Bug Description

Empty character classes like [^\S\s] (which logically match nothing) incorrectly match empty strings instead of failing.

Reproduction

package main

import (
    "fmt"
    "regexp"
    "github.com/coregx/coregex"
)

func main() {
    pattern := `[^\S\s]`
    input := "abc"

    // stdlib - correct (no match)
    reStd := regexp.MustCompile(pattern)
    fmt.Println("stdlib:", reStd.MatchString(input)) // false

    // coregex - incorrect (matches!)
    reCg := coregex.MustCompile(pattern)
    fmt.Println("coregex:", reCg.MatchString(input)) // true
}

Root Cause

In nfa/compile.go:365-367, when compileCharClass receives an empty rune slice, it calls compileEmptyMatch():

if len(ranges) == 0 {
    return c.compileEmptyMatch()  // WRONG!
}

compileEmptyMatch() creates an epsilon transition that matches empty string. But an empty character class should never match - it's an impossible condition.

Semantics

• [\S\s] = any character = OpAnyChar
• [^\S\s] = NOT (any character) = empty set = matches nothing
• Go's parser correctly sets Rune: [] for empty classes

Proposed Fix

Add compileNoMatch() function that creates an NFA fragment with no path from start to end, making it impossible to match. Use this for empty character classes instead of compileEmptyMatch().