Makes mocks more tolerant to argument changes of deps

tests/testthat/helper-chronos.R

@@ -12,7 +12,7 @@ stub_chronos_loaders <- function(also_mock_predict_core = FALSE) {
   dir.create(fake_dir, recursive = TRUE, showWarnings = FALSE)
 
   bindings <- list(
-    chronos2_download = function(model_id, revision, cache_dir) {
+    chronos2_download = function(..., revision) {
       list(
         model_dir = fake_dir,
         sha = if (grepl("^[0-9a-f]{40}$", revision)) {
@@ -22,7 +22,7 @@ stub_chronos_loaders <- function(also_mock_predict_core = FALSE) {
         }
       )
     },
-    chronos2_parse_config = function(path) {
+    chronos2_parse_config = function(...) {
       list(
         d_model = 384L,
         num_layers = 12L,
@@ -32,10 +32,10 @@ stub_chronos_loaders <- function(also_mock_predict_core = FALSE) {
         quantiles = (1:9) / 10
       )
     },
-    chronos2_model = function(config) {
-      structure(list(config = config), class = "fake_chronos_module")
+    chronos2_model = function(...) {
+      structure(list(), class = "fake_chronos_module")
     },
-    load_chronos2_weights = function(model, path) invisible(NULL)
+    load_chronos2_weights = function(...) invisible(NULL)
   )
 
   if (also_mock_predict_core) {
@@ -46,9 +46,8 @@ stub_chronos_loaders <- function(also_mock_predict_core = FALSE) {
     bindings$chronos2_predict_core <- function(
       object,
       context,
-      prediction_length = NULL,
-      past_covariates = NULL,
-      future_covariates = NULL
+      ...,
+      prediction_length = NULL
     ) {
       n_series <- if (is.list(context)) length(context) else 1L
       n_q <- length(object$config$quantiles)

tests/testthat/helper-icl.R

@@ -15,40 +15,27 @@ stub_tabicl_loaders <- function() {
   bindings <- list(
     # `tabicl_cache_lookup()` normally locates a cached checkpoint and errors if
     # none is available; return a throwaway directory instead.
-    tabicl_cache_lookup = function(task, call = rlang::caller_env()) {
+    tabicl_cache_lookup = function(...) {
       fake_dir
     },
     # Only `config$max_classes` is consumed at fit time: `> 0` marks a
     # classification checkpoint, `<= 0` a regression one. `tabicl_bridge()`
     # passes `file.path(path, files$config)`, so branch on the file name.
-    tabicl_parse_config = function(path) {
+    tabicl_parse_config = function(path, ...) {
       list(
         max_classes = if (grepl("classification", path)) 10L else 0L
       )
     },
     # `tabicl_load_model()` reads the safetensors weights; the fake object it
     # returns is never inspected because the inference cores below are mocked.
-    tabicl_load_model = function(
-      model_dir,
-      task,
-      device = "cpu",
-      verbose = FALSE
-    ) {
-      structure(list(task = task), class = "fake_tabicl_module")
+    tabicl_load_model = function(...) {
+      structure(list(), class = "fake_tabicl_module")
     },
     # Deterministic class probabilities: one row per test case, one column per
     # class, rows summing to 1. The arg-max is rotated across rows so `class`
     # predictions are non-constant (guards against the column extraction
     # collapsing to a single level).
-    tabicl_classifier_proba = function(
-      loaded,
-      x_train,
-      y_train,
-      x_test,
-      members,
-      temperature = 0.9,
-      device = "cpu"
-    ) {
+    tabicl_classifier_proba = function(loaded, x_train, y_train, x_test, ...) {
       n <- nrow(x_test)
       n_classes <- length(unique(y_train))
       proba <- matrix(0, nrow = n, ncol = n_classes)
@@ -59,16 +46,10 @@ stub_tabicl_loaders <- function() {
       }
       proba
     },
-    # Deterministic, non-constant regression predictions.
-    tabicl_regressor_mean = function(
-      loaded,
-      x_train,
-      y_train,
-      x_test,
-      members,
-      device = "cpu"
-    ) {
-      seq_len(nrow(x_test)) + mean(y_train)
+    # Deterministic, non-constant regression predictions. Only the `mean` output
+    # is stubbed, which covers `type = "numeric"`.
+    tabicl_regressor_stats = function(loaded, x_train, y_train, x_test, ...) {
+      list(mean = seq_len(nrow(x_test)) + mean(y_train))
     }
   )