diff mbox series

update-ref: add forward command to safely fast-forward refs

Message ID 20230223011530.47477-1-ronan@rjp.ie (mailing list archive)
State New, archived
Headers show
Series update-ref: add forward command to safely fast-forward refs | expand

Commit Message

Ronan Pigott Feb. 23, 2023, 1:15 a.m. UTC
forward is an update-ref command that behaves similarly to update, but
takes an additional argument, <ancestor>, and verifies that the new
value is a descendent of ancestor before updating. This is useful for
fast-forwarding prefetched refs.
---
I've found git fetch --prefetch to be useful in my workflow for
asynchronously updating git repos, but I'd really like a nice way to
update my remote tracking branches using the latest prefetched refs
without querying the remote in case the remote is temporarily
unavailable or I'm on a plane or something. I've settled on using

$ git for-each-ref refs/prefetch/ --format='update refs/%(refname:lstrip=2) %(refname)'

to prepare a transaction for git update-ref, but it has a defect: if I
have fetched the remote more recently than the last prefetch my remote
tracking branches will be sent into the past. "forward" is a convenience
feature for update-ref that will save the trouble of looping through
each ref and finding the merge base so it can be fast-forwarded.
 Documentation/git-update-ref.txt |  5 ++++
 builtin/update-ref.c             | 49 ++++++++++++++++++++++++++++++++
 t/t1400-update-ref.sh            | 29 +++++++++++++++++++
 3 files changed, 83 insertions(+)

Comments

Junio C Hamano Feb. 23, 2023, 5:14 p.m. UTC | #1
Ronan Pigott <ronan@rjp.ie> writes:

> forward is an update-ref command that behaves similarly to update, but
> takes an additional argument, <ancestor>, and verifies that the new
> value is a descendent of ancestor before updating. This is useful for
> fast-forwarding prefetched refs.

Why is this necessary?

 * Do you expect that you may not know the ancestry relationship
   between the <newvalue> and <ancestor> values when you need to
   compute them, in order to formulate the 'forward' command?

 * Is there a case where the relationship between <newvalue> and
   <ancestor> that was fast-forward when you formulated the
   'forward' command changes by the time the 'forward' command gets
   executed?

I do not see the reason why this new command is needed, unless one
or both of the above is what you are trying to address.

For example, existing "delete SP <ref> SP <oldvalue>" is to protect
the ref you want to delete, that used to have the oldvalue back when
you created the 'delete' command, from getting deleted when somebody
else changed it from the sidelines.  We can do

	oldvalue=$(git rev-parse refs/to/be/deleted^{object})
	echo delete refs/to/be/deleted $oldvalue | ...

and let the command notice if somebody else changed
refs/to/be/deleted in between.  It is similar to the second one
between the two I cited above, to make sure that your command does
not overwrite what somebody else did.



It may make some sense if the new <ancestor> thing is to replace the
<oldvalue> thing, though.  That is, if there were a three-commit chain

	A---B---C

where the ref you are trying to update currently points at A and you
want to update it to C.  You would observe that the current value is
A, and formulate the command line:

	update ref/to/be/updated C A

with the current system, and updating the ref is allowed only when
nobody touched the ref in the meantime.  It is _conceivable_ to say
that we are OK as long as the ref points at a decendant of A
(instead of pointing exactly at A), and is an ancestor of C (the
value we are updating to), with

	fast-forward ref/to/be/fast-forwarded C A

Then somebody else _could_ update the ref to B from the sideline,
but we notice that it is a descendant of A and an ancestor of C, and
we are still allowed to update it to C.

Even in that case, I am not sure how useful it would be, but at
least that use case I can see why it may make sense.

If you can write <newvalue> and <ancestor> on the command line, you
certainly should be able to compute "git merge-base".  And because
the commits are immutable, it won't change in the middle.

So, I am not very impressed.  Unless I am missing something, this
does not seem to be adding anything we cannot already do.

I wonder if

    git fetch . "+refs/prefetch/remotes/origin/*:refs/remotes/origin/*"

(with or without the '+' prefix, depending) what you are really
going after, though.
Ronan Pigott Feb. 23, 2023, 6:01 p.m. UTC | #2
> Unless I am missing something, this
> does not seem to be adding anything we cannot already do.

Correct. In my case it replaces:

MERGE_BASE=$(git merge-base refs/remotes/origin/master refs/prefetch/remotes/origin/master)
update refs/remotes/origin/master refs/prefetch/remotes/origin/master $MERGE_BASE

I just thought the "forward" form was a little more convenient since it could output it all in one for-each-ref invocation. I mean, I could always open .git/refs/* in vim but that doesn't mean I want to...

Originally I left it as a three argument command but figured it might as well be possible to update to the descendent of any given ref, so I separated <ancestor> from <refname>.
 
> I wonder if
> 
> git fetch . "+refs/prefetch/remotes/origin/*:refs/remotes/origin/*"
> 
> (with or without the '+' prefix, depending) what you are really
> going after, though.

Oh, that's perfect! I didn't consider that it was possible to "fetch" from the local repo. I think the refspec 'refs/prefetch/remotes/origin/*:refs/remotes/origin/*' does what I want — fast-forwarding my remote tracking branches to their latest prefetches.

In that case, if you don't feel the forward feature is useful feel free to drop it.

Thanks,
Ronan
diff mbox series

Patch

diff --git a/Documentation/git-update-ref.txt b/Documentation/git-update-ref.txt
index 48b6683071e6..138a478b437f 100644
--- a/Documentation/git-update-ref.txt
+++ b/Documentation/git-update-ref.txt
@@ -66,6 +66,7 @@  performs all modifications together.  Specify commands of the form:
 	delete SP <ref> [SP <oldvalue>] LF
 	verify SP <ref> [SP <oldvalue>] LF
 	option SP <opt> LF
+	forward SP <ref> SP <newvalue> SP <ancestor> [SP <oldvalue>] LF
 	start LF
 	prepare LF
 	commit LF
@@ -105,6 +106,10 @@  update::
 	after the update and/or a zero <oldvalue> to make sure the
 	ref does not exist before the update.
 
+forward::
+	Set <ref> to <newvalue> after verifying <oldvalue> only if
+	<newvalue> is a descendent of <ancestor>.
+
 create::
 	Create <ref> with <newvalue> after verifying it does not
 	exist.  The given <newvalue> may not be zero.
diff --git a/builtin/update-ref.c b/builtin/update-ref.c
index a84e7b47a206..cdffa52b2d93 100644
--- a/builtin/update-ref.c
+++ b/builtin/update-ref.c
@@ -1,5 +1,6 @@ 
 #include "cache.h"
 #include "config.h"
+#include "commit-reach.h"
 #include "refs.h"
 #include "builtin.h"
 #include "parse-options.h"
@@ -86,6 +87,11 @@  static char *parse_refname(const char **next)
  */
 #define PARSE_SHA1_ALLOW_EMPTY 0x02
 
+/*
+ * The value being parsed is <ancestor>, for forward updates.
+ */
+#define PARSE_SHA1_ANCESTOR 0x04
+
 /*
  * Parse an argument separator followed by the next argument, if any.
  * If there is an argument, convert it to a SHA-1, write it to sha1,
@@ -157,12 +163,16 @@  static int parse_next_oid(const char **next, const char *end,
  invalid:
 	die(flags & PARSE_SHA1_OLD ?
 	    "%s %s: invalid <oldvalue>: %s" :
+	    flags & PARSE_SHA1_ANCESTOR ?
+	    "%s %s: invalid <ancestor>: %s" :
 	    "%s %s: invalid <newvalue>: %s",
 	    command, refname, arg.buf);
 
  eof:
 	die(flags & PARSE_SHA1_OLD ?
 	    "%s %s: unexpected end of input when reading <oldvalue>" :
+	    flags & PARSE_SHA1_ANCESTOR ?
+	    "%s %s: unexpected end of input when reading <ancestor>" :
 	    "%s %s: unexpected end of input when reading <newvalue>",
 	    command, refname);
 }
@@ -211,6 +221,44 @@  static void parse_cmd_update(struct ref_transaction *transaction,
 	strbuf_release(&err);
 }
 
+static void parse_cmd_forward(struct ref_transaction *transaction,
+			     const char *next, const char *end)
+{
+	struct strbuf err = STRBUF_INIT;
+	char *refname;
+	struct object_id new_oid, anc_oid, old_oid;
+	int have_old;
+
+	refname = parse_refname(&next);
+	if (!refname)
+		die("forward: missing <ref>");
+
+	if (parse_next_oid(&next, end, &new_oid, "forward", refname, 0))
+		die("forward %s: missing <newvalue>", refname);
+
+	if (parse_next_oid(&next, end, &anc_oid, "forward", refname, PARSE_SHA1_ANCESTOR))
+		die("forward %s: missing <ancestor>", refname);
+
+	have_old = !parse_next_oid(&next, end, &old_oid, "forward", refname,
+				   PARSE_SHA1_OLD);
+
+	if (*next != line_termination)
+		die("forward %s: extra input: %s", refname, next);
+
+	if (!ref_newer(&new_oid, &anc_oid))
+		die("forward %s: not a fast-forward", refname);
+
+	if (ref_transaction_update(transaction, refname,
+				   &new_oid, have_old ? &old_oid : NULL,
+				   update_flags | create_reflog_flag,
+				   msg, &err))
+		die("%s", err.buf);
+
+	update_flags = default_flags;
+	free(refname);
+	strbuf_release(&err);
+}
+
 static void parse_cmd_create(struct ref_transaction *transaction,
 			     const char *next, const char *end)
 {
@@ -378,6 +426,7 @@  static const struct parse_cmd {
 	enum update_refs_state state;
 } command[] = {
 	{ "update",  parse_cmd_update,  3, UPDATE_REFS_OPEN },
+	{ "forward", parse_cmd_forward, 4, UPDATE_REFS_OPEN },
 	{ "create",  parse_cmd_create,  2, UPDATE_REFS_OPEN },
 	{ "delete",  parse_cmd_delete,  2, UPDATE_REFS_OPEN },
 	{ "verify",  parse_cmd_verify,  2, UPDATE_REFS_OPEN },
diff --git a/t/t1400-update-ref.sh b/t/t1400-update-ref.sh
index cf58cf025cd2..6ac90aca1917 100755
--- a/t/t1400-update-ref.sh
+++ b/t/t1400-update-ref.sh
@@ -785,6 +785,35 @@  test_expect_success 'stdin update ref works with right old value' '
 	test_cmp expect actual
 '
 
+test_expect_success 'stdin forward ref fails with wrong ancestor value' '
+	echo "forward $c $m~1 $m" >stdin &&
+	test_must_fail git update-ref --stdin <stdin 2>err &&
+	grep "forward $c: not a fast-forward" err &&
+	test_must_fail git rev-parse --verify -q $c
+'
+
+test_expect_success 'stdin forward ref fails with bad ancestor value' '
+	echo "forward $c $m does-not-exist" >stdin &&
+	test_must_fail git update-ref --stdin <stdin 2>err &&
+	grep "fatal: forward $c: invalid <ancestor>: does-not-exist" err &&
+	test_must_fail git rev-parse --verify -q $c
+'
+
+test_expect_success 'stdin forward ref fails with bad old value' '
+	echo "forward $c $m $m~1 does-not-exist" >stdin &&
+	test_must_fail git update-ref --stdin <stdin 2>err &&
+	grep "fatal: forward $c: invalid <oldvalue>: does-not-exist" err &&
+	test_must_fail git rev-parse --verify -q $c
+'
+
+test_expect_success 'stdin forward ref works with right ancestor value' '
+	echo "forward $b $m~1 $m~2" >stdin &&
+	git update-ref --stdin <stdin &&
+	git rev-parse $m~1 >expect &&
+	git rev-parse $b >actual &&
+	test_cmp expect actual
+'
+
 test_expect_success 'stdin delete ref fails with wrong old value' '
 	echo "delete $a $m~1" >stdin &&
 	test_must_fail git update-ref --stdin <stdin 2>err &&