

Something else these build tests are good for is testing pure_getters=true, which we can never do with ufuzz (not without getting an onslaught of false positives anyway)

[alexlamsl]

Hey, this is strange...

So that's because (r.default = r).__esModule = !0; can only be reduced to !0; due to side effects on LHS.

Won't affect output size except for beautify=true, and passes would certainly get rid of the dead meat easily as well so I guess we can leave it as is for now.