Narrow the merge preview's tree of a module that names its paths (issue 1758, S4, ADR 0499 §2)

ModuleTrees reads each named path's object with one git ls-tree and answers their
named tree, so the preview gives the value the indexer gives; a named path absent at
the commit fails by name rather than pinning a source that does not exist.

go/buildstep/trees.go

@@ -6,9 +6,12 @@ import (
 	"encoding/hex"
 	"fmt"
 	"os"
+	"path"
 	"path/filepath"
 	"strings"
 
+	"git.novox.be/novox/mesh-sdk/go/catalogue"
+	"git.novox.be/novox/mesh-sdk/go/delivery"
 	"git.novox.be/novox/mesh-sdk/go/forgekey"
 )
 
@@ -27,7 +30,8 @@ type ModuleTree struct {
 var lockFiles = []string{"go.sum", "package-lock.json"}
 
 // ModuleTrees are the trees, in commit, of every directory of the workspace dir that holds a module.json, by directory
-// (the repository's root is "."). The workspace is at commit, so the manifests and the lock files are read from it. env is
+// (the repository's root is "."); for a module whose every artifact names its paths, the named tree of those paths and
+// their objects (novox/hq ADR 0499 clause 2). The workspace is at commit, so the manifests and the lock files are read from it. env is
 // the environment git runs with (nil for none). Both build modules' fetch steps answer it for a merge.
 func ModuleTrees(ctx context.Context, run forgekey.Runner, env []string, dir, commit string) (map[string]ModuleTree, error) {
 	r := run(ctx, env, "git", "-C", dir, "ls-files", "--", "module.json", "*/module.json")
@@ -54,6 +58,9 @@ func ModuleTrees(ctx context.Context, run forgekey.Runner, env []string, dir, co
 			return nil, err
 		}
 		mt.Manifest = sha256Hex(raw)
+		if err := narrow(ctx, run, env, dir, commit, mdir, raw, &mt); err != nil {
+			return nil, err
+		}
 		for _, lock := range lockFiles {
 			if raw, err := os.ReadFile(filepath.Join(dir, filepath.FromSlash(mdir), lock)); err == nil {
 				if mt.Locks == nil {
@@ -67,4 +74,50 @@ func ModuleTrees(ctx context.Context, run forgekey.Runner, env []string, dir, co
 	return trees, nil
 }
 
+// narrow sets mt's Paths and its named Tree when the module's manifest names paths on every artifact (novox/hq ADR 0499
+// clause 2): one `git ls-tree -z` gives each named path's object at commit. The merge preview reads the manifest here, as
+// the indexer does, and no other build step reads paths from one (ADR 0495 clause 1). A manifest that does not parse
+// is not narrowed: the indexer gives it none. A named path absent at commit fails, naming the module and the path.
+func narrow(ctx context.Context, run forgekey.Runner, env []string, dir, commit, mdir string, raw []byte, mt *ModuleTree) error {
+	m, err := catalogue.ParseManifest(raw)
+	if err != nil {
+		return nil
+	}
+	paths, every := m.NamedPaths()
+	if !every {
+		return nil
+	}
+	full := make([]string, len(paths))
+	for i, p := range paths {
+		full[i] = path.Join(mdir, p)
+	}
+	r := run(ctx, env, "git", append([]string{"-C", dir, "ls-tree", "-z", commit, "--"}, full...)...)
+	if r.Status != 0 || r.Err != "" {
+		return fmt.Errorf("git ls-tree %s of the module %s: %s", commit, mdir, firstLine(r.Err+" "+r.Stderr))
+	}
+	objects := map[string]string{}
+	for _, l := range strings.Split(r.Stdout, "\x00") {
+		meta, name, ok := strings.Cut(l, "\t")
+		fields := strings.Fields(meta)
+		if !ok || len(fields) != 3 {
+			continue
+		}
+		objects[name] = fields[2]
+	}
+	mt.Paths = map[string]string{}
+	for i, p := range paths {
+		object := objects[full[i]]
+		if object == "" {
+			return fmt.Errorf("the module %s names the path %q, which is not in %s", mdir, p, commit)
+		}
+		mt.Paths[p] = object
+	}
+	named, err := delivery.NamedTree(mt.Manifest, mt.Paths)
+	if err != nil {
+		return fmt.Errorf("the module %s: %w", mdir, err)
+	}
+	mt.Tree = named
+	return nil
+}
+
 func sha256Hex(b []byte) string { h := sha256.Sum256(b); return hex.EncodeToString(h[:]) }

go/buildstep/trees_named_test.go

@@ -0,0 +1,75 @@
+package buildstep
+
+import (
+	"context"
+	"strings"
+	"testing"
+
+	"git.novox.be/novox/mesh-sdk/go/delivery"
+	"git.novox.be/novox/mesh-sdk/go/forgekey"
+)
+
+// A module whose every artifact names its paths takes its tree from them (novox/hq ADR 0499 clause 2): ModuleTrees asks
+// git for each named path's object at the commit, keeps them in Paths and answers their delivery.NamedTree as Tree; a
+// module naming none keeps its directory's tree, as before.
+func TestModuleTreesNarrowsAModuleWhoseArtifactsNameTheirPaths(t *testing.T) {
+	root := t.TempDir()
+	sdk := `{"module":"mesh-sdk","version":"1","build":{"artifacts":[{"name":"lib","kind":"package","language":"typescript","paths":["src","package.json"]}]}}`
+	writeTreeFile(t, root, "module.json", sdk)
+	writeTreeFile(t, root, "modules/b/module.json", `{"module":"b","version":"1"}`)
+	merge := strings.Repeat("c", 40)
+	var lsTree []string
+	run := func(ctx context.Context, env []string, name string, args ...string) forgekey.Ran {
+		switch args[2] {
+		case "ls-files":
+			return forgekey.Ran{Stdout: "module.json\nmodules/b/module.json\n"}
+		case "rev-parse":
+			return forgekey.Ran{Stdout: "tree-of-" + args[3] + "\n"}
+		case "ls-tree":
+			lsTree = append(lsTree, strings.Join(args[4:], " "))
+			return forgekey.Ran{Stdout: "100644 blob " + strings.Repeat("b", 40) + "\tpackage.json\x00" +
+				"040000 tree " + strings.Repeat("a", 40) + "\tsrc\x00"}
+		}
+		return forgekey.Ran{Status: 1, Stderr: "unexpected " + args[2]}
+	}
+	got, err := ModuleTrees(context.Background(), run, nil, root, merge)
+	if err != nil {
+		t.Fatal(err)
+	}
+	paths := map[string]string{"src": strings.Repeat("a", 40), "package.json": strings.Repeat("b", 40)}
+	want, _ := delivery.NamedTree(hexOf(sdk), paths)
+	if g := got["."]; g.Tree != want || len(g.Paths) != 2 || g.Paths["src"] != paths["src"] || g.Paths["package.json"] != paths["package.json"] {
+		t.Fatalf("the root module = %+v, want tree %s and paths %v", g, want, paths)
+	}
+	if len(lsTree) != 1 || lsTree[0] != merge+" -- package.json src" {
+		t.Fatalf("git ls-tree ran as %v", lsTree)
+	}
+	if g := got["modules/b"]; g.Tree != "tree-of-"+merge+":modules/b" || g.Paths != nil {
+		t.Fatalf("a module naming no path was narrowed: %+v", g)
+	}
+}
+
+// A named path absent at the commit fails the answer, naming the module and the path: a tree computed without it would
+// pin a build source that does not exist.
+func TestModuleTreesFailsNamingANamedPathAbsentAtTheCommit(t *testing.T) {
+	root := t.TempDir()
+	writeTreeFile(t, root, "modules/a/module.json", `{"module":"a","version":"1","build":{"artifacts":[{"name":"lib","kind":"package","language":"typescript","paths":["src","gone"]}]}}`)
+	run := func(ctx context.Context, env []string, name string, args ...string) forgekey.Ran {
+		switch args[2] {
+		case "ls-files":
+			return forgekey.Ran{Stdout: "modules/a/module.json\n"}
+		case "rev-parse":
+			return forgekey.Ran{Stdout: "t\n"}
+		case "ls-tree":
+			if strings.Join(args[4:], " ") != strings.Repeat("c", 40)+" -- modules/a/gone modules/a/src" {
+				return forgekey.Ran{Status: 1, Stderr: "wrong arguments " + strings.Join(args[3:], " ")}
+			}
+			return forgekey.Ran{Stdout: "040000 tree " + strings.Repeat("a", 40) + "\tmodules/a/src\x00"}
+		}
+		return forgekey.Ran{Status: 1}
+	}
+	_, err := ModuleTrees(context.Background(), run, nil, root, strings.Repeat("c", 40))
+	if err == nil || !strings.Contains(err.Error(), `"gone"`) || !strings.Contains(err.Error(), "modules/a") {
+		t.Fatalf("an absent named path gave %v", err)
+	}
+}