Skip to contentSkip to conversation

submodule--helper: set alternateLocation for cloned submodules

PATCH In review [email protected] 3 participants 7 messages Dec 8, 2016

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 submodule--helper: set alternateLocation for cloned submodules

From: "Vitaly \"_Vi\" Shukela" <[email protected]>

In 31224cbdc7 (clone: recursive and reference option triggers submodule alternates, 2016-08-17) a mechanism was added to have submodules referenced. It did not address _nested_ submodules, however.

This patch makes all not just the root repository, but also all submodules (recursively) have submodule.alternateLocation and submodule.alternateErrorStrategy configured, making Git search for possible alternates for nested submodules as well.

As submodule's alternate target does not end in .git/objects (rather .git/modules/qqqqqq/objects), this alternate target path restriction for in add_possible_reference_from_superproject relates from "*.git/objects" to just */objects".

New tests have been added to t7408-submodule-reference.

---

Notes:
Another round of fixups:

* Corrected code style
* Refactored and fixed the test

Previously test contained errorneous
test_must_fail, which was masked by
missing &&.

Mailmap patch omitted this time.

 builtin/submodule--helper.c    | 19 ++++++++++--
 t/t7408-submodule-reference.sh | 66 ++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 83 insertions(+), 2 deletions(-)
builtin/submodule--helper.c +17 -2
index 4beeda5..92fd676 100644
@@ -498,9 +498,9 @@ static int add_possible_reference_from_superproject(
 
 	/*
 	 * If the alternate object store is another repository, try the
-	 * standard layout with .git/modules/<name>/objects
+	 * standard layout with .git/(modules/<name>)+/objects
 	 */
-	if (ends_with(alt->path, ".git/objects")) {
+	if (ends_with(alt->path, "/objects")) {
 		char *sm_alternate;
 		struct strbuf sb = STRBUF_INIT;
 		struct strbuf err = STRBUF_INIT;
@@ -583,6 +583,7 @@ static int module_clone(int argc, const char **argv, const char *prefix)
 	struct strbuf rel_path = STRBUF_INIT;
 	struct strbuf sb = STRBUF_INIT;
 	struct string_list reference = STRING_LIST_INIT_NODUP;
+	char *sm_alternate = NULL, *error_strategy = NULL;
 
 	struct option module_clone_options[] = {
 		OPT_STRING(0, "prefix", &prefix,
@@ -672,6 +673,20 @@ static int module_clone(int argc, const char **argv, const char *prefix)
 		die(_("could not get submodule directory for '%s'"), path);
 	git_config_set_in_file(p, "core.worktree",
 			       relative_path(path, sm_gitdir, &rel_path));
+
+	/* setup alternateLocation and alternateErrorStrategy in the cloned submodule if needed */
+	git_config_get_string("submodule.alternateLocation", &sm_alternate);
+	if (sm_alternate)
+		git_config_set_in_file(p, "submodule.alternateLocation",
+					   sm_alternate);
+	git_config_get_string("submodule.alternateErrorStrategy", &error_strategy);
+	if (error_strategy)
+		git_config_set_in_file(p, "submodule.alternateErrorStrategy",
+					   error_strategy);
+
+	free(sm_alternate);
+	free(error_strategy);
+
 	strbuf_release(&sb);
 	strbuf_release(&rel_path);
 	free(sm_gitdir);
t/t7408-submodule-reference.sh +66 -0

70 lines not on this page · raw message

Signature · 1 line
2.10.2
#1 lore raw
Quoted 4 lines · On Wed, Dec 7, 2016 at 4:39 PM, <[email protected]> wrote:

Previously test contained errorneous
test_must_fail, which was masked by
missing &&.

I wonder if we could make either
the test_must_fail intelligent to detect such a broken && call chain or the test_expect_success macro to see for those broken chains.

Patch looks good to me except one very minor nit. :)

t/t7408-submodule-reference.sh test_expect_success 'ignoring missing submodule alternates passes clone and subm · line 157

Quoted 9 lines
t/t7408-submodule-reference.sh +9 -0
+test_expect_success 'nested submodule alternate in works and is actually used' '
+       test_when_finished "rm -rf supersuper-clone" &&
+       git clone --recursive --reference supersuper supersuper supersuper-clone &&
+       (
+               cd supersuper-clone &&
+               # test superproject has alternates setup correctly
+               test_alternate_is_used .git/objects/info/alternates . &&
+               # immediate submodule has alternate:
+               test_alternate_is_used .git/modules/subwithsub/objects/info/alternates subwithsub

here is a && missing ;)

#2 lore raw
Quoted 10 lines · Stefan Beller
Quoted 4 lines · On Wed, Dec 7, 2016 at 4:39 PM, <[email protected]> wrote:

Previously test contained errorneous
test_must_fail, which was masked by
missing &&.

I wonder if we could make either
the test_must_fail intelligent to detect such a broken && call chain or the test_expect_success macro to see for those broken chains.

I don't think test_must_fail is relevant for &&-chains. Even something like:

test_must_fail foo
bar

or:

bar
test_must_fail foo

will both trigger on the &&-chain linter, because it uses a magic exit code to detect the breakage. I think the problem is just that the &&-chain linter cannot peek inside subshells, and that's where the bug was in this case.

I wish we could improve that, but I spend a lot of brain cycles on it at one point and couldn't come up with a workable solution.

-Peff

#3 lore raw
Quoted 5 lines · Jeff King

will both trigger on the &&-chain linter, because it uses a magic exit code to detect the breakage. I think the problem is just that the &&-chain linter cannot peek inside subshells, and that's where the bug was in this case.

Uh, yeah in the subshell, but the patch v2 did have it not in subshells, I'll take another look.

Quoted 3 lines

I wish we could improve that, but I spend a lot of brain cycles on it at one point and couldn't come up with a workable solution.

Is it possible to overwrite what happens when you open a subshell with ( ) ? i.e. I imagine in the global test-setup we'd setup that ( ) not just opens /bin/sh
but a shell with benefits such that we can execute just one && chain.

author
#4 lore raw
In reply to Jeff King
Quoted 12 lines · Jeff King
Quoted 10 lines · Stefan Beller
Quoted 3 lines · On Wed, Dec 7, 2016 at 4:39 PM, <[email protected]> wrote:

Previously test contained errorneous
test_must_fail, which was masked by
missing &&.

I wonder if we could make either
the test_must_fail intelligent to detect such a broken && call chain or the test_expect_success macro to see for those broken chains.

Quoted 5 lines
Quoted 5 lines

I wish we could improve that, but I spend a lot of brain cycles on it at one point and couldn't come up with a workable solution.

-Peff

Why Git test use &&-chains instead of proper "set -e"?

#5 lore raw
Quoted 15 lines · On Thu, Dec 8, 2016 at 10:04 AM, vi0oss <[email protected]> wrote:
Quoted 14 lines · Jeff King
Quoted 11 lines · Stefan Beller
Quoted 3 lines · On Wed, Dec 7, 2016 at 4:39 PM, <[email protected]> wrote:

Previously test contained errorneous
test_must_fail, which was masked by
missing &&.

I wonder if we could make either
the test_must_fail intelligent to detect such a broken && call chain or the test_expect_success macro to see for those broken chains.

Quoted 7 lines
Quoted 5 lines
Quoted 5 lines

I wish we could improve that, but I spend a lot of brain cycles on it at one point and couldn't come up with a workable solution.

-Peff

Why Git test use &&-chains instead of proper "set -e"?

"Because set -e kills the shell and we would want to keep going until the test suite is finished and display a summary what failed" would be my first reaction, but let's dig into history:
bb79af9d09 might be helpful on that, but it doesn't explain why we use && chains.
I could not find any commit explaining the use of && chains.

e1970ce43abf might be interesting (the introduction of the test suite), as that did not contain && chains.

I guess it would be hard(er) to implement e.g. test_must_fail in an environment where -e is set.

#6 lore raw
In reply to vi0oss
Quoted 1 line · On Thu, Dec 08, 2016 at 09:04:46PM +0300, vi0oss wrote:

Why Git test use &&-chains instead of proper "set -e"?

Because "set -e" comes with all kinds of confusing corner cases. Using && chains is annoying, but rarely surprising.

One of my favorite examples is:

  set -e
  (
    false
    echo 1
  ) || {
    echo outcome=$?
    false
  }
  echo 2

which prints both "1" and "2".

Inside the subshell, "set -e" has no effect, and you cannot re-enable it by setting "-e" (it's suppressed entirely because we are on the left-hand side of an || conditional).

So you could write a function like this:

  foo() {
    do_one
    do_two
  }

that relies on catching the failure from do_one. And it works here:

set -e
foo

but not here:

set -e
if foo then
do_something
fi

And there's no way to make it work without adding back in the &&-chaining.

-Peff