test_arg = ->
  run_test [

    # wrong
    #"arg(-1)",
    #"-pi",

    "arg(pi)",
    "0",

    "arg(1+i)",
    "1/4*pi",

    "arg(1-i)",
    "-1/4*pi",

    "arg(-1+i)",
    "3/4*pi",

    "arg(-1-i)",
    "-3/4*pi",

    "arg((-1)^(1/3))",
    "1/3*pi",

    "arg(1+exp(i*pi/3))",
    "1/6*pi",

    "arg((-1)^(1/6)*exp(i*pi/6))",
    "1/3*pi",

    # check when not assuming real variables ----------
    "assumeRealVariables = 0",
    "",

    "arg(a)",
    "arg(a)",

    # TODO this is wrong
    #"arg(a*exp(b+i*pi/5))",
    #"1/5*pi",

    # this is wrong
    #"arg(-1)",
    #"-pi",

    # this is also highly debatable
    # take the example
    # a = -1-i
    # then arg(a) - arg(-a) should give pi
    # but arg(1+i) - arg(-1-i) gives -pi instead
    # "arg(-a)",
    # "-pi+arg(a)",

    "assumeRealVariables = 1",
    "",
    # --------------------------------------------------

    # TODO this is wrong.
    #"arg(a*exp(b+i*pi/5))",
    #"1/5*pi",

    # referencing the test above, if
    # a is positive:
    "arg(abs(a)*exp(b+i*pi/5))",
    "1/5*pi",

    # otherwise, if negative, we get:
    "arg((-8)*exp(b+i*pi/5))",
    "-4/5*pi",

    # if a is positive, zero
    # if a is negative, -pi, so
    # we can't say much
    "arg(a)",
    "arg(a)",

    # this is also wrong, this should
    # be either zero or pi
    #"arg(-a)",
    #"-pi+arg(a)",

    "arg(-(-1)^(1/3))",
    "-2/3*pi",

    "arg(-exp(i*pi/3))",
    "-2/3*pi",

    "arg(-i)",
    "-1/2*pi",

    "arg((a+b*i)/(c+d*i))",
    "arctan(b/a)-arctan(d/c)",

    "arg(((-1)^(1/2) / (3^(1/2)))^(1/2))",
    "1/4*pi",

    "arg((-1)^(1/6))",
    "1/6*pi",


  ]