Hand a build's fetch the named paths in a shared request, so no build module reads them from a manifest (issue 1758, ADR 0499 §4, ADR 0495 §1)

buildstep.FetchRequest is what the builder writes and the build modules read: the
repository, commit, ref, base and, for a module naming its paths, its directory and
those paths from the indexer's kept inputs.
This commit is contained in:
2026-10-22 17:29:58 +02:00
parent d32743499c
commit 7cf73ed76b
2 changed files with 172 additions and 0 deletions
+115
View File
@@ -0,0 +1,115 @@
package buildstep
import (
"fmt"
"path"
"regexp"
"strings"
)
var (
requestCommitOK = regexp.MustCompile(`^[0-9a-f]{40}$`)
requestRepositoryOK = regexp.MustCompile(`^[A-Za-z0-9][A-Za-z0-9._-]*/[A-Za-z0-9][A-Za-z0-9._-]*$`)
)
// FetchRequest is what a build module's fetch tool is asked: the repository and the full commit, the pull request's
// ref when only its head carries the commit, the trunk commit to merge onto for a merge preview, and, for a build of a
// module whose every artifact names its paths, that module's directory and its named paths (novox/hq ADR 0499 clause 4).
//
// **The paths come from the builder, not from a manifest** (ADR 0495 clause 1; the lead's decision L8 of R4e): the
// builder fills Dir and Paths from the indexer's Module value (Path, and KeptInputs2.Paths' keys), and the build module
// gives the build and test steps a workspace holding those paths, the manifest and its lock files. A request without
// them builds from the module's directory, as before.
type FetchRequest struct {
Repository string `json:"repository"`
Commit string `json:"commit"`
Ref string `json:"ref,omitempty"`
Base string `json:"base,omitempty"`
Dir string `json:"dir,omitempty"`
Paths []string `json:"paths,omitempty"`
}
// FetchRequestProperties are the JSON schema properties of FetchRequest's dir and paths, for a fetch tool's Input.
func FetchRequestProperties() map[string]any {
return map[string]any{
"dir": map[string]any{"type": "string", "description": "the module's directory in the repository (\".\" for its root), with paths"},
"paths": map[string]any{"type": "array", "items": map[string]any{"type": "string"},
"description": "the paths the module names, relative to dir, from the indexer's kept inputs: the build sees only these, the manifest and its lock files"},
}
}
// Args is r as a tool call's arguments.
func (r FetchRequest) Args() map[string]any {
args := map[string]any{"repository": r.Repository, "commit": r.Commit}
for k, v := range map[string]string{"ref": r.Ref, "base": r.Base, "dir": r.Dir} {
if v != "" {
args[k] = v
}
}
if len(r.Paths) > 0 {
args["paths"] = append([]string(nil), r.Paths...)
}
return args
}
// ParseFetchRequest reads a fetch tool's arguments; arguments it does not know (a build module's own) are left to the
// caller. It refuses a repository that is not <owner>/<name>, a commit or a base that is not 40 hex digits, paths with no
// module directory or a directory with no path, a directory leaving the repository and a path leaving the directory,
// naming each.
func ParseFetchRequest(args map[string]any) (FetchRequest, error) {
var r FetchRequest
r.Repository, _ = args["repository"].(string)
r.Commit, _ = args["commit"].(string)
r.Ref, _ = args["ref"].(string)
r.Base, _ = args["base"].(string)
r.Dir, _ = args["dir"].(string)
switch v := args["paths"].(type) {
case nil:
case []string:
r.Paths = append([]string(nil), v...)
case []any:
for _, p := range v {
s, ok := p.(string)
if !ok {
return FetchRequest{}, fmt.Errorf("paths is a list of strings")
}
r.Paths = append(r.Paths, s)
}
default:
return FetchRequest{}, fmt.Errorf("paths is a list of strings")
}
return r, r.Check()
}
// Check is what ParseFetchRequest refuses.
func (r FetchRequest) Check() error {
if !requestRepositoryOK.MatchString(r.Repository) || strings.Contains(r.Repository, "..") {
return fmt.Errorf("%q is not a repository, <owner>/<name>", r.Repository)
}
if !requestCommitOK.MatchString(r.Commit) {
return fmt.Errorf("%q is not a full commit (40 hex digits)", r.Commit)
}
if r.Base != "" && !requestCommitOK.MatchString(r.Base) {
return fmt.Errorf("%q is not a trunk commit (40 hex digits)", r.Base)
}
switch {
case len(r.Paths) > 0 && r.Dir == "":
return fmt.Errorf("the request names paths and names no module directory (dir) they are relative to")
case len(r.Paths) == 0 && r.Dir != "":
return fmt.Errorf("the request names a module directory and no path")
}
if r.Dir != "" && (outside(r.Dir) || strings.HasPrefix(r.Dir, "/")) {
return fmt.Errorf("the module directory %q leaves the repository", r.Dir)
}
for _, p := range r.Paths {
if p == "" || strings.HasPrefix(p, "/") || outside(p) || path.Clean(p) == "." {
return fmt.Errorf("the path %q leaves the module's directory or is not a path in it", p)
}
}
return nil
}
func outside(p string) bool {
c := path.Clean(p)
return c == ".." || strings.HasPrefix(c, "../")
}
+57
View File
@@ -0,0 +1,57 @@
package buildstep
import (
"reflect"
"strings"
"testing"
)
// A build's fetch is handed the paths of a module that names them (novox/hq ADR 0499 clause 4): the builder fills them
// from the indexer's kept inputs and the build module reads them from the call, never from the manifest (ADR 0495
// clause 1). The request survives the trip through a tool call's arguments.
func TestAFetchRequestCarriesTheNamedPathsThroughTheToolCall(t *testing.T) {
req := FetchRequest{Repository: "novox/mesh-sdk", Commit: strings.Repeat("a", 40), Ref: "refs/pull/7/head",
Dir: ".", Paths: []string{"package.json", "src"}}
args := req.Args()
// A tool call's arguments arrive as JSON: lists of any.
args["paths"] = []any{"package.json", "src"}
got, err := ParseFetchRequest(args)
if err != nil {
t.Fatal(err)
}
if !reflect.DeepEqual(got, req) {
t.Fatalf("got %+v, want %+v", got, req)
}
plain, err := ParseFetchRequest(map[string]any{"repository": "novox/x", "commit": strings.Repeat("b", 40),
"base": strings.Repeat("c", 40), "beside": []any{}})
if err != nil || plain.Paths != nil || plain.Base != strings.Repeat("c", 40) {
t.Fatalf("a request without paths: %+v, %v", plain, err)
}
ok := map[string]any{"repository": "novox/x", "commit": strings.Repeat("b", 40)}
for _, c := range []struct {
set map[string]any
says string
}{
{map[string]any{"repository": "novox"}, "is not a repository"},
{map[string]any{"repository": "novox/../x"}, "is not a repository"},
{map[string]any{"commit": "abc"}, "is not a full commit"},
{map[string]any{"base": "abc"}, "is not a trunk commit"},
{map[string]any{"paths": []any{"src"}}, "names no module directory"},
{map[string]any{"dir": "modules/a", "paths": []any{"../b"}}, "leaves the module's directory"},
{map[string]any{"dir": "../a", "paths": []any{"src"}}, "leaves the repository"},
{map[string]any{"dir": "modules/a", "paths": []any{3}}, "paths is a list of strings"},
{map[string]any{"dir": "modules/a"}, "names a module directory and no path"},
} {
args := map[string]any{}
for k, v := range ok {
args[k] = v
}
for k, v := range c.set {
args[k] = v
}
if _, err := ParseFetchRequest(args); err == nil || !strings.Contains(err.Error(), c.says) {
t.Errorf("%v: %v, wanted a refusal saying %q", c.set, err, c.says)
}
}
}