Read git metadata from the Repos API for Git CLI folders

.nextchanges/bundles/git-folder-metadata.md

@@ -0,0 +1 @@
+Fixed `bundle.git.branch`, `bundle.git.commit`, and `bundle.git.origin_url` being empty for bundles deployed from a workspace Git folder that has Git CLI access. The workspace API does not report git metadata for those folders, so it is now read from the Repos API instead.

acceptance/bundle/debug/fetch-repository-info-repos-error/databricks.yml

@@ -0,0 +1,2 @@
+bundle:
+  name: test-bundle

acceptance/bundle/debug/fetch-repository-info-repos-error/output.txt

@@ -0,0 +1,9 @@
+
+>>> [CLI] bundle debug fetch-repository-info --path /Workspace/Users/[USERNAME]/test-folder-[UNIQUE_NAME]
+Warn: failed to load git info for repo [NUMID]: Fault injected by test.
+{
+  "worktree_root": "/Workspace/Users/[USERNAME]/test-folder-[UNIQUE_NAME]",
+  "current_branch": "",
+  "latest_commit": "",
+  "origin_url": ""
+}

acceptance/bundle/debug/fetch-repository-info-repos-error/script

@@ -0,0 +1,16 @@
+# A Git folder outside /Repos has Git CLI access, so the workspace API reports no
+# git metadata for it and it is read from the Repos API instead. Make that read
+# fail: the worktree root must still be reported, with the metadata left empty
+# rather than failing the command.
+#
+# Registered before MSYS_NO_PATHCONV is exported, which on Windows would rewrite
+# the path Git Bash passes for this helper's own script.
+fault.py "GET /api/2.0/repos/*" 403 0 1
+
+# Keep Git Bash from rewriting the /Workspace path passed to --path.
+export MSYS_NO_PATHCONV=1
+
+folder=/Workspace/Users/$CURRENT_USER_NAME/test-folder-$UNIQUE_NAME
+$CLI repos create https://github.com/databricks/bundle-examples gitHub --path "$folder" > /dev/null
+
+trace $CLI bundle debug fetch-repository-info --path "$folder"

acceptance/bundle/debug/fetch-repository-info-repos-error/test.toml

@@ -0,0 +1,5 @@
+# Local only: the failure is injected into the testserver.
+Cloud = false
+
+# This exercises git.FetchRepositoryInfo, which is independent of the deploy engine.
+EnvMatrix.DATABRICKS_BUNDLE_ENGINE = ["direct"]

acceptance/bundle/debug/fetch-repository-info/databricks.yml

@@ -0,0 +1,2 @@
+bundle:
+  name: test-bundle

acceptance/bundle/debug/fetch-repository-info/output.txt

@@ -0,0 +1,40 @@
+
+=== the root of a Git folder under /Repos
+
+>>> [CLI] bundle debug fetch-repository-info --path /Workspace/Repos/[USERNAME]/test-repo-[UNIQUE_NAME]
+{
+  "worktree_root": "/Workspace/Repos/[USERNAME]/test-repo-[UNIQUE_NAME]",
+  "current_branch": "main",
+  "latest_commit": "[COMMIT_SHA]",
+  "origin_url": "https://github.com/databricks/bundle-examples"
+}
+
+=== a subdirectory of it, which resolves to the same root
+
+>>> [CLI] bundle debug fetch-repository-info --path /Workspace/Repos/[USERNAME]/test-repo-[UNIQUE_NAME]/subdir
+{
+  "worktree_root": "/Workspace/Repos/[USERNAME]/test-repo-[UNIQUE_NAME]",
+  "current_branch": "main",
+  "latest_commit": "[COMMIT_SHA]",
+  "origin_url": "https://github.com/databricks/bundle-examples"
+}
+
+=== the root of a Git folder outside /Repos
+
+>>> [CLI] bundle debug fetch-repository-info --path /Workspace/Users/[USERNAME]/test-folder-[UNIQUE_NAME]
+{
+  "worktree_root": "/Workspace/Users/[USERNAME]/test-folder-[UNIQUE_NAME]",
+  "current_branch": "main",
+  "latest_commit": "[COMMIT_SHA]",
+  "origin_url": "https://github.com/databricks/bundle-examples"
+}
+
+=== a path in no Git folder at all
+
+>>> [CLI] bundle debug fetch-repository-info --path /Workspace/Users/[USERNAME]
+{
+  "worktree_root": "",
+  "current_branch": "",
+  "latest_commit": "",
+  "origin_url": ""
+}

acceptance/bundle/debug/fetch-repository-info/script

@@ -0,0 +1,33 @@
+# The fake workspace reports "main", which is this repo's default branch too.
+url=https://github.com/databricks/bundle-examples
+
+# A Git folder under /Repos and one outside it. Once a Git folder has Git CLI
+# access the workspace API reports no git metadata for it, so both resolve
+# through the Repos API; the resolved metadata must be identical either way.
+repo=/Workspace/Repos/$CURRENT_USER_NAME/test-repo-$UNIQUE_NAME
+folder=/Workspace/Users/$CURRENT_USER_NAME/test-folder-$UNIQUE_NAME
+
+$CLI repos create $url gitHub --path "$repo" > /dev/null
+$CLI repos create $url gitHub --path "$folder" > /dev/null
+
+# A directory inside the Git folder, so the subdirectory case below resolves a
+# path that exists rather than a missing one.
+$CLI workspace mkdirs "$repo/subdir"
+
+cleanup() {
+    $CLI repos delete "$repo" 2>>LOG.cleanup
+    $CLI repos delete "$folder" 2>>LOG.cleanup
+}
+trap cleanup EXIT
+
+title "the root of a Git folder under /Repos\n"
+trace $CLI bundle debug fetch-repository-info --path "$repo"
+
+title "a subdirectory of it, which resolves to the same root\n"
+trace $CLI bundle debug fetch-repository-info --path "$repo/subdir"
+
+title "the root of a Git folder outside /Repos\n"
+trace $CLI bundle debug fetch-repository-info --path "$folder"
+
+title "a path in no Git folder at all\n"
+trace $CLI bundle debug fetch-repository-info --path "/Workspace/Users/$CURRENT_USER_NAME"

acceptance/bundle/debug/fetch-repository-info/test.toml

@@ -0,0 +1,13 @@
+Cloud = true
+
+# This exercises git.FetchRepositoryInfo, which is independent of the deploy engine.
+EnvMatrix.DATABRICKS_BUNDLE_ENGINE = ["direct"]
+
+# Keep Git Bash from rewriting the /Workspace paths passed to --path.
+Env.MSYS_NO_PATHCONV = "1"
+
+# The commit differs between the fake workspace and a real clone of the repo, so
+# the golden matches both.
+[[Repls]]
+Old = '\b[0-9a-f]{40}\b'
+New = '[COMMIT_SHA]'

acceptance/workspace/repos/create_with_provider/output.txt

@@ -5,6 +5,7 @@
 >>> [CLI] repos get [NUMID] -o json
 {
   "branch": "main",
+  "head_commit_id": "0a1b2c3d4e5f6a7b8e7dadd73e50a69d8ba47d8f",
   "id": [NUMID],
   "path": "/Repos/me@databricks.com/test-repo",
   "provider": "gitHub",
@@ -15,6 +16,7 @@
 >>> [CLI] repos get /Repos/me@databricks.com/test-repo -o json
 {
   "branch": "main",
+  "head_commit_id": "0a1b2c3d4e5f6a7b8e7dadd73e50a69d8ba47d8f",
   "id": [NUMID],
   "path": "/Repos/me@databricks.com/test-repo",
   "provider": "gitHub",

acceptance/workspace/repos/create_without_provider/output.txt

@@ -2,6 +2,7 @@
 >>> [CLI] repos create https://github.com/databricks/databricks-empty-ide-project.git --path /Repos/me@databricks.com/test-repo
 {
   "branch": "main",
+  "head_commit_id": "0a1b2c3d4e5f6a7b8e7dadd73e50a69d8ba47d8f",
   "id": [NUMID],
   "path": "/Repos/me@databricks.com/test-repo",
   "provider": "gitHub",

acceptance/workspace/repos/delete_by_path/output.txt

@@ -5,6 +5,7 @@
 >>> [CLI] repos get /Repos/me@databricks.com/test-repo -o json
 {
   "branch": "main",
+  "head_commit_id": "0a1b2c3d4e5f6a7b8e7dadd73e50a69d8ba47d8f",
   "id": [NUMID],
   "path": "/Repos/me@databricks.com/test-repo",
   "provider": "gitHub",

acceptance/workspace/repos/git_cli_folder/output.txt

@@ -6,6 +6,7 @@
 >>> [CLI] repos get /Workspace/Users/me@databricks.com/test-repo -o json
 {
   "branch": "main",
+  "head_commit_id": "0a1b2c3d4e5f6a7b8e7dadd73e50a69d8ba47d8f",
   "id": [NUMID],
   "path": "/Workspace/Users/me@databricks.com/test-repo",
   "provider": "gitHub",

acceptance/workspace/repos/update/output.txt

@@ -6,6 +6,7 @@
 >>> [CLI] repos get [NUMID] -o json
 {
   "branch": "update-by-id",
+  "head_commit_id": "0a1b2c3d4e5f6a7b8e7dadd73e50a69d8ba47d8f",
   "id": [NUMID],
   "path": "/Repos/me@databricks.com/test-repo",
   "provider": "gitHub",
@@ -17,6 +18,7 @@
 >>> [CLI] repos get [NUMID] -o json
 {
   "branch": "update-by-path",
+  "head_commit_id": "0a1b2c3d4e5f6a7b8e7dadd73e50a69d8ba47d8f",
   "id": [NUMID],
   "path": "/Repos/me@databricks.com/test-repo",
   "provider": "gitHub",

cmd/bundle/debug.go

@@ -18,5 +18,6 @@ func newDebugCommand() *cobra.Command {
 	cmd.AddCommand(debug.NewStatesCommand())
 	cmd.AddCommand(debug.NewRenderTemplateSchemaCommand())
 	cmd.AddCommand(debug.NewListTargetsCommand())
+	cmd.AddCommand(debug.NewFetchRepositoryInfoCommand())
 	return cmd
 }

cmd/bundle/debug/fetch_repository_info.go

@@ -0,0 +1,63 @@
+package debug
+
+import (
+	"encoding/json"
+
+	"github.com/databricks/cli/cmd/root"
+	"github.com/databricks/cli/libs/cmdctx"
+	"github.com/databricks/cli/libs/git"
+	"github.com/spf13/cobra"
+)
+
+// repositoryInfoOutput mirrors git.RepositoryInfo, which has no JSON tags.
+type repositoryInfoOutput struct {
+	WorktreeRoot  string `json:"worktree_root"`
+	CurrentBranch string `json:"current_branch"`
+	LatestCommit  string `json:"latest_commit"`
+	OriginURL     string `json:"origin_url"`
+}
+
+// NewFetchRepositoryInfoCommand returns a command that reports what
+// [git.FetchRepositoryInfoAPI] resolves for a path. It exists so that function
+// can be exercised through the CLI on both a fake and a real workspace; nothing
+// in the product calls it.
+//
+// It always reads the workspace API, which [git.FetchRepositoryInfo] only does on
+// a Databricks Runtime. That cannot be detected off-cluster, and reading .git is
+// covered by the TestFetchRepositoryInfoDotGit integration tests.
+func NewFetchRepositoryInfoCommand() *cobra.Command {
+	cmd := &cobra.Command{
+		Use:    "fetch-repository-info",
+		Short:  "Report the git metadata the workspace API resolves for a path",
+		Args:   root.NoArgs,
+		Hidden: true,
+	}
+
+	var path string
+	cmd.Flags().StringVar(&path, "path", ".", "Path to resolve git metadata for")
+
+	cmd.PreRunE = root.MustWorkspaceClient
+	cmd.RunE = func(cmd *cobra.Command, args []string) error {
+		ctx := cmd.Context()
+
+		info, err := git.FetchRepositoryInfoAPI(ctx, path, cmdctx.WorkspaceClient(ctx))
+		if err != nil {
+			return err
+		}
+
+		buf, err := json.MarshalIndent(repositoryInfoOutput{
+			WorktreeRoot:  info.WorktreeRoot,
+			CurrentBranch: info.CurrentBranch,
+			LatestCommit:  info.LatestCommit,
+			OriginURL:     info.OriginURL,
+		}, "", "  ")
+		if err != nil {
+			return err
+		}
+
+		_, err = cmd.OutOrStdout().Write(append(buf, '\n'))
+		return err
+	}
+
+	return cmd
+}

integration/libs/git/git_fetch_test.go

@@ -9,7 +9,6 @@ import (
 	"testing"
 
 	"github.com/databricks/cli/integration/internal/acc"
-	"github.com/databricks/cli/internal/testutil"
 	"github.com/databricks/cli/libs/dbr"
 	"github.com/databricks/cli/libs/git"
 	"github.com/stretchr/testify/assert"
@@ -48,13 +47,6 @@ func ensureWorkspacePrefix(root string) string {
 }
 
 func TestFetchRepositoryInfoAPI_FromRepo(t *testing.T) {
-	// On GCP the Repos API returns no branch, commit, or origin URL for a
-	// freshly-cloned repo, so FetchRepositoryInfo comes back empty and the
-	// assertions in assertFullGitInfo fail.
-	if testutil.GetCloud(t) == testutil.GCP {
-		t.Skip("Skipping on GCP: Repos API does not return git metadata")
-	}
-
 	ctx, wt := acc.WorkspaceTest(t)
 	targetPath := ensureWorkspacePrefix(acc.TemporaryRepo(wt, examplesRepoUrl))
 

libs/git/info.go

@@ -16,6 +16,7 @@ import (
 	"github.com/databricks/databricks-sdk-go"
 	"github.com/databricks/databricks-sdk-go/apierr"
 	"github.com/databricks/databricks-sdk-go/client"
+	"github.com/databricks/databricks-sdk-go/service/workspace"
 )
 
 type RepositoryInfo struct {
@@ -31,6 +32,7 @@ type RepositoryInfo struct {
 type gitInfo struct {
 	Branch       string `json:"branch"`
 	HeadCommitID string `json:"head_commit_id"`
+	ID           int64  `json:"id"`
 	Path         string `json:"path"`
 	URL          string `json:"url"`
 }
@@ -50,7 +52,7 @@ func FetchRepositoryInfo(ctx context.Context, path string, w *databricks.Workspa
 	var info RepositoryInfo
 	var err error
 	if strings.HasPrefix(path, "/Workspace/") && dbr.RunsOnRuntime(ctx) {
-		info, err = fetchRepositoryInfoAPI(ctx, path, w)
+		info, err = FetchRepositoryInfoAPI(ctx, path, w)
 	} else {
 		info, err = fetchRepositoryInfoDotGit(ctx, path)
 	}
@@ -65,7 +67,14 @@ func FetchRepositoryInfo(ctx context.Context, path string, w *databricks.Workspa
 	return info, err
 }
 
-func fetchRepositoryInfoAPI(ctx context.Context, path string, w *databricks.WorkspaceClient) (RepositoryInfo, error) {
+// FetchRepositoryInfoAPI reads the metadata from the workspace API, which is what
+// FetchRepositoryInfo does on a Databricks Runtime. Exported so that
+// `bundle debug fetch-repository-info --workspace-api` can reach this path
+// off-runtime; the product calls it through FetchRepositoryInfo.
+//
+// A path that does not exist is reported as fs.ErrNotExist; callers that treat
+// that as "no repository here" must normalize it the way FetchRepositoryInfo does.
+func FetchRepositoryInfoAPI(ctx context.Context, path string, w *databricks.WorkspaceClient) (RepositoryInfo, error) {
 	result := RepositoryInfo{}
 
 	apiClient, err := client.New(w.Config)
@@ -108,6 +117,21 @@ func fetchRepositoryInfoAPI(ctx context.Context, path string, w *databricks.Work
 		result.LatestCommit = gi.HeadCommitID
 		result.CurrentBranch = gi.Branch
 		result.WorktreeRoot = fixedPath
+
+		// A Git folder with Git CLI access does not store the git metadata on the
+		// workspace object, so get-status returns only the id and path for it. The
+		// Repos API still has the metadata, and git_info.id identifies the Git
+		// folder root even when the queried path is a subdirectory of it.
+		if gi.ID != 0 && gi.Branch == "" && gi.HeadCommitID == "" && gi.URL == "" {
+			repo, err := w.Repos.Get(ctx, workspace.GetRepoRequest{RepoId: gi.ID})
+			if err != nil {
+				log.Warnf(ctx, "failed to load git info for repo %d: %s", gi.ID, err)
+			} else {
+				result.OriginURL = repo.Url
+				result.LatestCommit = repo.HeadCommitId
+				result.CurrentBranch = repo.Branch
+			}
+		}
 	} else {
 		log.Infof(ctx, "Failed to load git info from %s", apiEndpoint)
 	}