exclude dead branches

[Rich-Harris]

Follow-up to #208 (comment). This PR excludes dead branches, wherever it's possible to determine that through static analysis.

This goes slightly further than what e.g. UglifyJS does:

var obj = {};
obj.foo = function () {
  console.log( 'this function should be excluded' );
};

if ( false ) obj.foo();

Rollup will exclude obj.foo but UglifyJS won't, as far as I can tell (there may be an option that I don't know about).

Alongside https://github.com/rollup/rollup-plugin-replace, this should make it easier to make different bundles for different targets (e.g. browser vs node, dev vs prod).

[Victorystick]

Any reason to use != here while you use === above?

[Rich-Harris]

yeah, it's to do with strict vs non-strict equality. though now that you mention it, i think it'd be better to have separate nodesAreStrictEqual and nodesAreNotStrictEqual functions

[Victorystick]

What I mean is, what if an identifier has the value NaN?

The assumption a == a -> true is invalid.

Edit: Looks very dry and nice otherwise.

[Rich-Harris]

What I mean is, what if an identifier has the value NaN?

Ah. Good point. Bloody NaN! Is there any more useless feature in any programming language?

So do we skip identifier equality checks altogether? I guess the only alternative is to have an option controlling whether or not to care about NaN, which is territory I'd rather not get into

[Victorystick]

I fear the best thing to do is to skip-identifiers. It's unlikely to end up with NaNs like that, but it could happen. Unless we can be sure of the identifiers' values.

Say our Declarations had a value property, then we could compare actual values. const variables would be easy. It'd be a best-effort thing regardless.

[Rich-Harris]

agreed

[Victorystick]

@Rich-Harris Mind if I go ahead and merge this?

[Rich-Harris]

Yep let's do it. We should probably merge #253 too and release it as 0.21 (so that people can pin 0.20 if it turns out the algorithm fails in certain scenarios)