Forward args to _get_remote_config() and honour core/no_scm if present (#10719)

* Forward args to _get_remote_config and honour core/no_scm if present

* Pass full config, except cache, to Repo()

* Avoid modifying original kwargs contents

* Simplify if config

* Add test

* Relax error matching.. .. Avoid windows vs unix paths

* Update dvc/repo/open_repo.py

---------

Co-authored-by: skshetry <[email protected]>

dvc/dependency/repo.py

@@ -146,6 +146,10 @@ class RepoDependency(Dependency):
         else:
             config = Config.load_file(conf)
 
+        # Setup config to the new DVCFileSystem to use the remote repo, but rely on the
+        # local cache instead of the remote's cache. This avoids re-streaming of data,
+        # but messes up the call to `_get_remote_config()` downstream, which will need
+        # to ignore cache parameters.
         config["cache"] = self.repo.config["cache"]
         config["cache"]["dir"] = self.repo.cache.local_cache_dir
 

dvc/repo/open_repo.py

@@ -1,3 +1,4 @@
+import copy
 import os
 import tempfile
 import threading
@@ -50,7 +51,7 @@ def open_repo(url, *args, **kwargs):
     if os.path.exists(url):
         url = os.path.abspath(url)
         try:
-            config = _get_remote_config(url)
+            config = _get_remote_config(url, *args, **kwargs)
             config.update(kwargs.get("config") or {})
             kwargs["config"] = config
             return Repo(url, *args, **kwargs)
@@ -97,9 +98,24 @@ def clean_repos():
         _remove(path)
 
 
-def _get_remote_config(url):
+def _get_remote_config(url, *args, **kwargs):
     try:
-        repo = Repo(url)
+        # Deepcopy to prevent modifying the original `kwargs['config']`
+        config = copy.deepcopy(kwargs.get("config"))
+
+        # Import operations will use this function to get the remote's cache. However,
+        # while the `url` sent will point to the external repo, the cache information
+        # in `kwargs["config"]["cache"]["dir"]`) will point to the local repo,
+        # see `dvc/dependency/repo.py:RepoDependency._make_fs()`
+        #
+        # This breaks this function, since we'd be instructing `Repo()` to use the wrong
+        # cache to being with. We need to remove the cache info from `kwargs["config"]`
+        # to read the actual remote repo data.
+        if config:
+            config.pop("cache", None)
+
+        repo = Repo(url, config=config)
+
     except NotDvcRepoError:
         return {}
 

tests/func/api/test_data.py

@@ -5,6 +5,7 @@ from funcy import first, get_in
 
 from dvc import api
 from dvc.exceptions import OutputNotFoundError, PathMissingError
+from dvc.scm import CloneError, SCMError
 from dvc.testing.api_tests import TestAPI  # noqa: F401
 from dvc.testing.tmp_dir import make_subrepo
 from dvc.utils.fs import remove
@@ -65,6 +66,38 @@ def test_get_url_from_remote(tmp_dir, erepo_dir, cloud, local_cloud):
     )
 
 
+def test_get_url_ignore_scm(tmp_dir, dvc, cloud, scm):
+    tmp_dir.add_remote(config=cloud.config)
+    tmp_dir.dvc_gen("foo", "foo", commit="add foo")
+
+    repo_posix = tmp_dir.as_posix()
+    expected_url = (cloud / "files" / "md5" / "ac/bd18db4cc2f85cedef654fccc4a4d8").url
+
+    # Test baseline with scm
+    assert api.get_url("foo", repo=repo_posix) == expected_url
+
+    # Simulate gitless environment (e.g. deployed container)
+    (tmp_dir / ".git").rename(tmp_dir / "gitless_environment")
+
+    # Test failure mode when trying to access with git
+    with pytest.raises(SCMError, match="is not a git repository"):
+        api.get_url("foo", repo=repo_posix)
+
+    # Test successful access by ignoring git
+    assert (
+        api.get_url("foo", repo=repo_posix, config={"core": {"no_scm": True}})
+        == expected_url
+    )
+
+    # Addressing repos with `file://` triggers git, so it fails in a gitless environment
+    repo_url = f"file://{repo_posix}"
+    with pytest.raises(
+        CloneError,
+        match="SCM error",
+    ):
+        api.get_url("foo", repo=repo_url, config={"core": {"no_scm": True}})
+
+
 def test_open_external(tmp_dir, erepo_dir, cloud):
     erepo_dir.add_remote(config=cloud.config)