# tests/testthat/test-diagnostics.R

# ---- fixtures -------------------------------------------------------------
make_dat <- function(n = 40, seed = 99) {
  set.seed(seed)
  x <- seq(0, 180, length.out = n)
  y <- 3 + 0.03 * x - 0.00008 * x^2 + rnorm(n, 0, 0.2)
  data.frame(nitrogen = x, yield = y)
}

dat <- make_dat()

# ---- gof_stats ------------------------------------------------------------
test_that("gof_stats returns named list with all keys", {
  m   <- fit_linear(dat, "yield ~ nitrogen")
  gf  <- gof_stats(m)
  expect_named(gf, c("R2", "adj_R2", "RMSE", "MAE", "AIC", "BIC"))
})

test_that("gof_stats R2 is in (0, 1) for a sensible model", {
  m  <- fit_linear(dat, "yield ~ nitrogen")
  gf <- gof_stats(m)
  expect_gt(gf$R2, 0)
  expect_lt(gf$R2, 1)
})

test_that("gof_stats RMSE is positive", {
  m  <- fit_linear(dat, "yield ~ nitrogen")
  gf <- gof_stats(m)
  expect_gt(gf$RMSE, 0)
})

test_that("gof_stats works for agriNLS", {
  m  <- fit_nonlinear(dat, "nitrogen", "yield", "quadratic")
  gf <- gof_stats(m)
  expect_named(gf, c("R2", "adj_R2", "RMSE", "MAE", "AIC", "BIC"))
})

test_that("gof_stats works on raw lm object", {
  m  <- lm(yield ~ nitrogen, data = dat)
  gf <- gof_stats(m)
  expect_named(gf, c("R2", "adj_R2", "RMSE", "MAE", "AIC", "BIC"))
})

# ---- compare_models -------------------------------------------------------
test_that("compare_models returns a data frame", {
  m1  <- fit_linear(dat, "yield ~ nitrogen")
  m2  <- fit_nonlinear(dat, "nitrogen", "yield", "quadratic")
  cmp <- compare_models(linear = m1, quadratic = m2)
  expect_s3_class(cmp, "data.frame")
})

test_that("compare_models has expected columns", {
  m1  <- fit_linear(dat, "yield ~ nitrogen")
  m2  <- fit_linear(dat, "yield ~ nitrogen + I(nitrogen^2)")
  cmp <- compare_models(m1 = m1, m2 = m2)
  expect_true(all(c("model","AIC","BIC","RMSE","delta_AIC") %in% names(cmp)))
})

test_that("compare_models ranks best model first (lowest AIC)", {
  m1  <- fit_linear(dat, "yield ~ 1")          # intercept only — poor
  m2  <- fit_linear(dat, "yield ~ nitrogen")    # better
  cmp <- compare_models(intercept = m1, linear = m2)
  expect_equal(cmp$model[1], "linear")
})

test_that("compare_models works with unnamed args", {
  m1  <- fit_linear(dat, "yield ~ nitrogen")
  m2  <- fit_linear(dat, "yield ~ nitrogen + I(nitrogen^2)")
  cmp <- compare_models(m1, m2)
  expect_equal(nrow(cmp), 2)
  expect_true(all(grepl("^model", cmp$model)))
})

test_that("delta_AIC of best model is 0", {
  m1  <- fit_linear(dat, "yield ~ 1")
  m2  <- fit_linear(dat, "yield ~ nitrogen")
  cmp <- compare_models(int = m1, lin = m2)
  expect_equal(cmp$delta_AIC[1], 0)
})

# ---- residual_check -------------------------------------------------------
test_that("residual_check returns list of 4 ggplot objects", {
  skip_if_not_installed("ggplot2")
  m   <- fit_linear(dat, "yield ~ nitrogen")
  plt <- residual_check(m)
  expect_type(plt, "list")
  expect_length(plt, 4)
  for (p in plt) expect_s3_class(p, "gg")
})

# ---- model_summary --------------------------------------------------------
test_that("model_summary.agriLM prints without error", {
  m <- fit_linear(dat, "yield ~ nitrogen")
  expect_output(model_summary(m), "agriReg")
})

test_that("model_summary.agriNLS prints without error", {
  m <- fit_nonlinear(dat, "nitrogen", "yield", "quadratic")
  expect_output(model_summary(m), "agriReg")
})

# ---- plot_fit -------------------------------------------------------------
test_that("plot_fit dispatches to correct plot method", {
  skip_if_not_installed("ggplot2")
  m <- fit_linear(dat, "yield ~ nitrogen")
  p <- plot_fit(m)
  expect_s3_class(p, "gg")
})