Git development
 help / color / mirror / Atom feed
* [PATCH] sequencer: refuse to revert shallow boundary commits by default
@ 2026-10-21 10:09 Nakul V S via GitGitGadget
  2026-10-21 15:20 ` Phillip Wood
  0 siblings, 1 reply; 4+ messages in thread
From: Nakul V S via GitGitGadget @ 2026-10-21 10:09 UTC (permalink / raw)
  To: git; +Cc: Patrick Steinhardt, Junio C Hamano, Nakul V S, Nakul V S

From: Nakul V S <vsnakul@gmail.com>

When a repository is cloned with --depth=1, Git records the boundary
commit in .git/shallow and rewrites its parent links via the graft
mechanism, making it appear as a root commit. Reverting such a commit
will therefore delete all tracked files, which is almost always
unintended by the user.

Add a check in do_pick_commit() that detects when the target commit is
a shallow boundary (nr_parent < 0 in the commit graft) and refuses the
operation by default. A new --allow-shallow-boundary flag is provided
for the rare cases where the user genuinely intends this behavior.

Signed-off-by: Nakul V S <vsnakul@gmail.com>
---
    sequencer: refuse to revert shallow boundary commits by default

Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-2273%2Fvsnakul%2Fsequencer-shallow-boundary-check-v1
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-2273/vsnakul/sequencer-shallow-boundary-check-v1
Pull-Request: https://github.com/gitgitgadget/git/pull/2273

 Documentation/git-revert.adoc | 12 +++++++++++-
 builtin/revert.c              |  4 +++-
 sequencer.c                   | 14 ++++++++++++++
 sequencer.h                   |  1 +
 t/t3501-revert-cherry-pick.sh | 25 +++++++++++++++++++++++++
 5 files changed, 54 insertions(+), 2 deletions(-)

diff --git a/Documentation/git-revert.adoc b/Documentation/git-revert.adoc
index ffba365e63..4395df62e2 100644
--- a/Documentation/git-revert.adoc
+++ b/Documentation/git-revert.adoc
@@ -8,7 +8,7 @@ git-revert - Revert some existing commits
 SYNOPSIS
 --------
 [verse]
-'git revert' [--[no-]edit] [-n] [-m <parent-number>] [-s] [-S[<keyid>]] <commit>...
+'git revert' [--[no-]edit] [-n] [-m <parent-number>] [-s] [-S[<keyid>]] [--allow-shallow-boundary] <commit>...
 'git revert' (--continue | --skip | --abort | --quit)
 
 DESCRIPTION
@@ -122,6 +122,16 @@ include::rerere-options.adoc[]
 	configuration variable can be used to enable this option by
 	default.
 
+--allow-shallow-boundary::
+	By default, 'git revert' refuses to revert a commit that is a
+	shallow boundary (i.e. a commit listed in `.git/shallow`),
+	because its true parent history is absent from the clone and
+	reverting it would remove all tracked files. This option
+	overrides that safeguard and allows the revert to proceed.
++
+To restore full history before reverting, use
+`git fetch --unshallow` instead.
+
 
 SEQUENCER SUBCOMMANDS
 ---------------------
diff --git a/builtin/revert.c b/builtin/revert.c
index 52100a20cb..f27397fdec 100644
--- a/builtin/revert.c
+++ b/builtin/revert.c
@@ -24,7 +24,7 @@
  */
 
 static const char * const revert_usage[] = {
-	N_("git revert [--[no-]edit] [-n] [-m <parent-number>] [-s] [-S[<keyid>]] <commit>..."),
+	N_("git revert [--[no-]edit] [-n] [-m <parent-number>] [-s] [-S[<keyid>]] [--allow-shallow-boundary] <commit>..."),
 	N_("git revert (--continue | --skip | --abort | --quit)"),
 	NULL
 };
@@ -128,6 +128,8 @@ static int run_sequencer(int argc, const char **argv, const char *prefix,
 		OPT_BOOL('e', "edit", &opts->edit, N_("edit the commit message")),
 		OPT_NOOP_NOARG('r', NULL),
 		OPT_BOOL('s', "signoff", &opts->signoff, N_("add a Signed-off-by trailer")),
+		OPT_BOOL(0, "allow-shallow-boundary", &opts->allow_shallow_boundary,
+			 N_("allow reverting a shallow boundary commit")),
 		OPT_CALLBACK('m', "mainline", opts, N_("parent-number"),
 			     N_("select mainline parent"), option_parse_m),
 		OPT_RERERE_AUTOUPDATE(&opts->allow_rerere_auto),
diff --git a/sequencer.c b/sequencer.c
index efbfd1315b..6de59bf6aa 100644
--- a/sequencer.c
+++ b/sequencer.c
@@ -49,6 +49,7 @@
 #include "rebase-interactive.h"
 #include "reset.h"
 #include "branch.h"
+#include "shallow.h"
 
 #define GIT_REFLOG_ACTION "GIT_REFLOG_ACTION"
 
@@ -2366,6 +2367,19 @@ static enum pick_result do_pick_commit(struct repository *r,
 	}
 	discard_index(r->index);
 
+	if (is_repository_shallow(r)) {
+		struct commit_graft *graft = lookup_commit_graft(r, &commit->object.oid);
+		if (graft && graft->nr_parent < 0 && !opts->allow_shallow_boundary) {
+			return error(_("%s: commit %s is a shallow boundary commit.\n"
+				       "Reverting it will remove all tracked files because its\n"
+				       "true parent history is not present in this clone.\n"
+				       "Use --allow-shallow-boundary to override, or\n"
+				       "'git fetch --unshallow' to restore full history first."),
+				     command_to_string(command),
+				     oid_to_hex(&commit->object.oid));
+		}
+	}
+
 	if (!commit->parents)
 		parent = NULL;
 	else if (commit->parents->next) {
diff --git a/sequencer.h b/sequencer.h
index a5d12a1070..f334bc5357 100644
--- a/sequencer.h
+++ b/sequencer.h
@@ -57,6 +57,7 @@ struct replay_opts {
 	int committer_date_is_author_date;
 	int ignore_date;
 	int commit_use_reference;
+	int allow_shallow_boundary;
 
 	struct strvec trailer_args;
 
diff --git a/t/t3501-revert-cherry-pick.sh b/t/t3501-revert-cherry-pick.sh
index 939e7a16a6..508464004f 100755
--- a/t/t3501-revert-cherry-pick.sh
+++ b/t/t3501-revert-cherry-pick.sh
@@ -256,4 +256,29 @@ test_expect_success 'cherry-pick is unaware of --reference (for now)' '
 	test_grep "^usage: git cherry-pick" actual
 '
 
+# Tests for shallow boundary safeguard
+
+test_expect_success 'setup shallow boundary tests' '
+	git init src &&
+	test_commit -C src A &&
+	test_commit -C src B &&
+	git clone --no-local --depth=1 src shallow-repo
+'
+
+test_expect_success 'revert of shallow boundary commit is refused by default' '
+	test_must_fail git -C shallow-repo revert --no-edit HEAD 2>err &&
+	test_grep "shallow boundary commit" err
+'
+
+test_expect_success 'revert of shallow boundary succeeds with --allow-shallow-boundary' '
+	git -C shallow-repo revert --no-edit --allow-shallow-boundary HEAD
+'
+
+test_expect_success 'revert in non-shallow repo is unaffected' '
+	git init non-shallow &&
+	test_commit -C non-shallow A &&
+	test_commit -C non-shallow B &&
+	git -C non-shallow revert --no-edit HEAD
+'
+
 test_done

base-commit: 47fdf2208c5b94c8cc01c106b8f7c85efb91c926
-- 
gitgitgadget

^ permalink raw reply related	[flat|nested] 4+ messages in thread

* Re: [PATCH] sequencer: refuse to revert shallow boundary commits by default
  2026-10-21 10:09 [PATCH] sequencer: refuse to revert shallow boundary commits by default Nakul V S via GitGitGadget
@ 2026-10-21 15:20 ` Phillip Wood
  2026-10-21 19:11   ` Junio C Hamano
  0 siblings, 1 reply; 4+ messages in thread
From: Phillip Wood @ 2026-10-21 15:20 UTC (permalink / raw)
  To: Nakul V S via GitGitGadget, git
  Cc: Patrick Steinhardt, Junio C Hamano, Nakul V S

Hi Nakul

On 21/10/2026 11:09, Nakul V S via GitGitGadget wrote:
> From: Nakul V S <vsnakul@gmail.com>
> 
> When a repository is cloned with --depth=1, Git records the boundary
> commit in .git/shallow and rewrites its parent links via the graft
> mechanism, making it appear as a root commit. Reverting such a commit
> will therefore delete all tracked files, which is almost always
> unintended by the user.
> 
> Add a check in do_pick_commit() that detects when the target commit is
> a shallow boundary (nr_parent < 0 in the commit graft) and refuses the
> operation by default. A new --allow-shallow-boundary flag is provided
> for the rare cases where the user genuinely intends this behavior.

Thanks for working on this, it is a useful and timely improvement as 
there was some discussion about improving the behavior of git with 
regard to shallow boundaries at the recent git merge conference. I 
wonder if it would be better to refuse to cherry-pick and rebase a 
shallow boundary commit as well. Do you know of a sensible use for 
reverting a shallow boundary or is the new option added out of caution 
in case someone wants it?

> diff --git a/builtin/revert.c b/builtin/revert.c
> index 52100a20cb..f27397fdec 100644
> --- a/builtin/revert.c
> +++ b/builtin/revert.c
> @@ -24,7 +24,7 @@
>    */
>   
>   static const char * const revert_usage[] = {
> -	N_("git revert [--[no-]edit] [-n] [-m <parent-number>] [-s] [-S[<keyid>]] <commit>..."),
> +	N_("git revert [--[no-]edit] [-n] [-m <parent-number>] [-s] [-S[<keyid>]] [--allow-shallow-boundary] <commit>..."),

For such a rarely used option I don't think we need to add it to the 
synopsis (we already omit some other options like --rerere-autoupdate)

> diff --git a/sequencer.c b/sequencer.c
> index efbfd1315b..6de59bf6aa 100644
> --- a/sequencer.c
> +++ b/sequencer.c
> @@ -2366,6 +2367,19 @@ static enum pick_result do_pick_commit(struct repository *r,
>   	}
>   	discard_index(r->index);
>   
> +	if (is_repository_shallow(r)) {
> +		struct commit_graft *graft = lookup_commit_graft(r, &commit->object.oid);
> +		if (graft && graft->nr_parent < 0 && !opts->allow_shallow_boundary) {

It would be good to extract this into a helper function so that we don't 
have to copy the code to detect a shallow boundary each time we need to 
detect one.

Thanks

Phillip

> +			return error(_("%s: commit %s is a shallow boundary commit.\n"
> +				       "Reverting it will remove all tracked files because its\n"
> +				       "true parent history is not present in this clone.\n"
> +				       "Use --allow-shallow-boundary to override, or\n"
> +				       "'git fetch --unshallow' to restore full history first."),
> +				     command_to_string(command),
> +				     oid_to_hex(&commit->object.oid));
> +		}
> +	}
> +
>   	if (!commit->parents)
>   		parent = NULL;
>   	else if (commit->parents->next) {
> diff --git a/sequencer.h b/sequencer.h
> index a5d12a1070..f334bc5357 100644
> --- a/sequencer.h
> +++ b/sequencer.h
> @@ -57,6 +57,7 @@ struct replay_opts {
>   	int committer_date_is_author_date;
>   	int ignore_date;
>   	int commit_use_reference;
> +	int allow_shallow_boundary;
>   
>   	struct strvec trailer_args;
>   
> diff --git a/t/t3501-revert-cherry-pick.sh b/t/t3501-revert-cherry-pick.sh
> index 939e7a16a6..508464004f 100755
> --- a/t/t3501-revert-cherry-pick.sh
> +++ b/t/t3501-revert-cherry-pick.sh
> @@ -256,4 +256,29 @@ test_expect_success 'cherry-pick is unaware of --reference (for now)' '
>   	test_grep "^usage: git cherry-pick" actual
>   '
>   
> +# Tests for shallow boundary safeguard
> +
> +test_expect_success 'setup shallow boundary tests' '
> +	git init src &&
> +	test_commit -C src A &&
> +	test_commit -C src B &&
> +	git clone --no-local --depth=1 src shallow-repo
> +'
> +
> +test_expect_success 'revert of shallow boundary commit is refused by default' '
> +	test_must_fail git -C shallow-repo revert --no-edit HEAD 2>err &&
> +	test_grep "shallow boundary commit" err
> +'
> +
> +test_expect_success 'revert of shallow boundary succeeds with --allow-shallow-boundary' '
> +	git -C shallow-repo revert --no-edit --allow-shallow-boundary HEAD
> +'
> +
> +test_expect_success 'revert in non-shallow repo is unaffected' '
> +	git init non-shallow &&
> +	test_commit -C non-shallow A &&
> +	test_commit -C non-shallow B &&
> +	git -C non-shallow revert --no-edit HEAD
> +'
> +
>   test_done
> 
> base-commit: 47fdf2208c5b94c8cc01c106b8f7c85efb91c926


^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [PATCH] sequencer: refuse to revert shallow boundary commits by default
  2026-10-21 15:20 ` Phillip Wood
@ 2026-10-21 19:11   ` Junio C Hamano
  2026-10-26 10:58     ` Phillip Wood
  0 siblings, 1 reply; 4+ messages in thread
From: Junio C Hamano @ 2026-10-21 19:11 UTC (permalink / raw)
  To: Phillip Wood
  Cc: Nakul V S via GitGitGadget, git, Patrick Steinhardt, Nakul V S

Phillip Wood <phillip.wood123@gmail.com> writes:

> Hi Nakul
>
> On 21/10/2026 11:09, Nakul V S via GitGitGadget wrote:
>> From: Nakul V S <vsnakul@gmail.com>
>> 
>> When a repository is cloned with --depth=1, Git records the boundary
>> commit in .git/shallow and rewrites its parent links via the graft
>> mechanism, making it appear as a root commit. Reverting such a commit
>> will therefore delete all tracked files, which is almost always
>> unintended by the user.
>> 
>> Add a check in do_pick_commit() that detects when the target commit is
>> a shallow boundary (nr_parent < 0 in the commit graft) and refuses the
>> operation by default. A new --allow-shallow-boundary flag is provided
>> for the rare cases where the user genuinely intends this behavior.
>
> Thanks for working on this, it is a useful and timely improvement as 
> there was some discussion about improving the behavior of git with 
> regard to shallow boundaries at the recent git merge conference. I 
> wonder if it would be better to refuse to cherry-pick and rebase a 
> shallow boundary commit as well. Do you know of a sensible use for 
> reverting a shallow boundary or is the new option added out of caution 
> in case someone wants it?

Very good point.

The next natural question is where the modification should go if we
want to make sure we protect these shallow boundaries not just from
revert but also from "cherry-pick", "rebase", and possibly "commit
--amend".  Is sequencer.c:do_pick_commit() the right place to cover
all of them?  If not, would it make sense to have a helper function
that different code paths to protect shallow boundaries from these
operations can call, and add the first call to that helper funcion
to the location this patch adds new code to?

Thanks.

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [PATCH] sequencer: refuse to revert shallow boundary commits by default
  2026-10-21 19:11   ` Junio C Hamano
@ 2026-10-26 10:58     ` Phillip Wood
  0 siblings, 0 replies; 4+ messages in thread
From: Phillip Wood @ 2026-10-26 10:58 UTC (permalink / raw)
  To: Junio C Hamano
  Cc: Nakul V S via GitGitGadget, git, Patrick Steinhardt, Nakul V S

On 21/10/2026 20:11, Junio C Hamano wrote:
> Phillip Wood <phillip.wood123@gmail.com> writes:
>> On 21/10/2026 11:09, Nakul V S via GitGitGadget wrote:
>>
>> Thanks for working on this, it is a useful and timely improvement as
>> there was some discussion about improving the behavior of git with
>> regard to shallow boundaries at the recent git merge conference. I
>> wonder if it would be better to refuse to cherry-pick and rebase a
>> shallow boundary commit as well. Do you know of a sensible use for
>> reverting a shallow boundary or is the new option added out of caution
>> in case someone wants it?
> 
> Very good point.
> 
> The next natural question is where the modification should go if we
> want to make sure we protect these shallow boundaries not just from
> revert but also from "cherry-pick", "rebase", and possibly "commit
> --amend".  Is sequencer.c:do_pick_commit() the right place to cover
> all of them?  If not, would it make sense to have a helper function
> that different code paths to protect shallow boundaries from these
> operations can call, and add the first call to that helper funcion
> to the location this patch adds new code to?

 From the user's point of view, I think do_pick_commit() is a bit late 
to error out. It would be nicer to refuse to add a shallow commit to the 
todo list in the first place which means we'd need a helper function 
that's called from sequencer_make_script(), make_script_with_merges(), 
single_pick() and sequencer_pick_revisions(). That is a bit of a pain 
but would make for a better UI.

Thanks

Phillip


^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2026-10-26 10:58 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-10-21 10:09 [PATCH] sequencer: refuse to revert shallow boundary commits by default Nakul V S via GitGitGadget
2026-10-21 15:20 ` Phillip Wood
2026-10-21 19:11   ` Junio C Hamano
2026-10-26 10:58     ` Phillip Wood

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox