Add a test that required submodules are initialized

This adds the new test, `test_required_submodule_is_initialized`, to
`test_fixture_health.py`. For the gitdb and smmap submodules, the
test asserts that the working tree directory exists and contains a
`.git` marker (file or directory). The failure message reminds the
developer to run `git submodule update --init --recursive`.

This is a regression test for a different contract than the preexisting
`test_fixture_dir_is_trusted_by_git` in `test_fixture_health.py`. That
test verifies git's willingness to operate in each fixture directory
(the `safe.directory` / dubious-ownership contract). This one verifies
that the directories are populated at all.

When the gitdb and smmap submodules aren't populated, the rest of the
suite fails in ways that don't name the cause: a cascade of failures
across `test_docs.py`, `test_repo.py`, and `test_submodule.py` -- tests
that need submodule content but don't set out to check for it. That
cascade was the failure mode of gitpython-developers#1713 on the Arch Linux build, where
`init-tests-after-clone.sh` had stopped running `git submodule update`
on CI. Wherever the submodules are actually missing, this test reports
the missing path directly instead.

The test skips when the source tree is not a git clone (no `.git` at
`REPO_ROOT`). This is to accommodate setups that run pytest from an
extracted release tarball and prepare submodules in a separately-pointed
tree via `GIT_PYTHON_TEST_GIT_REPO_BASE` (e.g. OpenIndiana's package
build). In such setups, `git submodule update` cannot operate on the
source tree, and the submodule paths checked here would never be
populated there, but the test suite still works because `rorepo`
points elsewhere.

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: EliahKagan

test/test_fixture_health.py

@@ -45,6 +45,19 @@
     ),
 ]
 
+# Submodule working trees that must be present and initialized for the
+# test suite to operate normally: gitdb at `git/ext/gitdb`, and smmap
+# nested inside gitdb at `git/ext/gitdb/gitdb/ext/smmap`. The paths
+# below are anchored at REPO_ROOT (the GitPython source tree), not at
+# any rorepo redirection target.
+SUBMODULE_DIRS = [
+    pytest.param(REPO_ROOT / "git" / "ext" / "gitdb", id="gitdb"),
+    pytest.param(
+        REPO_ROOT / "git" / "ext" / "gitdb" / "gitdb" / "ext" / "smmap",
+        id="smmap",
+    ),
+]
+
 
 @pytest.mark.parametrize("fixture_dir", FIXTURE_DIRS)
 def test_fixture_dir_is_trusted_by_git(fixture_dir: Path) -> None:
@@ -89,3 +102,40 @@ def test_fixture_dir_is_trusted_by_git(fixture_dir: Path) -> None:
         "This usually means the directory is not an initialized git "
         "repository (its `.git` marker may be stale or pointing elsewhere)."
     )
+
+
+@pytest.mark.parametrize("submodule_dir", SUBMODULE_DIRS)
+def test_required_submodule_is_initialized(submodule_dir: Path) -> None:
+    """The submodule's working tree is present and initialized.
+
+    Failure means the source tree is a git clone but the submodule's
+    working tree hasn't been populated. Skipped when the source tree
+    itself isn't a git clone (e.g. an extracted release tarball), since
+    ``git submodule update`` cannot operate there; setups that handle
+    submodules in a separately-prepared tree (via
+    ``GIT_PYTHON_TEST_GIT_REPO_BASE``) are exempted from this check.
+    """
+    if not (REPO_ROOT / ".git").exists():
+        pytest.skip(
+            "Source tree is not a git clone (no .git in REPO_ROOT); submodules "
+            "cannot be initialized via `git submodule update` here. Setups "
+            "that prepare submodules in a separately-pointed tree (via "
+            "GIT_PYTHON_TEST_GIT_REPO_BASE) are exempted from this check."
+        )
+    # The assertion messages below recommend `git submodule update --init
+    # --recursive` rather than `init-tests-after-clone.sh`, even though the
+    # latter is the documented entry point for first-time test setup. Two
+    # reasons: the script performs `git reset --hard` operations that can
+    # destroy local work, and #1713 showed the script itself can carry
+    # submodule-init regressions, in which case recommending it would lead
+    # developers in a circle. The direct git command is a safe minimal fix
+    # for this test's specific failure mode and bypasses any such regression.
+    assert submodule_dir.is_dir(), (
+        f"Submodule working tree missing: {submodule_dir}.\n"
+        "Run `git submodule update --init --recursive` from the repo root."
+    )
+    assert (submodule_dir / ".git").exists(), (
+        f"Submodule directory exists but has no .git marker: {submodule_dir}.\n"
+        "The submodule hasn't been initialized. "
+        "Run `git submodule update --init --recursive` from the repo root."
+    )