[lukaszsamson]

Found with SpecLint on elixir 1.20.4

lib/complex.ex:692: warning: Complex.abs_squared/1: inferred return includes :infinity or :nan, which the spec does not declare
    @spec abs_squared(t() | number() | non_finite_number()) :: number()
    inferred: (%Complex{re: :nan, im: :nan} or :nan) -> :nan
    inferred: (%Complex{} or :infinity or :neg_infinity) -> :infinity
    inferred: (%Complex{re: float() or integer(), im: float() or integer()} or float() or integer()) -> dynamic(float() or integer())
lib/complex.ex:1623: warning: Complex.csc/1: inferred return includes :infinity or :nan or :neg_infinity or
  (%Complex{
     re: :infinity or :nan or :neg_infinity or float(),
     im: :infinity or :nan or :neg_infinity or float()
   } and not %Complex{re: float() or integer(), im: float() or integer()}) or float() or integer(), which the spec does not declare
    @spec csc(t()) :: t()
    @spec csc(number()) :: number()
    inferred: (%Complex{im: :nan} or %Complex{re: :nan} or
  %Complex{re: float() or integer(), im: :infinity or :nan or :neg_infinity} or
  %Complex{
    re: :infinity or :nan or :neg_infinity,
    im: :infinity or :nan or :neg_infinity or float() or integer()
  } or :infinity or :nan or :neg_infinity or
  (%Complex{im: not :nan} and
     not (%Complex{re: :nan} or
            %Complex{re: :infinity or :nan or :neg_infinity, im: float() or integer()} or
            %Complex{re: float() or integer(), im: :infinity or :nan or :neg_infinity})) or
  (%Complex{re: :infinity or :nan or :neg_infinity, im: :infinity or :nan or :neg_infinity} and
     not (%Complex{im: :nan} or %Complex{re: :nan})) or float() or integer()) -> dynamic(
  %Complex{
    re: :infinity or :nan or :neg_infinity or float(),
    im: :infinity or :nan or :neg_infinity or float()
  } or :infinity or :nan or :neg_infinity or float() or integer()
)
lib/complex.ex:350: warning: Complex.to_polar/1: inferred clause (float() or integer()) -> {float() or integer(), integer()} has a return disjoint from the spec return {:infinity or :nan or :neg_infinity or float(), :infinity or :nan or :neg_infinity or float()}
    @spec to_polar(t() | number() | non_finite_number()) ::
  {float() | non_finite_number(), float() | non_finite_number()}
    inferred: (%Complex{}) -> dynamic({:infinity or :nan or float() or integer(), term()})
    inferred: (not %Complex{}) -> dynamic({not %Complex{}, term()})
    inferred: (float() or integer()) -> dynamic({float() or integer(), integer()})
    inferred: (:infinity) -> {:infinity, integer()}
    inferred: (:neg_infinity) -> dynamic({:infinity, term()})
    inferred: (:nan) -> {:nan, :nan}

The functions return non-finite values and integers: Complex.abs_squared(:nan) #=> :nan, Complex.csc(0) #=> :infinity, Complex.to_polar(3) #=> {3, 0}. The specs now say so, reusing non_finite_number(). Not changed: Complex.csc(Complex.new(0, 0)) returns a %Complex{} with NaN parts, which t() (number parts) does not cover; I left t() alone.