Skip to contentSkip to conversation

commit-graph: when incompatible with graphs, indicate why

PATCH In review Johannes Schindelin via GitGitGadget 4 participants 4 messages Feb 11, 2021 – Feb 25, 2021

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 commit-graph: when incompatible with graphs, indicate why

From: Johannes Schindelin <[email protected]>

When gc.writeCommitGraph = true, it is possible that the commit-graph is _still_ not written: replace objects, grafts and shallow repositories are incompatible with the commit-graph feature.

Under such circumstances, we need to indicate to the user why the commit-graph was not written instead of staying silent about it.

---
Be clear why commit-graph was skipped

After repairing my local checkout
[https://github.com/gitgitgadget/git/pull/873], I was puzzled that the commit-graph file was not written. Turns out that I still had almost a dozen replace objects. But I only found out that they were blocking the commit-graph when I stepped through git gc in a debugger. This is my attempt to make it more straight-forward to recover from similar situations in the future.

Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-875%2Fdscho%2Fwarn-if-commit-graph-is-skipped-v1
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-875/dscho/warn-if-commit-graph-is-skipped-v1
Pull-Request: https://github.com/gitgitgadget/git/pull/875

 commit-graph.c | 14 +++++++++++---
 1 file changed, 11 insertions(+), 3 deletions(-)
commit-graph.c +11 -3
index 65410602714e..9ad176fa7c8e 100644
@@ -205,16 +205,24 @@ static int commit_graph_compatible(struct repository *r)
 
 	if (read_replace_refs) {
 		prepare_replace_object(r);
-		if (hashmap_get_size(&r->objects->replace_map->map))
+		if (hashmap_get_size(&r->objects->replace_map->map)) {
+			warning(_("repository contains replace objects; "
+			       "skipping commit-graph"));
 			return 0;
+		}
 	}
 
 	prepare_commit_graft(r);
 	if (r->parsed_objects &&
-	    (r->parsed_objects->grafts_nr || r->parsed_objects->substituted_parent))
+	    (r->parsed_objects->grafts_nr || r->parsed_objects->substituted_parent)) {
+		warning(_("repository contains (deprecated) grafts; "
+		       "skipping commit-graph"));
 		return 0;
-	if (is_repository_shallow(r))
+	}
+	if (is_repository_shallow(r)) {
+		warning(_("repository is shallow; skipping commit-graph"));
 		return 0;
+	}
 
 	return 1;
 }

base-commit: f9f2520108bab26a750bcbb00518dc27672cf0a2

Signature · 1 line
gitgitgadget
#1 lore raw
Quoted 8 lines · Johannes Schindelin via GitGitGadget

From: Johannes Schindelin <[email protected]>

When gc.writeCommitGraph = true, it is possible that the commit-graph is _still_ not written: replace objects, grafts and shallow repositories are incompatible with the commit-graph feature.

Under such circumstances, we need to indicate to the user why the commit-graph was not written instead of staying silent about it.

This feedback is valuable for these corner cases, especially now that the commit-graph is getting less and less "optional" as time goes on.

Quoted 3 lines
commit-graph.c +3 -0
+		if (hashmap_get_size(&r->objects->replace_map->map)) {
+			warning(_("repository contains replace objects; "
+			       "skipping commit-graph"));

...

commit-graph.c static int commit_graph_compatible(struct repository *r) · line 223

Quoted 1 line
commit-graph.c +1 -0
+		warning(_("repository is shallow; skipping commit-graph"));

These warnings make sense to me.

Thanks,
-Stolee

#2 lore raw
Quoted 14 lines · Derrick Stolee
Quoted 8 lines · Johannes Schindelin via GitGitGadget

From: Johannes Schindelin <[email protected]>

When gc.writeCommitGraph = true, it is possible that the commit-graph is _still_ not written: replace objects, grafts and shallow repositories are incompatible with the commit-graph feature.

Under such circumstances, we need to indicate to the user why the commit-graph was not written instead of staying silent about it.

This feedback is valuable for these corner cases, especially now that the commit-graph is getting less and less "optional" as time goes on.

Quoted 7 lines
Quoted 3 lines
commit-graph.c +3 -0
+		if (hashmap_get_size(&r->objects->replace_map->map)) {
+			warning(_("repository contains replace objects; "
+			       "skipping commit-graph"));

...

commit-graph.c static int commit_graph_compatible(struct repository *r) · line 223

Quoted 1 line
commit-graph.c +1 -0
+		warning(_("repository is shallow; skipping commit-graph"));

These warnings make sense to me.

I take that as your Acked-by.

Thanks, both.

#3 lore raw

noisy "warning: repository contains (deprecated) grafts; skipping commit-graph" warnings

In reply to Johannes Schindelin via GitGitGadget
Quoted 8 lines · Johannes Schindelin via GitGitGadget

From: Johannes Schindelin <[email protected]>

When gc.writeCommitGraph = true, it is possible that the commit-graph is _still_ not written: replace objects, grafts and shallow repositories are incompatible with the commit-graph feature.

Under such circumstances, we need to indicate to the user why the commit-graph was not written instead of staying silent about it.

This change really needs to be changed in some way or other, but unfortunately the commit message has little/no information about when these warnings are expected, and no tests.

I'm assuming you were targeting the write_commit_graph() caller of commit_graph_compatible(). In that case this somewhat makes sense I guess.

But we also call this at a distance when simply checking if we can lookup things in the commit-graph, observe this on the current git master:

    $ git clone --depth 1 --no-tags --single-branch --branch master https://github.com/git/git.git /tmp/git.git
    [...]
    warning: repository contains (deprecated) grafts; skipping commit-graph
    warning: repository contains (deprecated) grafts; skipping commit-graph

In that case we reach this via "parse_object()" in the "clone" process, and print it twice because our rev-list child also spews the warning at us.

Perhaps a better approach here is to pass down some flag and e.g. write it only from "git gc" and friends?

Quoted 53 lines

---
Be clear why commit-graph was skipped

After repairing my local checkout
[https://github.com/gitgitgadget/git/pull/873], I was puzzled that the commit-graph file was not written. Turns out that I still had almost a dozen replace objects. But I only found out that they were blocking the commit-graph when I stepped through git gc in a debugger. This is my attempt to make it more straight-forward to recover from similar situations in the future.

Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-875%2Fdscho%2Fwarn-if-commit-graph-is-skipped-v1
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-875/dscho/warn-if-commit-graph-is-skipped-v1
Pull-Request: https://github.com/gitgitgadget/git/pull/875

 commit-graph.c | 14 +++++++++++---
 1 file changed, 11 insertions(+), 3 deletions(-)
commit-graph.c +11 -3
index 65410602714e..9ad176fa7c8e 100644
@@ -205,16 +205,24 @@ static int commit_graph_compatible(struct repository *r)
 
 	if (read_replace_refs) {
 		prepare_replace_object(r);
-		if (hashmap_get_size(&r->objects->replace_map->map))
+		if (hashmap_get_size(&r->objects->replace_map->map)) {
+			warning(_("repository contains replace objects; "
+			       "skipping commit-graph"));
 			return 0;
+		}
 	}
 
 	prepare_commit_graft(r);
 	if (r->parsed_objects &&
-	    (r->parsed_objects->grafts_nr || r->parsed_objects->substituted_parent))
+	    (r->parsed_objects->grafts_nr || r->parsed_objects->substituted_parent)) {
+		warning(_("repository contains (deprecated) grafts; "
+		       "skipping commit-graph"));
 		return 0;
-	if (is_repository_shallow(r))
+	}
+	if (is_repository_shallow(r)) {
+		warning(_("repository is shallow; skipping commit-graph"));
 		return 0;
+	}
 
 	return 1;
 }

base-commit: f9f2520108bab26a750bcbb00518dc27672cf0a2