test: replacing tmpdir with tmp_path in tests_pytorch/loops (#1…

…9646)

refactored tmpdir from the tets_pytorch/loops dir

tests/tests_pytorch/loops/optimization/test_closure.py

@@ -18,14 +18,14 @@
 from lightning.pytorch.utilities.exceptions import MisconfigurationException
 
 
-def test_optimizer_step_no_closure_raises(tmpdir):
+def test_optimizer_step_no_closure_raises(tmp_path):
     class TestModel(BoringModel):
         def optimizer_step(self, epoch=None, batch_idx=None, optimizer=None, optimizer_closure=None, **_):
             # does not call `optimizer_closure()`
             pass
 
     model = TestModel()
-    trainer = Trainer(default_root_dir=tmpdir, fast_dev_run=1)
+    trainer = Trainer(default_root_dir=tmp_path, fast_dev_run=1)
     with pytest.raises(MisconfigurationException, match="The closure hasn't been executed"):
         trainer.fit(model)
 
@@ -39,12 +39,12 @@ def step(self, closure=None):
             return BrokenSGD(self.layer.parameters(), lr=0.1)
 
     model = TestModel()
-    trainer = Trainer(default_root_dir=tmpdir, fast_dev_run=1)
+    trainer = Trainer(default_root_dir=tmp_path, fast_dev_run=1)
     with pytest.raises(MisconfigurationException, match="The closure hasn't been executed"):
         trainer.fit(model)
 
 
-def test_closure_with_no_grad_optimizer(tmpdir):
+def test_closure_with_no_grad_optimizer(tmp_path):
     """Test that the closure is guaranteed to run with grad enabled.
 
     There are certain third-party library optimizers
@@ -67,6 +67,6 @@ def training_step(self, batch, batch_idx):
         def configure_optimizers(self):
             return NoGradAdamW(self.parameters(), lr=0.1)
 
-    trainer = Trainer(default_root_dir=tmpdir, fast_dev_run=1)
+    trainer = Trainer(default_root_dir=tmp_path, fast_dev_run=1)
     model = TestModel()
     trainer.fit(model)

tests/tests_pytorch/loops/optimization/test_manual_loop.py

@@ -28,7 +28,7 @@ def test_manual_result():
     assert result.extra == asdict
 
 
-def test_warning_invalid_trainstep_output(tmpdir):
+def test_warning_invalid_trainstep_output(tmp_path):
     class InvalidTrainStepModel(BoringModel):
         def __init__(self):
             super().__init__()
@@ -38,7 +38,7 @@ def training_step(self, batch, batch_idx):
             return 5
 
     model = InvalidTrainStepModel()
-    trainer = Trainer(default_root_dir=tmpdir, fast_dev_run=1)
+    trainer = Trainer(default_root_dir=tmp_path, fast_dev_run=1)
 
     with pytest.raises(MisconfigurationException, match="return a Tensor or have no return"):
         trainer.fit(model)

tests/tests_pytorch/loops/optimization/test_optimizer_loop.py

@@ -70,15 +70,15 @@ def __getitem__(self, key: str) -> T:
         (OutputMapping({"a": 5}), "the 'loss' key needs to be present"),
     ],
 )
-def test_warning_invalid_trainstep_output(tmpdir, case):
+def test_warning_invalid_trainstep_output(tmp_path, case):
     output, match = case
 
     class InvalidTrainStepModel(BoringModel):
         def training_step(self, batch, batch_idx):
             return output
 
     model = InvalidTrainStepModel()
-    trainer = Trainer(default_root_dir=tmpdir, fast_dev_run=1)
+    trainer = Trainer(default_root_dir=tmp_path, fast_dev_run=1)
 
     with pytest.raises(MisconfigurationException, match=match):
         trainer.fit(model)

tests/tests_pytorch/loops/test_all.py

@@ -87,13 +87,13 @@ def on_predict_batch_end(self, outputs, batch, *_):
         pytest.param("mps", marks=RunIf(mps=True)),
     ],
 )
-def test_callback_batch_on_device(tmpdir, accelerator):
+def test_callback_batch_on_device(tmp_path, accelerator):
     """Test that the batch object sent to the on_*_batch_start/end hooks is on the right device."""
     batch_callback = BatchHookObserverCallback()
 
     model = BatchHookObserverModel()
     trainer = Trainer(
-        default_root_dir=tmpdir,
+        default_root_dir=tmp_path,
         max_steps=1,
         limit_train_batches=1,
         limit_val_batches=1,

tests/tests_pytorch/loops/test_evaluation_loop.py

@@ -27,12 +27,12 @@
 
 
 @mock.patch("lightning.pytorch.loops.evaluation_loop._EvaluationLoop._on_evaluation_epoch_end")
-def test_on_evaluation_epoch_end(eval_epoch_end_mock, tmpdir):
+def test_on_evaluation_epoch_end(eval_epoch_end_mock, tmp_path):
     """Tests that `on_evaluation_epoch_end` is called for `on_validation_epoch_end` and `on_test_epoch_end` hooks."""
     model = BoringModel()
 
     trainer = Trainer(
-        default_root_dir=tmpdir, limit_train_batches=2, limit_val_batches=2, max_epochs=2, enable_model_summary=False
+        default_root_dir=tmp_path, limit_train_batches=2, limit_val_batches=2, max_epochs=2, enable_model_summary=False
     )
 
     trainer.fit(model)
@@ -89,7 +89,7 @@ def _get_dataloader():
 @mock.patch(
     "lightning.pytorch.trainer.connectors.logger_connector.logger_connector._LoggerConnector.log_eval_end_metrics"
 )
-def test_log_epoch_metrics_before_on_evaluation_end(update_eval_epoch_metrics_mock, tmpdir):
+def test_log_epoch_metrics_before_on_evaluation_end(update_eval_epoch_metrics_mock, tmp_path):
     """Test that the epoch metrics are logged before the `on_evaluation_end` hook is fired."""
     order = []
     update_eval_epoch_metrics_mock.side_effect = lambda _: order.append("log_epoch_metrics")
@@ -99,14 +99,14 @@ def on_validation_end(self):
             order.append("on_validation_end")
             super().on_validation_end()
 
-    trainer = Trainer(default_root_dir=tmpdir, fast_dev_run=1, enable_model_summary=False, num_sanity_val_steps=0)
+    trainer = Trainer(default_root_dir=tmp_path, fast_dev_run=1, enable_model_summary=False, num_sanity_val_steps=0)
     trainer.fit(LessBoringModel())
 
     assert order == ["log_epoch_metrics", "on_validation_end"]
 
 
 @RunIf(min_cuda_gpus=1)
-def test_memory_consumption_validation(tmpdir):
+def test_memory_consumption_validation(tmp_path):
     """Test that the training batch is no longer in GPU memory when running validation.
 
     Cannot run with MPS, since there we can only measure shared memory and not dedicated, which device has how much
@@ -155,7 +155,7 @@ def validation_step(self, batch, batch_idx):
     trainer = Trainer(
         accelerator="gpu",
         devices=1,
-        default_root_dir=tmpdir,
+        default_root_dir=tmp_path,
         fast_dev_run=2,
         enable_model_summary=False,
     )
@@ -460,7 +460,7 @@ def test_step(self, batch, batch_idx):
     assert seen == expected
 
 
-def test_evaluation_loop_when_batch_idx_argument_is_not_given(tmpdir):
+def test_evaluation_loop_when_batch_idx_argument_is_not_given(tmp_path):
     class TestModel(BoringModel):
         def __init__(self) -> None:
             super().__init__()
@@ -476,7 +476,7 @@ def test_step(self, batch):
             return {"y": self.step(batch)}
 
     trainer = Trainer(
-        default_root_dir=tmpdir,
+        default_root_dir=tmp_path,
         fast_dev_run=1,
         logger=False,
         enable_checkpointing=False,

tests/tests_pytorch/loops/test_evaluation_loop_flow.py

@@ -22,7 +22,7 @@
 from tests_pytorch.helpers.deterministic_model import DeterministicModel
 
 
-def test__eval_step__flow(tmpdir):
+def test__eval_step__flow(tmp_path):
     """Tests that only training_step can be used."""
 
     class TestModel(DeterministicModel):
@@ -45,7 +45,7 @@ def backward(self, loss):
 
     model = TestModel()
     trainer = Trainer(
-        default_root_dir=tmpdir,
+        default_root_dir=tmp_path,
         limit_train_batches=2,
         limit_val_batches=2,
         max_epochs=2,
@@ -71,7 +71,7 @@ def backward(self, loss):
     assert opt_closure_result.item() == 171
 
 
-def test__eval_step__epoch_end__flow(tmpdir):
+def test__eval_step__epoch_end__flow(tmp_path):
     """Tests that only training_step can be used."""
 
     class TestModel(DeterministicModel):
@@ -96,7 +96,7 @@ def backward(self, loss):
 
     model = TestModel()
     trainer = Trainer(
-        default_root_dir=tmpdir,
+        default_root_dir=tmp_path,
         limit_train_batches=2,
         limit_val_batches=2,
         max_epochs=2,

tests/tests_pytorch/loops/test_fetchers.py

@@ -168,7 +168,7 @@ def measure() -> float:
 
 
 @pytest.mark.parametrize("automatic_optimization", [False, True])
-def test_fetching_dataloader_iter_opt(automatic_optimization, tmpdir):
+def test_fetching_dataloader_iter_opt(automatic_optimization, tmp_path):
     class TestModel(BoringModel):
         def __init__(self, *args, automatic_optimization: bool = False, **kwargs):
             super().__init__(*args, **kwargs)
@@ -207,7 +207,7 @@ def on_train_epoch_end(self):
             assert self.count == 64
 
     model = TestModel(automatic_optimization=automatic_optimization)
-    trainer = Trainer(default_root_dir=tmpdir, max_epochs=1, accelerator="cpu")
+    trainer = Trainer(default_root_dir=tmp_path, max_epochs=1, accelerator="cpu")
     trainer.fit(model)
 
 
@@ -290,9 +290,9 @@ def train_dataloader(self):
         return DataLoader(RandomDataset(BATCH_SIZE, DATASET_LEN))
 
 
-def test_training_step_with_dataloader_iter(tmpdir) -> None:
+def test_training_step_with_dataloader_iter(tmp_path) -> None:
     """A baseline functional test for `training_step` with dataloader access."""
-    trainer = Trainer(max_epochs=1, default_root_dir=tmpdir, accelerator="cpu")
+    trainer = Trainer(max_epochs=1, default_root_dir=tmp_path, accelerator="cpu")
     m = AsyncBoringModel()
     trainer.fit(m)
     assert m.num_batches_processed == DATASET_LEN, f"Expect all {DATASET_LEN} batches to be processed."
@@ -406,7 +406,7 @@ def length(iterable, limit):
 
 
 @pytest.mark.parametrize("trigger_stop_iteration", [False, True])
-def test_stop_iteration_with_dataloader_iter(trigger_stop_iteration, tmpdir):
+def test_stop_iteration_with_dataloader_iter(trigger_stop_iteration, tmp_path):
     """Verify that StopIteration properly terminates the training when this is triggered from the current
     `dataloader_iter`"""
     EXPECT_NUM_BATCHES_PROCESSED = 2
@@ -428,7 +428,7 @@ def train_dataloader(self):
                 return DataLoader(RandomDataset(BATCH_SIZE, 2 * EXPECT_NUM_BATCHES_PROCESSED))
             return DataLoader(RandomDataset(BATCH_SIZE, EXPECT_NUM_BATCHES_PROCESSED))
 
-    trainer = Trainer(max_epochs=1, default_root_dir=tmpdir, accelerator="cpu")
+    trainer = Trainer(max_epochs=1, default_root_dir=tmp_path, accelerator="cpu")
     m = TestModel(trigger_stop_iteration)
     trainer.fit(m)
     expected = EXPECT_NUM_BATCHES_PROCESSED
@@ -437,7 +437,7 @@ def train_dataloader(self):
     assert m.num_batches_processed == expected
 
 
-def test_transfer_hooks_with_unpacking(tmpdir):
+def test_transfer_hooks_with_unpacking(tmp_path):
     """This test asserts the `transfer_batch` hooks are called only once per batch."""
 
     class RandomDictDataset(RandomDataset):
@@ -476,7 +476,7 @@ def validation_step(self, batch, batch_idx):
             x, _ = batch
             return super().validation_step(x, batch_idx)
 
-    trainer = Trainer(default_root_dir=tmpdir, max_epochs=1, num_sanity_val_steps=0)
+    trainer = Trainer(default_root_dir=tmp_path, max_epochs=1, num_sanity_val_steps=0)
     dm = BoringDataModule()
     trainer.fit(TestModel(), datamodule=dm)
     assert dm.count_called_on_before_batch_transfer == 4

tests/tests_pytorch/loops/test_flow_warnings.py

@@ -22,12 +22,12 @@ def training_step(self, batch, batch_idx):
         return self.step(batch[0])
 
 
-def test_no_depre_without_epoch_end(tmpdir):
+def test_no_depre_without_epoch_end(tmp_path):
     """Tests that only training_step can be used."""
     model = TestModel()
 
     trainer = Trainer(
-        default_root_dir=tmpdir,
+        default_root_dir=tmp_path,
         limit_train_batches=2,
         limit_val_batches=2,
         max_epochs=2,