[BUG?] git submodule update --remote assumes 'origin' remote

I think I've found a bit of an edge case in
`git submodule update --remote` [1] that got easier to reproduce in
2.30+ under certain configurations.
`git submodule--helper print-default-remote` defaults to 'origin' when
run in a submodule in a detached HEAD state or on a branch with no
remote tracking branch [1].  In git < 2.30, the only way for that
default to be incorrect required a user to manually change remote names
with something like `git -C ./some-sm remote rename origin foo`.
`git rev-parse`ing that ref would naturally fail, but at least it's
because the user took manual steps to break git's assumptions.

Under git 2.30+ with clone.defaultRemoteName set however, submodules are
created with that remote name instead of 'origin'.  The same behavior
occurs, but this time without direct user action - only a simple
`git submodule init && git submodule update --remote` is required.

I'm terribly sorry about that regression - I've only just started
working on a few projects that use submodules heavily and probably could
have found this sooner.  I'm happy to fix it, but would *super*
appreciate a little guidance.  It seems the intention is "use the remote
that has the url found in the superproject's .submodules entry", that'd
require `git submodule--helper print-default-remote` to be called from
the superproject from what I can tell.  I've experimented with
introducing fallbacks to `remote_for_branch` in `remote.c` [2] as an
alternative:

1. use remote tracking branch; or
2. if there's only one remote, use that; or
3. if config.defaultRemoteName is set, use that; or
4. fall back to "origin"

This seems to work (at the very least, no tests fail?), but leaves
`cd ./some-sm; git remote add foo bar; git remote rename origin baz`
open to the original behavior.

Something tells me y'all will have a very simple solution that I'm
missing :)

Again, I'm so sorry for introducing a regression here!
Sean Barag

[1] https://github.com/git/git/blob/225365fb5195e804274ab569ac3cc4919451dc7f/git-submodule.sh#L571-L584
[2] https://github.com/git/git/blob/225365fb5195e804274ab569ac3cc4919451dc7f/remote.c#L477-L487

On Mon, Mar 01 2021, Sean Barag wrote:

I think I've found a bit of an edge case in
`git submodule update --remote` [1] that got easier to reproduce in
2.30+ under certain configurations.
`git submodule--helper print-default-remote` defaults to 'origin' when
run in a submodule in a detached HEAD state or on a branch with no
remote tracking branch [1].  In git < 2.30, the only way for that
default to be incorrect required a user to manually change remote names
with something like `git -C ./some-sm remote rename origin foo`.
`git rev-parse`ing that ref would naturally fail, but at least it's
because the user took manual steps to break git's assumptions.

Under git 2.30+ with clone.defaultRemoteName set however, submodules are
created with that remote name instead of 'origin'.  The same behavior
occurs, but this time without direct user action - only a simple
`git submodule init && git submodule update --remote` is required.

I'm terribly sorry about that regression - I've only just started
working on a few projects that use submodules heavily and probably could
have found this sooner.  I'm happy to fix it, but would *super*
appreciate a little guidance.  It seems the intention is "use the remote
that has the url found in the superproject's .submodules entry", that'd
require `git submodule--helper print-default-remote` to be called from
the superproject from what I can tell.  I've experimented with
introducing fallbacks to `remote_for_branch` in `remote.c` [2] as an
alternative:

1. use remote tracking branch; or
2. if there's only one remote, use that; or
3. if config.defaultRemoteName is set, use that; or
4. fall back to "origin"

This seems to work (at the very least, no tests fail?), but leaves
`cd ./some-sm; git remote add foo bar; git remote rename origin baz`
open to the original behavior.

Something tells me y'all will have a very simple solution that I'm
missing :)

Again, I'm so sorry for introducing a regression here!
Sean Barag

Just a quick note, I gather you're talking about de9ed3ef37 (clone:
allow configurable default for `-o`/`--origin`, 2020-10-01) which was
first released in v2.30.0.

Just noting that in the context of this bug already being in a release,
since we're now in the 2.31.0-rc0 period. Meaning likely not a release
blocker in the next few weeks.

On Mon, Mar 01 2021, Sean Barag wrote:

I've experimented with
introducing fallbacks to `remote_for_branch` in `remote.c` [2] as an
alternative:

1. use remote tracking branch; or
2. if there's only one remote, use that; or
3. if config.defaultRemoteName is set, use that; or
4. fall back to "origin"

This seems to work (at the very least, no tests fail?), but leaves
`cd ./some-sm; git remote add foo bar; git remote rename origin baz`
open to the original behavior.

I forgot to mention that this approach requires `get_default_remote` in
`submodule--helper.c` to use the name of the remote returned by
`remote_get(NULL)`.  That was probably obvious to regular mailing list
readers :D

On Mon, Mar 1, 2021, at 8:38 AM, Ævar Arnfjörð Bjarmason wrote:

Just a quick note, I gather you're talking about de9ed3ef37 (clone:
allow configurable default for `-o`/`--origin`, 2020-10-01) which was
first released in v2.30.0.

Yes, that's exactly the commit.  I knew I forgot to include something.

Just noting that in the context of this bug already being in a
release, since we're now in the 2.310-rc0 period. Meaning likely not
a release blocker in the next few weeks.

That makes sense - thanks for clarifying!