Please use brackets, not a semicolon, for empty branches.

I fixed that.
8dd203e

Would using return values on the set-expression-related methods, rather than a flag, be a workable way to track this? Looking at the code around this, tracking this value with a property seems verbose and difficult to see through.

I will try to make that change. It will take some time, so please wait.

I fixed that. d09dc4e

What is the idea behind this block? Does regexp_nonEmptyClassRanges as called through regexp_classContents not consume all the set content? What does the comment that says "Unreachable since ..." mean?

This process is to issue the same "Invalid character in character class" message as V8.
Without this process, broken expressions such as [A&&] will display the message "Unterminated character class".

Acorn as a whole does not try to mimic the precise error messages from browsers. I think it would be preferable to have less code and complexity (also here) rather than V8-aligned error messages.

Thanks for your comment. I will remove those processes that have become too complicated.

826f6f7

Here too the side effect of eat seems to 'leak' when the condition is false.

I didn't quite understand. What does leak mean? I thought we could get that status back with state.pos = start at L1060. Could you tell me what else I need to do?

My bad—I forgot to delete this comment after spotting the pos reset. Still not a fan of this pattern though, see my other comment.

I think I understand that you don't like state.pos = start. But for this part, I don't think it makes sense to do it any other way.
The way I think without eat is to do:

pp.regexp_eatClassSetRange = function(state) {
  const start = state.pos
  if (this.regexp_eatClassSetCharacter(state)) {
    const left = state.lastIntValue
    if (state.current() === 0x2D /* - */) {
      state.advance()
      if (this.regexp_eatClassSetCharacter(state)) {
        const right = state.lastIntValue
        if (left !== -1 && right !== -1 && left > right) {
          state.raise("Range out of order in character class")
        }
        return true
      }
    }
    state.pos = start
  }
  return false
}

state.current()+state.advance() is consequently the same as eat(), and state.pos = start cannot be removed yet. If you know a better way, could you please advise?

Could we maybe add some lookahead or multi-character eat mechanism to RegExpValidationState? This repeated side-effecting with eat only to restore the position after seems both hard to read and error-prone.

I added a function that eats multiple characters and changed it to use it.

0659703