Re: [PATCH] Introduce get_octopus_merge_bases() in commit.c

3 messages, 3 authors, 2016-06-15 · open the first message on its own page

Re: [PATCH] Introduce get_octopus_merge_bases() in commit.c

From: Junio C Hamano <hidden>
Date: 2016-06-15 22:44:43

Miklos Vajna [off-list ref] writes:
+struct commit_list *get_octopus_merge_bases(struct commit_list *in, int cleanup)
+{
+	struct commit_list *ret, *i;
+
+	ret = merge_bases_many(in->item, in->next);
+	if (cleanup)
+		for(i = in; i; i = i->next)
+			clear_commit_marks(i->item, all_flags);
+	return ret;
+}
I suspect either me or you are confused, but how is this exactly used?

The code for merge_bases_many(), at least the one I showed you a few days
ago, is not a replacement for "show-branch --merge-base", and I do not
think you would want to use it as such in the rewrite of git-merge, if you
are trying to replace this part of git-merge.sh:

        case "$#" in
        1)
                common=$(git merge-base --all $head "$@")
                ;;
        *)
                common=$(git show-branch --merge-base $head "$@")
                ;;
        esac

The purpose of merge-base-many code was to improve this line in the
git-merge-octopus.sh:

	common=$(git merge-base --all $MRC $SHA1) ||
		die "Unable to find common commit with $SHA1"

Instead of keeping a single MRC, we can compute the merge-base-many
between the SHA1 (i.e. the one we are looking at right now -- it is fed as
"one") and all the previous SHA1's we have already looked at (they become
"two's"), like this:

	common($git merge-base-many $SHA1 $SHA1_SO_FAR)

and you would have at the end of the loop:

	SHA1_SO_FAR="$SHA1_SO_FAR$SHA1 "

or something.

Re: [PATCH] Introduce get_octopus_merge_bases() in commit.c

From: Miklos Vajna <hidden>
Date: 2016-06-15 22:44:43

On Mon, Jun 09, 2008 at 04:06:12PM -0700, Junio C Hamano [off-list ref] wrote:
Miklos Vajna [off-list ref] writes:
quoted
+struct commit_list *get_octopus_merge_bases(struct commit_list *in, int cleanup)
+{
+	struct commit_list *ret, *i;
+
+	ret = merge_bases_many(in->item, in->next);
+	if (cleanup)
+		for(i = in; i; i = i->next)
+			clear_commit_marks(i->item, all_flags);
+	return ret;
+}
I suspect either me or you are confused, but how is this exactly used?

The code for merge_bases_many(), at least the one I showed you a few days
ago, is not a replacement for "show-branch --merge-base", and I do not
think you would want to use it as such in the rewrite of git-merge, if you
are trying to replace this part of git-merge.sh:

        case "$#" in
        1)
                common=$(git merge-base --all $head "$@")
                ;;
        *)
                common=$(git show-branch --merge-base $head "$@")
                ;;
        esac
Then I think I'm the one who is confused. My original
get_octopus_merge_bases() - in commit e13c0c2 - was exactly to replace
this part.

So, may I assume that the original get_octopus_merge_bases() is OK for
this purpose?
The purpose of merge-base-many code was to improve this line in the
git-merge-octopus.sh:

	common=$(git merge-base --all $MRC $SHA1) ||
		die "Unable to find common commit with $SHA1"

Instead of keeping a single MRC, we can compute the merge-base-many
between the SHA1 (i.e. the one we are looking at right now -- it is fed as
"one") and all the previous SHA1's we have already looked at (they become
"two's"), like this:

	common($git merge-base-many $SHA1 $SHA1_SO_FAR)

and you would have at the end of the loop:

	SHA1_SO_FAR="$SHA1_SO_FAR$SHA1 "

or something.
Actually I do not want to touch git-merge-octopus.sh before
builtin-merge is not ready. (I try to "do one thing and do it
well".)

Thanks.

Re: [PATCH] Introduce get_octopus_merge_bases() in commit.c

From: Johannes Schindelin <hidden>
Date: 2016-06-15 22:44:43

Hi,

On Mon, 9 Jun 2008, Junio C Hamano wrote:
Miklos Vajna [off-list ref] writes:
quoted
+struct commit_list *get_octopus_merge_bases(struct commit_list *in, int cleanup)
+{
+	struct commit_list *ret, *i;
+
+	ret = merge_bases_many(in->item, in->next);
+	if (cleanup)
+		for(i = in; i; i = i->next)
+			clear_commit_marks(i->item, all_flags);
+	return ret;
+}
I suspect either me or you are confused, but how is this exactly used?

The code for merge_bases_many(), at least the one I showed you a few days
ago, is not a replacement for "show-branch --merge-base", and I do not
think you would want to use it as such in the rewrite of git-merge, if you
are trying to replace this part of git-merge.sh:

        case "$#" in
        1)
                common=$(git merge-base --all $head "$@")
                ;;
        *)
                common=$(git show-branch --merge-base $head "$@")
                ;;
        esac

The purpose of merge-base-many code was to improve this line in the
git-merge-octopus.sh:

	common=$(git merge-base --all $MRC $SHA1) ||
		die "Unable to find common commit with $SHA1"

Instead of keeping a single MRC, we can compute the merge-base-many
between the SHA1 (i.e. the one we are looking at right now -- it is fed as
"one") and all the previous SHA1's we have already looked at (they become
"two's"), like this:

	common($git merge-base-many $SHA1 $SHA1_SO_FAR)

and you would have at the end of the loop:

	SHA1_SO_FAR="$SHA1_SO_FAR$SHA1 "

or something.
Ah!  Even I thought that merge_bases_many() was meant for the show-branch 
--merge-base case.

However, your remark about optimizing for the two-head case got me 
thinking: should we not rather use the simple algorithm Miklos proposed 
for octopus_merge_bases(), even if it is suboptimal?

Octopus is such a rare case that it is more important to have a robust, 
working code, than a fast one, right?  Especially since Octopus will be 
exercized much less often, and therefore has a higher chance of hiding a 
bug.

Ciao,
Dscho
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help