fix: staging error in GitHubConnector (#529)

* fix: staging error in GitHub connector

Fixes an issue when calling stage on a github connector. We were calling
_checkout_revision before cloining/downloading the repo which caused an
error when trying to interact with the local directory.

The fix is simple, move the call to _checkout_revision to after the
download step.

* add test case covering the bug

* add another assertion for better validation of staging logic

garden_ai/model_connectors/model_connector.py

@@ -154,15 +154,16 @@ def stage(self) -> str:
             ConnectorStagingError: If something goes wrong during staging.
         """
         try:
-            self._checkout_revision()
-
             # pull from the repo if we have already downloaded it.
             if self._pull_if_downloaded():
-                return str(self.local_dir)
+                model_dir = str(self.local_dir)
             else:
                 # otherwise download the repo
-                return self._download()
+                model_dir = self._download()
 
+            # ensure the correct commit is checked out
+            self._checkout_revision()
+            return model_dir
         except Exception as e:
             raise ConnectorStagingError(str(self.repo_url), None) from e
 

tests/model_connectors/test_github_conn.py

@@ -1,7 +1,8 @@
 import base64
+from pathlib import Path
 import pytest
 
-from garden_ai.model_connectors import GitHubConnector
+from garden_ai.model_connectors import GitHubConnector, create_connector
 from garden_ai.model_connectors.exceptions import (
     ConnectorLFSError,
     ConnectorInvalidRepoIdError,
@@ -141,3 +142,33 @@ def test_init_raises_on_non_github_url(
         GitHubConnector(
             repo_url="https://some.repo.example.com",
         )
+
+
+@pytest.mark.integration
+def test_stage_works_on_empty_local_dir(
+    tmp_path,
+):
+    """This tests calling `stage` on a GitHubConnector when the local directory is empty.
+
+    This test may fail if there is no internet access or github.com is unreachable.
+
+    This was created as part of the resolution of this issue:
+    https://github.com/Garden-AI/garden/issues/521
+    """
+    c = create_connector(
+        "https://github.com/WillEngler/model-perovskite-ASR", local_dir=tmp_path
+    )
+    assert isinstance(c, GitHubConnector)
+
+    # This might still throw errors if the network is down or github is unreachable
+    model_dir = c.stage()
+
+    # Make sure we have cloned the repo
+    git_dir = Path(model_dir) / ".git"
+    assert git_dir.exists()
+
+    # Ensure we have checked out the correct revision
+    head = git_dir / "HEAD"
+    with open(head, "r") as f:
+        head_rev = f.readline().strip()
+    assert head_rev == c.revision