test_that("illegal initializations are rejected", {
  expect_silent(LogNormDistribution$new(1.0, 1.0))
  expect_error(LogNormDistribution$new("1", 1.0), class = "p1_not_numeric")
  expect_error(LogNormDistribution$new(1.0, "1"), class = "p2_not_numeric")
  expect_error(LogNormDistribution$new(1.0, 1.0, "LN8"),
               class = "parametrization_not_supported")
})

test_that("distribution name is correct", {
  ln <- LogNormDistribution$new(1.0, 1.0)
  expect_identical(ln$distribution(), "LN(1,1)")
  ln <- LogNormDistribution$new(1.0, 1.0, "LN2")
  expect_identical(ln$distribution(), "LN(1,1)")
})

test_that("mean, mode, sd and quantiles are returned correctly", {
  mu <- 0.0
  sigma <- 0.25
  ln <- LogNormDistribution$new(p1 = mu, p2 = sigma, "LN1")
  expect_intol(ln$mean(), exp(mu + (sigma ^ 2L) / 2L), 0.01)
  sd <- sqrt((exp(sigma ^ 2L) - 1L) * (exp(2L * mu + sigma ^ 2L)))
  expect_intol(ln$SD(), sd, 0.01)
  expect_intol(ln$mode(), exp(mu - sigma ^ 2L), 0.01)
  # quantiles
  expect_intol(ln$quantile(0.5), exp(mu), 0.01)
  erf.inv <- function(x) qnorm((x + 1L) / 2L) / sqrt(2L)
  ln.quant <- function(mu, sigma, p) {
    exp(mu + sqrt(2L * sigma ^ 2L) * erf.inv(2L * p - 1L))
  }
  expect_intol(ln$quantile(0.25), ln.quant(mu, sigma, 0.25), 0.01)
  expect_intol(ln$quantile(0.75), ln.quant(mu, sigma, 0.75), 0.01)
})

test_that("quantile function checks inputs and has correct output", {
  mu <- 0.0
  sigma <- 0.25
  ln <- LogNormDistribution$new(p1 = mu, p2 = sigma, "LN1")
  probs <- c(0.1, 0.2, 0.5)
  expect_silent(ln$quantile(probs))
  probs <- c(0.1, NA, 0.5)
  expect_error(ln$quantile(probs), class = "probs_not_defined")
  probs <- c(0.1, "boo", 0.5)
  expect_error(ln$quantile(probs), class = "probs_not_numeric")
  probs <- c(0.1, 0.4, 1.5)
  expect_error(ln$quantile(probs), class = "probs_out_of_range")
  probs <- c(0.1, 0.2, 0.5)
  expect_length(ln$quantile(probs), 3L)
})

test_that("random sampling is from a log normal distribution", {
  mu <- 0.0
  sigma <- 0.25
  n <- 1000L
  ln <- LogNormDistribution$new(mu, sigma, "LN1")
  # check mean
  ln$sample(TRUE)
  expect_identical(ln$r(), exp(mu + 0.5 * sigma ^ 2L))
  # check sampling
  samp <- vapply(seq_len(n), FUN.VALUE = 1.0, FUN = function(i) {
    ln$sample()
    rv <- ln$r()
    return(rv)
  })
  expect_length(samp, n)
  # expect sample mean and sd to fall within 99.9% CI; test
  # expected to fail 0.1% of the time, exclude from CRAN
  skip_on_cran()
  ht <- ks.test(samp, rlnorm(n, meanlog = mu, sdlog = sigma))
  expect_gt(ht$p.value, 0.001)
})

test_that("parametrizations are linked", {
  mu <- 0.0
  sigma <- 0.25
  # 1
  ln1 <- LogNormDistribution$new(mu, sigma, "LN1")
  # 2
  ln2 <- LogNormDistribution$new(mu, sigma ^ 2L, "LN2")
  expect_identical(ln1$mean(), ln2$mean())
  expect_identical(ln1$mode(), ln2$mode())
  # 3
  ln3 <- LogNormDistribution$new(exp(mu), sigma, "LN3")
  expect_intol(ln1$mean(), ln3$mean(), 0.01)
  expect_intol(ln1$mode(), ln3$mode(), 0.01)
  # 4
  ln4 <- LogNormDistribution$new(exp(mu), sqrt(exp(sigma ^ 2L) - 1L), "LN4")
  expect_intol(ln1$mean(), ln4$mean(), 0.01)
  expect_intol(ln1$mode(), ln4$mode(), 0.01)
  # 5
  ln5 <- LogNormDistribution$new(mu, 1L / sigma ^ 2L, "LN5")
  expect_intol(ln1$mean(), ln5$mean(), 0.01)
  expect_intol(ln1$mode(), ln5$mode(), 0.01)
  # 6
  ln6 <- LogNormDistribution$new(exp(mu), exp(sigma), "LN6")
  expect_intol(ln1$mean(), ln6$mean(), 0.01)
  expect_intol(ln1$mode(), ln6$mode(), 0.01)
  # 7
  ln7 <- LogNormDistribution$new(
    exp(mu + (sigma ^ 2L) / 2L),
    exp(mu + (sigma ^ 2L) / 2L) * sqrt(exp(sigma ^ 2L) - 1L), "LN7"
  )
  expect_intol(ln1$mean(), ln7$mean(), 0.01)
  expect_intol(ln1$mode(), ln7$mode(), 0.01)
})