flakes: for detected git repository now assume shallow clones by default

Shallow clones are faster to access because we don't have to compute the
revCount, which in sparse checkouts might not even exists.

This is especially useful in combination with lazy trees in mind on large
repository such as nixpkgs.

src/libflake/flakeref.cc

@@ -168,8 +168,10 @@ std::pair<FlakeRef, std::string> parsePathFlakeRefWithFragment(
                         parsedURL.query.insert_or_assign("dir", subdir);
                     }
 
-                    if (pathExists(flakeRoot + "/.git/shallow"))
-                        parsedURL.query.insert_or_assign("shallow", "1");
+                    if (!parsedURL.query.count("shallow")) {
+                       // We assume shallow by default, so we don't need to compute the revCount, which is an expensive operation.
+                       parsedURL.query.insert_or_assign("shallow", "1");
+                    }
 
                     return fromParsedURL(fetchSettings, std::move(parsedURL), isFlake);
                 }

tests/functional/fetchGitShallow.sh

@@ -65,3 +65,8 @@ fi
 # Verify that we can shallow fetch the worktree
 git -C "$TEST_ROOT/shallow-worktree" rev-list --count HEAD >/dev/null
 nix eval --impure --raw --expr "(builtins.fetchGit { url = \"file://$TEST_ROOT/shallow-worktree\"; shallow = true; }).rev"
+
+# Normal flake operation work because they use shallow by default
+pushd "$TEST_ROOT/shallow-worktree"
+nix flake metadata
+popd