Dead-code elimination

October 24, 2018

Imagine you have the code below:

function foo() {
   var a = foobar();
   var b = 8;
   return 10;
}

after passing this code through uglify-js, you will get the code below:

function foo(){
  foobar();
  return 10;
}

uglify-js will do dead code removal, ie: it will remove code that will not affect the program results. Since we defined a and b in the function foo and we are not using a and b within the foo, uglify-js safely removes it from the uglified code.

However, if you can see from the uglified code, foobar() is preserved. This is because calling foobar may have side effects on the program, so uglify-js preserves foobar(). But since we do not need the return value of foobar(), which is assigned to a, a got removed by uglify-js.


What if we know that foobar() will not caused any side effects?

How do we tell uglify-js that if we are not using the return value from foobar(), please remove it as well?


Turns out that uglify-js will treat a function call as โ€œpureโ€ if there is a comment annotation /*@__PURE__*/ or /*#__PURE__*/ immediately precedes the call.

For example: /* @__PURE__ */ foobar() !

// before uglify
function foo() {
  var a = /*@__PURE__*/foobar();
  var b = 8;
  return 10;
}

// after uglify
function foo(){return 10}

Note that now foobar() is being removed!


So, how is this useful?

As a library author, you can write a babel plugin to mark your function to be pure, so that uglify-js can drop the function call if the return value is not being used in the code. (There is a util from babel to do just the job of marking function as pure ๐Ÿ˜Ž)