Ensure default case folding only happens when default is last

[jridgewell]

I failed to consider if multiple branches could evaluate to the test condition. Eg,

switch (true) {
    case definitely_true:
    default:
      return 'definitely';
    case maybe_true:
      return 'maybe';
}

Here, both definitely_true and maybe_true can be === true. And we can't eliminate the case definitely_true:, because it prevents us from falling into the maybe_true branch (when it is true).

Fixes #1083

[elad-yosifon]

I think this code is valuable, we only need to limit it to a more specific use case.
whenever there is a switch with an expressions that does not need to be evaluated and are not true (false should be completely folded), and all the case expressions are side effects free.

switch(false){  // <--- should be removed entirely
  //...
}

switch(true){  // <--- optimization should bail
  //...
}

switch(var1++){ // <-- optimization should bail
  case (B + 1):
  // ...
}

switch(var1){
  case (B + 1): // <-- fine
  case (B++): // <-- optimization should bail
  // ...
}

@jridgewell plz LMK if you have other reservations...

[jridgewell]

(false should be completely folded),

I don't think that's correct. Any switch statement which has cases that aren't mutually exclusive can cause this issue, and the only way we could know otherwise is if every case expression was statically evaluable. I don't think there is a way to perform reordering, we can only fold equivalent cases that appear directly next to each other.

[elad-yosifon]

@jridgewell plz ignore my last reply.. was probably out of focus or something.. total nonsense.

If everything can be statically evaluate, I think we should apply the reordering optimization. It is actually very beneficial for one of my company products., and will probably help other projects as well - especially with Webpack and DCE using switch statements.

[jridgewell]

It is actually very beneficial for one of my company products., and will probably help other projects as well

If you want to open a PR that only performs the reordering when each case expression is static, I'd be happy to integrate that instead of reverting the original.

[elad-yosifon]

Sure.. that was my plan for this weekend. 🙏

[jridgewell]

Updated this PR.

[fabiosantoscode]

Great stuff, thanks!