import caffeine_lang/errors.{type CompilationError}
import caffeine_query_language/ast.{
  type Comparator, type CqlParsed, type Exp, type Operator, type TimeSliceExp,
  Add, Div, GreaterThan, GreaterThanOrEqualTo, LessThan, LessThanOrEqualTo, Mul,
  OperatorExpr, Primary, PrimaryExp, PrimaryWord, Sub, TimeSliceExp,
  TimeSliceExpr, Word,
}
import gleam/bool
import gleam/float
import gleam/option.{type Option}
import gleam/result
import gleam/string

/// Codeunit (UTF-16 on JS, UTF-8 byte on Erlang) at index, or -1 if out of bounds.
/// Used in per-position scans where we look for ASCII tokens — bypassing
/// gleam_stdlib's grapheme-safe slicing avoids an O(N) Intl.Segmenter setup
/// per char (which turns the scans into O(N^2)).
@external(erlang, "parser_ffi", "code_unit_at")
@external(javascript, "./parser_ffi.mjs", "code_unit_at")
fn code_unit_at(s: String, i: Int) -> Int

/// True if `haystack[pos..pos+|needle|]` exactly equals `needle` (codeunit-wise).
@external(erlang, "parser_ffi", "substring_equals_at")
@external(javascript, "./parser_ffi.mjs", "substring_equals_at")
fn substring_equals_at(haystack: String, pos: Int, needle: String) -> Bool

/// Codeunit count — the indexing space used by code_unit_at / substring_equals_at /
/// slice_codeunits. NOT grapheme count (which `string.length` returns and which
/// also walks the string via Intl.Segmenter on JS).
@external(erlang, "parser_ffi", "code_unit_length")
@external(javascript, "./parser_ffi.mjs", "code_unit_length")
fn code_unit_length(s: String) -> Int

/// Slice by codeunit range. Mirrors `string.slice` but skips the grapheme walk.
@external(erlang, "parser_ffi", "slice_codeunits")
@external(javascript, "./parser_ffi.mjs", "slice_codeunits")
fn slice_codeunits(s: String, start: Int, len: Int) -> String

// ASCII codeunits for parens used in count_parens.
const open_paren: Int = 0x28

const close_paren: Int = 0x29

/// Parses a CQL expression string into an Exp AST node.
/// Returns an error if the input cannot be parsed.
@internal
pub fn parse_expr(input: String) -> Result(Exp(CqlParsed), String) {
  let trimmed = string.trim(input)

  case is_fully_parenthesized(trimmed) {
    True -> {
      let inner = string.slice(trimmed, 1, string.length(trimmed) - 2)
      use inner_exp <- result.try(parse_expr(inner))
      Ok(Primary(PrimaryExp(inner_exp)))
    }
    False -> {
      let operators = [#("+", Add), #("-", Sub), #("*", Mul), #("/", Div)]
      try_operators(trimmed, operators)
    }
  }
}

fn is_fully_parenthesized(input: String) -> Bool {
  string.starts_with(input, "(")
  && string.ends_with(input, ")")
  && { string.length(input) >= 2 && is_balanced_parens(input, 1, 1) }
}

fn try_operators(
  input: String,
  operators: List(#(String, Operator)),
) -> Result(Exp(CqlParsed), String) {
  case operators {
    [] -> {
      case try_parse_keyword_expr(input) {
        Ok(exp) -> Ok(exp)
        Error(err) -> {
          use <- bool.guard(
            when: string.starts_with(input, "time_slice(")
              && string.ends_with(input, ")"),
            return: Error(err),
          )
          let word = Word(input)
          Ok(Primary(PrimaryWord(word)))
        }
      }
    }
    [#(op_str, op), ..rest] -> {
      case find_operator(input, op_str) {
        Ok(#(left, right)) -> {
          use left_exp <- result.try(parse_expr(left))
          use right_exp <- result.try(parse_expr(right))
          Ok(OperatorExpr(left_exp, right_exp, op))
        }
        Error(_) -> try_operators(input, rest)
      }
    }
  }
}

/// Attempts to parse a keyword expression like "time_slice(...)".
/// Returns Error if the input is not a keyword expression.
fn try_parse_keyword_expr(input: String) -> Result(Exp(CqlParsed), String) {
  use <- bool.guard(
    when: !{
      string.starts_with(input, "time_slice(") && string.ends_with(input, ")")
    },
    return: Error("Not a keyword expression"),
  )
  let prefix_len = string.length("time_slice(")
  let inner_len = string.length(input) - prefix_len - 1
  let inner = string.slice(input, prefix_len, inner_len)
  use spec <- result.try(parse_time_slice_spec(inner))
  Ok(TimeSliceExpr(spec))
}

/// Parses the inner content of a time_slice expression.
/// Format: "<query> <comparator> <threshold> per <interval>"
/// Example: "avg:system.cpu > 80 per 300s"
fn parse_time_slice_spec(input: String) -> Result(TimeSliceExp, String) {
  let trimmed = string.trim(input)

  case trimmed {
    "" -> Error("Empty time_slice expression")
    _ -> {
      use #(query, comparator, rest) <- result.try(find_comparator(trimmed))

      let query_trimmed = string.trim(query)
      case query_trimmed {
        "" -> Error("Missing query in time_slice expression")
        _ -> {
          use #(threshold_str, interval_str) <- result.try(split_on_per(rest))
          use threshold <- result.try(parse_threshold(threshold_str))
          use interval_seconds <- result.try(parse_interval(interval_str))

          Ok(TimeSliceExp(
            query: query_trimmed,
            comparator: comparator,
            threshold: threshold,
            interval_seconds: interval_seconds,
          ))
        }
      }
    }
  }
}

/// Finds a comparator in the input and splits into (query, comparator, rest).
fn find_comparator(
  input: String,
) -> Result(#(String, Comparator, String), String) {
  let comparators = [
    #(">=", GreaterThanOrEqualTo),
    #("<=", LessThanOrEqualTo),
    #(">", GreaterThan),
    #("<", LessThan),
  ]

  find_comparator_loop(input, comparators)
}

fn find_comparator_loop(
  input: String,
  comparators: List(#(String, Comparator)),
) -> Result(#(String, Comparator, String), String) {
  case comparators {
    [] -> Error("No comparator found in time_slice expression")
    [#(comp_str, comp), ..rest] -> {
      case find_substring_position(input, comp_str) {
        option.Some(pos) -> {
          let query = slice_codeunits(input, 0, pos)
          let rest_start = pos + code_unit_length(comp_str)
          let rest_len = code_unit_length(input) - rest_start
          let rest_str = slice_codeunits(input, rest_start, rest_len)
          Ok(#(query, comp, rest_str))
        }
        option.None -> find_comparator_loop(input, rest)
      }
    }
  }
}

/// Finds the position of a substring in a string. Returns a codeunit position.
fn find_substring_position(haystack: String, needle: String) -> Option(Int) {
  find_substring_position_loop(
    haystack,
    needle,
    0,
    code_unit_length(needle),
    code_unit_length(haystack),
  )
}

fn find_substring_position_loop(
  haystack: String,
  needle: String,
  pos: Int,
  needle_len: Int,
  haystack_len: Int,
) -> Option(Int) {
  use <- bool.guard(when: pos + needle_len > haystack_len, return: option.None)
  use <- bool.guard(
    when: substring_equals_at(haystack, pos, needle),
    return: option.Some(pos),
  )
  find_substring_position_loop(
    haystack,
    needle,
    pos + 1,
    needle_len,
    haystack_len,
  )
}

/// Splits on "per" keyword, returning (threshold_str, interval_str).
fn split_on_per(input: String) -> Result(#(String, String), String) {
  case find_substring_position(input, "per") {
    option.Some(pos) -> {
      let threshold_str = string.trim(slice_codeunits(input, 0, pos))
      let rest_start = pos + 3
      let rest_len = code_unit_length(input) - rest_start
      let interval_str =
        string.trim(slice_codeunits(input, rest_start, rest_len))
      Ok(#(threshold_str, interval_str))
    }
    option.None -> Error("Missing 'per' keyword in time_slice expression")
  }
}

/// Parses a threshold value as a float.
fn parse_threshold(input: String) -> Result(Float, String) {
  let trimmed = string.trim(input)
  case trimmed {
    "" -> Error("Missing threshold in time_slice expression")
    _ ->
      case float.parse(trimmed) {
        Ok(f) -> Ok(f)
        Error(_) ->
          case parse_int_as_float(trimmed) {
            Ok(f) -> Ok(f)
            Error(_) ->
              Error(
                "Invalid threshold '" <> trimmed <> "' in time_slice expression",
              )
          }
      }
  }
}

/// Parses an integer string as a float.
fn parse_int_as_float(input: String) -> Result(Float, String) {
  use <- bool.guard(
    when: string.contains(input, "."),
    return: Error("Not an integer"),
  )
  float.parse(input <> ".0")
  |> result.map_error(fn(_) { "Invalid number" })
}

/// Parses an interval like "10s", "5m", "1h", "500ms", "1d" into seconds.
fn parse_interval(input: String) -> Result(Float, String) {
  let trimmed = string.trim(input)
  case trimmed {
    "" -> Error("Missing interval in time_slice expression")
    _ -> {
      let len = string.length(trimmed)
      // Check for the 2-char "ms" unit before falling back to 1-char units.
      let #(unit, number_part) = case
        len >= 3 && string.slice(trimmed, len - 2, 2) == "ms"
      {
        True -> #("ms", string.slice(trimmed, 0, len - 2))
        False -> #(
          string.slice(trimmed, len - 1, 1),
          string.slice(trimmed, 0, len - 1),
        )
      }

      use multiplier <- result.try(case unit {
        "ms" -> Ok(0.001)
        "s" -> Ok(1.0)
        "m" -> Ok(60.0)
        "h" -> Ok(3600.0)
        "d" -> Ok(86_400.0)
        _ ->
          Error(
            "Invalid interval unit '"
            <> unit
            <> "' (expected ms, s, m, h, or d)",
          )
      })

      use number <- result.try(case float.parse(number_part) {
        Ok(f) -> Ok(f)
        Error(_) ->
          case parse_int_as_float(number_part) {
            Ok(f) -> Ok(f)
            Error(_) -> Error("Invalid interval number '" <> number_part <> "'")
          }
      })

      Ok(number *. multiplier)
    }
  }
}

fn find_operator(
  input: String,
  operator: String,
) -> Result(#(String, String), CompilationError) {
  find_rightmost_operator_at_level(input, operator, 0, 0, -1)
}

/// Checks if parentheses are balanced in the input string starting from a position.
/// Used to validate parenthesized expressions during parsing.
@internal
pub fn is_balanced_parens(input: String, pos: Int, count: Int) -> Bool {
  is_balanced_parens_loop(input, pos, count, code_unit_length(input))
}

/// Internal loop with pre-computed input length.
fn is_balanced_parens_loop(
  input: String,
  pos: Int,
  count: Int,
  input_len: Int,
) -> Bool {
  use <- bool.guard(when: pos >= input_len, return: count == 0)
  let new_count = count_parens(count, input, pos)
  let does_not_close_too_early = !{ { new_count == 0 } && pos != input_len - 1 }

  does_not_close_too_early
  && is_balanced_parens_loop(input, pos + 1, new_count, input_len)
}

/// Finds the rightmost occurrence of an operator at parenthesis level 0.
/// Returns the left and right parts of the expression split at the operator.
@internal
pub fn find_rightmost_operator_at_level(
  input: String,
  operator: String,
  start_pos: Int,
  paren_level: Int,
  rightmost_pos: Int,
) -> Result(#(String, String), CompilationError) {
  find_rightmost_operator_at_level_loop(
    input,
    operator,
    start_pos,
    paren_level,
    rightmost_pos,
    code_unit_length(operator),
    code_unit_length(input),
  )
}

/// Internal loop with pre-computed lengths.
fn find_rightmost_operator_at_level_loop(
  input: String,
  operator: String,
  start_pos: Int,
  paren_level: Int,
  rightmost_pos: Int,
  operator_length: Int,
  input_len: Int,
) -> Result(#(String, String), CompilationError) {
  case start_pos >= input_len {
    True ->
      case rightmost_pos {
        -1 -> Error(errors.cql_parser_error(msg: "Operator not found"))
        pos -> {
          let left = string.trim(slice_codeunits(input, 0, pos))
          let right_start = pos + operator_length
          let right_length = input_len - right_start
          let right =
            string.trim(slice_codeunits(input, right_start, right_length))
          Ok(#(left, right))
        }
      }
    False -> {
      let new_paren_level = count_parens(paren_level, input, start_pos)
      let new_rightmost_pos = case
        new_paren_level == 0 && substring_equals_at(input, start_pos, operator)
      {
        True -> start_pos
        False -> rightmost_pos
      }

      find_rightmost_operator_at_level_loop(
        input,
        operator,
        start_pos + 1,
        new_paren_level,
        new_rightmost_pos,
        operator_length,
        input_len,
      )
    }
  }
}

fn count_parens(cur_count: Int, input: String, pos: Int) -> Int {
  case code_unit_at(input, pos) {
    c if c == open_paren -> cur_count + 1
    c if c == close_paren -> cur_count - 1
    _ -> cur_count
  }
}