Regenerator relies on Function#name, breaks minifiers

[monsanto]

See here.

The author of this code tries to be clever and set GF.name in the line above, but this actually isn't possible to do (try it in Chrome or Firefox, it will just ignore the assignment).

Having "GeneratorFunction" as the constructor name is mandated by the specification. However, I would argue it's not a very important part of the specification, and that convenient minification is much more important. At the very least, the throw line should be removed, so people who don't care what the constructor name is aren't affected. It's worth noting that Regenerator itself does not care what the constructor name is.

However, there is a case for removing the "GeneratorFunction" trick altogether: it would allow for removing the mark() call in the following, improving the readability of the generated code:

root@smogon-dev:~$ echo 'function* foo() { }' | 6to5
"use strict";

var foo = regeneratorRuntime.mark(function foo() {
  return regeneratorRuntime.wrap(function foo$(_context) {
    while (true) switch (_context.prev = _context.next) {
      case 0:
      case "end": return _context.stop();
    }
  }, foo, this);
});

Here's the issue which lead to mark being added: facebook/regenerator#48. The primary reason seems to be allowing the co library to perform some magic to distinguish between regular functions and generator functions. I guess I'm not swayed by this because co isn't very useful when you have async functions (like we do).

[monsanto]

Also see mishoo/UglifyJS#552

[sebmck]

I've done a possible fix. I'm going to spend some time rewriting the regenerator runtime and simplifying it. It'll have to be released with 2.0.0 though due to breaking changes and incompatible API.

[jlongster]

@sebmck I'm glad you fixed it, but how are you going to stay up-to-date with regenerator's runtime? It sounds like you want to completely fork it. Would be nice to see if we could merge changes upstream.

[sebmck]

@jlongster We wouldn't. Maintenance of the regenerator runtime is minimal compared to everything else.

[sebmck]

Thanks! Released as of 1.15.0.

[benjamn]

Please don't spend any time rewriting the regenerator runtime without looping me in!

I'm sincerely confused why no one CC'd me here. You didn't even want to know why that assertion exists? Come on, people: facebook/regenerator#156

[jamiebuilds]

@benjamn I'm sorry you weren't CC'd. 6to5 moves at a really fast pace than other tools so this kind of thing becomes somewhat necessary. However, it's a goal to bring bug fixes and features back into the original libraries.

[monsanto]

@benjamn We know why the assertion exists, in fact I discuss it a bit in the first post. I personally don't care if the constructor is the correct name and don't want an error raised if it isn't. I feel easy minification is much, much more important than spec compliance in this corner case.

Note that if you DO care what the constructor name is, you can configure your preferred minifier to not mangle that name. We can even put a note in the documentation.

[aaronshaf]

In 6to5 2.13.3 I am still getting GeneratorFunction renamed? when using alongside uglifyjs.

[sebmck]

@aaronshaf Where are you getting the polyfill/runtime from? regenerator/runtime.js doesn't include the GeneratorFunction renamed? error anymore.