Merge pull request #1377 from entireio/cor-389-data-api-context-aware

data-api: follow auth context via /.well-known/entire-api.json

Author: toothbrush

cmd/entire/cli/activity_cmd_test.go

@@ -36,6 +36,10 @@ func TestRunActivity_SilencesContextCanceled(t *testing.T) {
 		return nil, context.Canceled
 	})
 	t.Cleanup(auth.SetManagerForTest(t, mgr))
+	// Force discovery-unavailable so ResolveDataAPIToken takes the static
+	// fallback through the singleton test manager above, rather than making a
+	// real network fetch to the configured data host.
+	t.Cleanup(auth.SetResolveContextForAPIForTest(t, auth.DiscoveryUnavailableForTest))
 
 	var out, errOut bytes.Buffer
 	err := runActivity(t.Context(), &out, &errOut)
@@ -67,6 +71,10 @@ func TestRunActivity_PrintsLoginHintOnNotLoggedIn(t *testing.T) {
 		return nil, errors.New("unreachable")
 	})
 	t.Cleanup(auth.SetManagerForTest(t, mgr))
+	// Force discovery-unavailable so ResolveDataAPIToken takes the static
+	// fallback through the singleton test manager above, rather than making a
+	// real network fetch to the configured data host.
+	t.Cleanup(auth.SetResolveContextForAPIForTest(t, auth.DiscoveryUnavailableForTest))
 
 	var out, errOut bytes.Buffer
 	err := runActivity(t.Context(), &out, &errOut)

cmd/entire/cli/api_client.go

@@ -38,10 +38,12 @@ func NewAuthenticatedAPIClient(ctx context.Context, insecureHTTP bool) (*api.Cli
 		}
 	}
 
-	// tokenmanager validates Resource as a strict origin URL; strip any path
-	// the operator may have included in ENTIRE_API_BASE_URL before handing
-	// it across the package boundary.
-	token, err := auth.TokenForResource(ctx, api.OriginOnly(dataURL))
+	// ResolveDataAPIToken discovers which login context the data host trusts
+	// (via its /.well-known/entire-api.json) and exchanges that context's
+	// token for the advertised audience, falling back to static resolution
+	// when the host doesn't advertise discovery. It normalises dataURL to an
+	// origin internally.
+	token, err := auth.ResolveDataAPIToken(ctx, dataURL)
 	if err != nil {
 		if errors.Is(err, auth.ErrNotLoggedIn) {
 			// Wrap the original err (not the sentinel) so any context

cmd/entire/cli/auth.go

@@ -75,18 +75,6 @@ func newAuthSessionsClient(coreURL, token string) *api.Client {
 	return api.NewClientWithBaseURL(token, coreURL).WithAuthSessionsPath(coreAuthSessionsPath)
 }
 
-// resolveAuthHostToken returns a bearer scoped for the auth host (entire-core).
-// For the auth host's own origin the tokenmanager hits the same-host shortcut
-// and returns the stored login JWT unchanged — keeping the entire:session
-// scope that core's session endpoints (and /me) require, with no STS exchange.
-func resolveAuthHostToken(ctx context.Context) (string, error) {
-	token, err := auth.TokenForResource(ctx, api.OriginOnly(api.AuthBaseURL()))
-	if err != nil {
-		return "", fmt.Errorf("resolve auth-host token: %w", err)
-	}
-	return token, nil
-}
-
 // isKeychainTokenRejected reports whether err indicates the stored
 // keyring token can't authenticate against entire-core. Failure modes that
 // collapse into the single "the user must re-login" branch:
@@ -167,7 +155,7 @@ func newAuthStatusCmd() *cobra.Command {
 			if err := requireSecureBaseURL(insecureHTTPAuth); err != nil {
 				return err
 			}
-			target, err := resolveStatusTarget(auth.NewContextStore(), auth.Contexts, api.AuthBaseURL())
+			target, err := resolveStatusTarget(cmd.Context(), auth.NewContextStore(), auth.Contexts, auth.RefreshedLoginToken, api.AuthBaseURL())
 			if err != nil {
 				return err
 			}
@@ -208,6 +196,11 @@ type authSessionLister func(ctx context.Context, coreURL, token string) ([]api.A
 // name. Injected for testability; production wires auth.Contexts.
 type contextsProvider func() ([]*contexts.Context, string, error)
 
+// loginTokenResolver returns a usable login JWT for a context, transparently
+// re-minting an expired one from the stored refresh token. Injected so status
+// tests don't reach the network; production wires auth.RefreshedLoginToken.
+type loginTokenResolver func(ctx context.Context, c *contexts.Context) (string, error)
+
 // statusTarget is the resolved core to act against: the active context's
 // CoreURL + its session token, or (no active context) the configured
 // AuthBaseURL + legacy keyring entry. Shared by `auth status` (profile +
@@ -219,17 +212,29 @@ type statusTarget struct {
 	totalContexts int
 }
 
-// resolveStatusTarget picks the core + token for `entire auth status`. The
-// active contexts.json context wins (so `auth use` retargets status onto that
-// login server); otherwise it falls back to the legacy keyring entry keyed by
-// the configured auth host.
+// resolveStatusTarget picks the core + token for `entire auth status` (and
+// `logout`). The active contexts.json context wins (so `auth use` retargets
+// status onto that login server); otherwise it falls back to the legacy keyring
+// entry keyed by the configured auth host.
+//
+// For the active context the token is resolved through resolveLogin, which
+// transparently re-mints an expired login JWT from the stored refresh token.
+// This is the point of the refresh: an expired-but-refreshable session must
+// report "logged in", not "re-login" — the same false negative
+// auth.ResolveControlPlaneTarget already avoids for org/repo/project/grant.
+// `logout` benefits too: the refreshed bearer can authenticate the revoke call
+// instead of failing on an expired token. When refresh fails (revoked family,
+// network, opaque token), we fall back to the stored token and let the /me
+// liveness probe be the arbiter — preserving the accurate "no longer valid"
+// outcome for a genuinely dead session (ErrReauthRequired → expired token →
+// 401 → re-login).
 //
 // A genuine contexts.json read/parse error is surfaced, not swallowed — a
 // missing file reads as "no contexts" (no error), so an error here means the
 // file is corrupt or unreadable, which the user must see. This keeps status
 // symmetric with the control-plane commands (auth.ResolveControlPlaneTarget),
 // which fail the same way rather than silently degrading to a stale identity.
-func resolveStatusTarget(store tokenStore, listContexts contextsProvider, fallbackBaseURL string) (statusTarget, error) {
+func resolveStatusTarget(ctx context.Context, store tokenStore, listContexts contextsProvider, resolveLogin loginTokenResolver, fallbackBaseURL string) (statusTarget, error) {
 	all, current, err := listContexts()
 	if err != nil {
 		return statusTarget{}, fmt.Errorf("load contexts: %w", err)
@@ -239,6 +244,12 @@ func resolveStatusTarget(store tokenStore, listContexts contextsProvider, fallba
 		if c.Name != current || c.CoreURL == "" {
 			continue
 		}
+		// Prefer a refreshed token; fall back to the raw stored token so a
+		// refresh failure degrades to today's behaviour rather than dropping
+		// to the legacy entry.
+		if tok, terr := resolveLogin(ctx, c); terr == nil && tok != "" {
+			return statusTarget{coreURL: c.CoreURL, token: tok, activeContext: c.Name, totalContexts: total}, nil
+		}
 		if tok, terr := auth.LoginTokenForContext(c); terr == nil && tok != "" {
 			return statusTarget{coreURL: c.CoreURL, token: tok, activeContext: c.Name, totalContexts: total}, nil
 		}

cmd/entire/cli/auth/data_api.go

@@ -0,0 +1,118 @@
+package auth
+
+import (
+	"context"
+	"errors"
+	"net/http"
+	"net/url"
+	"time"
+
+	"github.com/entireio/cli/cmd/entire/cli/api"
+	"github.com/entireio/cli/internal/entireclient/clusterdiscovery"
+	"github.com/entireio/cli/internal/entireclient/contexts"
+	"github.com/entireio/cli/internal/entireclient/discovery"
+)
+
+// dataAPIDiscoveryTimeout bounds the one /.well-known/entire-api.json GET we
+// add per data-API command. Kept short: on any failure we fall back to static
+// resolution, so a slow or absent endpoint must not stall the command.
+const dataAPIDiscoveryTimeout = 8 * time.Second
+
+// resolveContextForAPIFunc is the shape of the discovery seam: it mirrors
+// clusterdiscovery.ResolveContextForAPI (ctx, configDir, cacheDir, apiHost,
+// httpClient, debugf).
+type resolveContextForAPIFunc func(context.Context, string, string, string, *http.Client, clusterdiscovery.DebugFunc) (*contexts.Context, error)
+
+// resolveContextForAPI is the discovery seam, swapped in tests so they don't
+// reach the network. See SetResolveContextForAPIForTest for cross-package tests.
+var resolveContextForAPI resolveContextForAPIFunc = clusterdiscovery.ResolveContextForAPI
+
+// SetResolveContextForAPIForTest overrides the /.well-known/entire-api.json
+// discovery seam and returns a cleanup func. Tests in other packages that
+// exercise a data-API command (activity/search/dispatch/recap) MUST install
+// this — otherwise ResolveDataAPIToken makes a real network call to the
+// configured data host and bypasses any SetManagerForTest fallback seam. Pass
+// a func returning clusterdiscovery.ErrDiscoveryUnavailable to force the static
+// fallback path. Test-only.
+func SetResolveContextForAPIForTest(t interface{ Helper() }, fn resolveContextForAPIFunc) func() {
+	t.Helper()
+	prev := resolveContextForAPI
+	resolveContextForAPI = fn
+	return func() { resolveContextForAPI = prev }
+}
+
+// DiscoveryUnavailableForTest is a ready-made SetResolveContextForAPIForTest
+// value that forces the discovery-unavailable fallback (no network), so a
+// cross-package test exercises the static TokenForResource path deterministically.
+func DiscoveryUnavailableForTest(context.Context, string, string, string, *http.Client, clusterdiscovery.DebugFunc) (*contexts.Context, error) {
+	return nil, clusterdiscovery.ErrDiscoveryUnavailable
+}
+
+// ResolveDataAPIToken returns a bearer for the data API at dataBaseURL.
+//
+// It dials the API's /.well-known/entire-api.json to learn which login
+// server(s) the API trusts and which audience to exchange for, picks the
+// matching local auth context (active-wins-if-eligible → sole → explicit
+// choice), and exchanges that context's login JWT for the advertised audience
+// at that context's core. This is what makes
+//
+//	ENTIRE_API_BASE_URL=https://partial.to entire activity
+//
+// authenticate as the partial.to login even while the active context is a
+// prod entire.io login — without the operator also setting ENTIRE_AUTH_BASE_URL.
+//
+// When the API doesn't advertise discovery (404 / unreachable / 503 /
+// malformed — e.g. a deployment predating the well-known), it falls back to
+// the pre-discovery static path (TokenForResource through the singleton
+// manager) so behaviour is never worse than before. A reachable API whose
+// context selection fails (no eligible context, or several with none active)
+// surfaces that error directly — the user must log in or pick one.
+//
+// Callers that honour --insecure-http-auth must call EnableInsecureHTTP before
+// invoking this (as they already do); the per-context exchange and the static
+// fallback both read that global opt-in.
+func ResolveDataAPIToken(ctx context.Context, dataBaseURL string) (string, error) {
+	dataOrigin := api.OriginOnly(dataBaseURL)
+	host, ok := hostOf(dataOrigin)
+	if !ok {
+		// Can't derive a host to discover against — use static resolution.
+		return TokenForResource(ctx, dataOrigin)
+	}
+
+	// Bridge any pre-contexts.json login so the resolver can match it, mirroring
+	// the git remote helper's cold-boot path. Best-effort: a migration failure
+	// must not block resolution.
+	_, _ = MigrateLegacyLoginContext() //nolint:errcheck // best-effort bridge; resolution proceeds regardless
+
+	dctx, cancel := context.WithTimeout(ctx, dataAPIDiscoveryTimeout)
+	defer cancel()
+	httpClient := &http.Client{Timeout: dataAPIDiscoveryTimeout}
+
+	selected, err := resolveContextForAPI(dctx, contexts.DefaultConfigDir(), discovery.DefaultCacheDir(), host, httpClient, nil)
+	if errors.Is(err, clusterdiscovery.ErrDiscoveryUnavailable) {
+		// Old deployment / not rolled out / transient — preserve today's behaviour.
+		return TokenForResource(ctx, dataOrigin)
+	}
+	if err != nil {
+		return "", err
+	}
+
+	// Exchange for the data host origin; the token manager derives the RFC 8693
+	// audience from it, which is the aud the API requires (aud == base URI).
+	allowInsecure := insecureHTTPEnabled() || isLoopbackHTTP(selected.CoreURL)
+	provider, err := NewRefreshingResourceProvider(selected, dataOrigin, nil, allowInsecure)
+	if err != nil {
+		return "", err
+	}
+	return provider(ctx)
+}
+
+// hostOf returns the host[:port] of an origin URL, ok=false when it can't be
+// parsed into a host.
+func hostOf(origin string) (string, bool) {
+	u, err := url.Parse(origin)
+	if err != nil || u.Host == "" {
+		return "", false
+	}
+	return u.Host, true
+}