task template: base_copy without mv of the image tree

src/datasmith/harbor_adapter/template/tests/prepare.sh

@@ -109,12 +109,16 @@ fi
 
 SETUP_PHASE="base_copy"
 # lsv_measure.py times this unpatched copy (with its build) against the patched repo. Overlayfs cannot rename
-# image directories, so the repo is moved out and copied back to make both trees renamable.
+# image directories, so both trees are copies. No mv of the image tree: on a merged directory with ~10^5 git objects,
+# mv stopped on entries already gone (numpy#12445, pandas#40007); rm -rf skips them.
 if [ "${FC_LSV_PAIRED:-1}" != "0" ]; then
   cd /
-  rm -rf /workspace/.fc_base
-  mv /workspace/repo /workspace/.fc_base
-  cp -a /workspace/.fc_base /workspace/repo
+  rm -rf /workspace/.fc_base /workspace/.fc_new
+  cp -a /workspace/repo /workspace/.fc_base
+  cp -a /workspace/repo /workspace/.fc_new
+  rm -rf /workspace/repo
+  mv /workspace/.fc_new /workspace/repo
+  [ "$(git -C /workspace/repo rev-parse HEAD)" = "$(git -C /workspace/.fc_base rev-parse HEAD)" ]
   cd /workspace/repo
 fi
 

tests/docker/test_setup_base_copy.py

@@ -0,0 +1,27 @@
+import subprocess
+from pathlib import Path
+
+SETUP = Path(__file__).resolve().parents[2] / "src/datasmith/harbor_adapter/template/tests/prepare.sh"
+
+
+def block(root: Path) -> str:
+    text = SETUP.read_text()
+    start, end = text.index('SETUP_PHASE="base_copy"'), text.index('SETUP_PHASE="lsv_init"')
+    return text[start:end].replace("/workspace", str(root))
+
+
+def test_base_copy_makes_two_equal_trees_without_moving_the_repo(tmp_path):
+    repo = tmp_path / "repo"
+    repo.mkdir()
+    git = lambda *a: subprocess.run(["git", "-C", str(repo), *a], check=True, capture_output=True, text=True).stdout
+    git("init", "-q")
+    (repo / "a.py").write_text("x = 1\n")
+    git("add", "-A")
+    git("-c", "user.name=t", "-c", "user.email=t@t", "commit", "-q", "-m", "base")
+    head = git("rev-parse", "HEAD")
+    subprocess.run(["bash", "-c", "set -euo pipefail\n" + block(tmp_path)], check=True, capture_output=True)
+    for tree in (repo, tmp_path / ".fc_base"):
+        assert (tree / "a.py").read_text() == "x = 1\n"
+        assert subprocess.run(["git", "-C", str(tree), "rev-parse", "HEAD"], capture_output=True, text=True).stdout == head
+    assert not (tmp_path / ".fc_new").exists()
+    assert " mv /workspace/repo " not in SETUP.read_text()

tests/docker/test_setup_baseline_commit.py

@@ -0,0 +1,33 @@
+import re
+import subprocess
+import time
+from pathlib import Path
+
+SETUP = Path(__file__).resolve().parents[2] / "src/datasmith/harbor_adapter/template/tests/prepare.sh"
+
+
+def baseline_block(sha_file: Path) -> str:
+    text = SETUP.read_text()
+    start, end = text.index('SETUP_PHASE="baseline_commit"'), text.index('SETUP_PHASE="base_copy"')
+    stage = (SETUP.parents[1] / "shared" / "stage_tree.sh").read_text()
+    block = re.sub(r'\{% filter indent\(2\) %\}\n\{% include "shared/stage_tree.sh" %\}\n\{% endfilter %\}\n', lambda _: stage, text[start:end])
+    return block.replace("/opt/fc_baseline_sha", str(sha_file))
+
+
+def test_baseline_commit_starts_no_background_gc(tmp_path):
+    repo = tmp_path / "repo"
+    repo.mkdir()
+    git = lambda *a: subprocess.run(["git", "-C", str(repo), *a], check=True, capture_output=True, text=True).stdout
+    git("init", "-q")
+    for i in range(2000):
+        (repo / f"f{i}.txt").write_text(f"{i}\n")
+    git("add", "-A")
+    git("-c", "user.name=t", "-c", "user.email=t@t", "-c", "gc.auto=0", "commit", "-q", "-m", "base")
+    git("config", "gc.auto", "1")
+    (repo / "edit.txt").write_text("image edit\n")
+    subprocess.run(["bash", "-c", "set -euo pipefail\n" + baseline_block(tmp_path / "sha")], cwd=repo, check=True,
+                   capture_output=True)
+    time.sleep(2)
+    assert git("log", "-1", "--format=%s").strip() == "fc-baseline"
+    assert not list((repo / ".git/objects/pack").glob("*.pack"))
+    assert int(git("count-objects").split()[0]) > 2000