submodule--helper: check clone.defaultRemoteName in print-default-remote

From: Sean Barag <[email protected]>

git submodule update --remote previously assumed that submodules in detached-HEAD state (or with no remote configured via branch.$branchname.remote) should always fetch from "origin". While this was often true, users could rename the "origin" remote within that submodule. This would result in a pair of fatal errors:

fatal: Needed a single revision
fatal: Unable to find current origin/HEAD revision in submodule path 'path/to/submodule'

Since de9ed3ef37 (clone: allow configurable default for -o/--origin, 2020-10-01) (released in git 2.30) however, that edge-case was trivially easy to get into. Simply set clone.defaultRemoteName, clone a repo with submodules, run git submodule update --init, and the next git submodule update --remote would fail. Support clone.defaultRemoteName in submodules by checking for the configured default remote name before falling back to "origin".

---
[RFC PATCH] submodule--helper: check clone.defaultRemoteName in print-default-remote

What used to be an edge case with git submodule update --remote got significantly easier to reproduce in 2.30+ with clone.defaultRemoteName set in git config. git-submodule.sh relies on git submodule--helper print-default-remote to determine which remote to fetch from, which falls back to "origin" when an explicit remote can't be found (e.g. detached HEAD states). Where clone.defaultRemoteName is set, however, that fallback to "origin" is no longer appropriate, because the "origin" remote doesn't exist. Either way, this results in a fatal error:

fatal: Needed a single revision
fatal: Unable to find current origin/HEAD revision in submodule path 'path/to/submodule'

This RFC patch adds a check for clone.defaultRemoteName in submodule-helper's get_default_remote(), but would love some feedback before this becomes a proper [PATCH] email. This works in my initial testing and in the added test case, but it feels like I'm missing something.

Should I add test cases that don't rely on detached HEAD state perhaps? I'm not thrilled with the duplication of git_config_get_string calls for clone.defaultRemoteName, but extracting that logic to a static char *get_clone_default_remote_name() felt worse. I'd be happy to have any feedback y'all can provide :)

From a high-level, I suppose this leaves the pre-2.30 edge case
unresolved: users can still rename a submodule's remote to something else (something != clone.defaultRemoteName if it's set, or something != "origin" if it's not set) and trigger this failure. Perhaps a long-term solution would involve querying the list of remotes for each submodule, and if there's only one remote assuming it's the correct one to fetch from, with the proposed solution as a fallback for multi-remote situations.

Anyway, please let me know what I can do to clean this up, make it more robust, etc. before it becomes a more formal patch submission. Thanks in advance! Sean Barag

Related conversations:

* The original bug report:
https://lore.kernel.org/git/[email protected]

* An adjacent conversation about submodule--helper and remote names:
https://lore.kernel.org/git/[email protected]

Philippe Blain [email protected], Atharva Raykar [email protected], Christian Couder [email protected], Shourya Shukla [email protected], Emily Shaffer [email protected]

Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-1112%2Fsjbarag%2Fcheck-clone.defaultremotename-in-submodule-helper-print-default-remote-v1
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-1112/sjbarag/check-clone.defaultremotename-in-submodule-helper-print-default-remote-v1
Pull-Request: https://github.com/gitgitgadget/git/pull/1112

 builtin/submodule--helper.c | 35 +++++++++++++++++++++++++----------
 t/t7400-submodule-basic.sh  | 18 ++++++++++++++++++
 2 files changed, 43 insertions(+), 10 deletions(-)

index 9b25a508e6a..2d1468508ff 100644
@@ -38,18 +38,33 @@ static char *get_default_remote(void)
 	if (!refname)
 		die(_("No such ref: %s"), "HEAD");
 
-	/* detached HEAD */
-	if (!strcmp(refname, "HEAD"))
-		return xstrdup("origin");
+	/* detached HEAD should use the configured default remote name or fall back to "origin" */
+	if (!strcmp(refname, "HEAD")) {
+		strbuf_addstr(&sb, "clone.defaultRemoteName");
 
-	if (!skip_prefix(refname, "refs/heads/", &refname))
-		die(_("Expecting a full ref name, got %s"), refname);
+		if (!git_config_get_string(sb.buf, &dest))
+			ret = dest;
+		else
+			ret = xstrdup("origin");
+	} else {
+		if (!skip_prefix(refname, "refs/heads/", &refname))
+			die(_("Expecting a full ref name, got %s"), refname);
 
-	strbuf_addf(&sb, "branch.%s.remote", refname);
-	if (git_config_get_string(sb.buf, &dest))
-		ret = xstrdup("origin");
-	else
-		ret = dest;
+		strbuf_addf(&sb, "branch.%s.remote", refname);
+		if (!git_config_get_string(sb.buf, &dest))
+			/* use configured remote name if one is present */
+			ret = dest;
+		else {
+			/* otherwise use configured *default* remote name, falling back to "origin" */
+			strbuf_reset(&sb);
+			strbuf_addstr(&sb, "clone.defaultRemoteName");
+
+			if (!git_config_get_string(sb.buf, &dest))
+				ret = dest;
+			else
+				ret = xstrdup("origin");
+		}
+	}
 
 	strbuf_release(&sb);
 	return ret;

index e7cec2e457a..ce0801321a9 100755
@@ -1416,4 +1416,22 @@ test_expect_success 'recursive clone respects -q' '
 	test_must_be_empty actual
 '
 
+test_expect_success 'submodule update --remote respects clone.defaultRemoteName' '
+	test_when_finished "rm -rf multisuper_clone" &&
+
+	# expect is intentionally empty.
+	# no output expected, since submodule "sub0" will be up to date
+	> expect &&
+
+	git clone multisuper multisuper_clone &&
+	git -C multisuper_clone submodule update --init -- sub0 &&
+	git -C multisuper_clone/sub0 remote rename origin upstream &&
+	git \
+		-C multisuper_clone \
+		-c clone.defaultRemoteName=upstream \
+		submodule update --remote -- sub0 \
+		2>&1 >actual &&
+	test_cmp expect actual
+'
+
 test_done

base-commit: e83ba647f7c61cf945690d6a0bd8c172a6498dc8

gitgitgadget

Whoops, my CC list got mangled. Fixed.