chain-studio/lib/data/friendly_error.dart
flemming-it ea2cd87b5e fix(runs): stop claiming the hub is unreachable on RPC-level errors
The runs monitor folded every load failure into 'hub not reachable',
contradicting the sidebar's green connected dot whenever the hub
answered but the RPC failed — most visibly against a pre-0.22 hub
whose version predates the ListInvocations RPC (UNIMPLEMENTED).

Classify the failure instead (top-level, unit-tested):
- UNIMPLEMENTED -> 'this view needs a newer hub version' with a
  doctor-page link (the update banner lives there)
- FAILED_PRECONDITION from the detached gate -> the regular
  feature-off empty state with the guide button
- UNAVAILABLE / DEADLINE_EXCEEDED / socket-level failures -> the
  honest 'hub not reachable' state (unchanged)
- everything else -> a load-failed state with the friendly error
  and a copyable detail box

The error view is a public callback-driven widget so the tests pump
each variant without a live hub. New DE+EN strings for the too-old
and load-failed states; grpcCodeOf/grpcMessageOf exposed from the
friendly-error mapper instead of duplicating the duck-typing.

Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
2026-07-18 17:33:11 +02:00

268 lines
9 KiB
Dart

// Friendly-error mapper. Turns a thrown object (typically a
// `GrpcError` from package:grpc) into a short human sentence
// plus an optional recovery hint, ready to drop into the UI.
//
// Why this exists: Studio used to render `GrpcError.toString()`
// verbatim — operators saw walls of text like
// "gRPC Error (code: 13, codeName: INTERNAL, message: ..., details: [], rawResponse: null, trailers: {...})".
// The friendly mapper folds every gRPC code into a one-line
// explanation we can localise, plus a recovery action when one
// is obvious (open settings, restart hub, retry).
import '../l10n/app_localizations.dart';
/// Result of running an error through [friendlyError]. The
/// `headline` is the short user-facing sentence; `detail` is
/// the verbatim original error string (kept around in case the
/// operator wants to copy-paste it into a bug report). `hint`
/// is a one-sentence recovery suggestion or `null`.
class FriendlyError {
final String headline;
final String detail;
final String? hint;
const FriendlyError({
required this.headline,
required this.detail,
this.hint,
});
// The on-disk error log serialises thrown objects via toString —
// keep the whole story (headline, hint, verbatim detail) in one
// readable record.
@override
String toString() => [
headline,
?hint,
if (detail.isNotEmpty && detail != headline) detail,
].join('\n');
}
/// Map an arbitrary thrown object to a [FriendlyError]. Always
/// returns a value — never throws — so callers can drop the
/// result straight into UI without try/catch ceremony.
FriendlyError friendlyError(Object error, AppLocalizations l) {
// A pre-built FriendlyError passes through unchanged — call
// sites that already know the precise story (e.g. the setup
// wizard's CLI-version-skew case) construct one directly and
// still route through the shared presentation helpers.
if (error is FriendlyError) return error;
// We deliberately don't import package:grpc here so Studio
// doesn't have to add it to its own pubspec — the dependency
// lives one layer down in chain_client_sdk. `GrpcError` has a
// stable `.code` (int) and `.message` (String?) shape; we
// duck-type on those instead of an `is` check.
final code = _intField(error, 'code');
final detail = _stringField(error, 'message') ?? '';
// Hub-specific pattern detection — runs BEFORE the gRPC-code
// switch so a precise hit ("approval rejected by ...") wins
// over the generic "permission denied" label even though
// both correspond to gRPC PERMISSION_DENIED.
final specific = _matchHubPattern(detail, l);
if (specific != null) return specific;
if (code != null) {
switch (code) {
case 3: // INVALID_ARGUMENT
return FriendlyError(
headline: l.errInvalidArgument,
detail: detail,
hint: l.errInvalidArgumentHint,
);
case 5: // NOT_FOUND
return FriendlyError(
headline: l.errNotFound,
detail: detail,
hint: l.errNotFoundHint,
);
case 6: // ALREADY_EXISTS
return FriendlyError(
headline: l.errAlreadyExists,
detail: detail,
hint: null,
);
case 7: // PERMISSION_DENIED
return FriendlyError(
headline: l.errPermissionDenied,
detail: detail,
hint: l.errPermissionDeniedHint,
);
case 9: // FAILED_PRECONDITION
return FriendlyError(
headline: l.errFailedPrecondition,
detail: detail,
hint: l.errFailedPreconditionHint,
);
case 13: // INTERNAL
return FriendlyError(
headline: l.errInternal,
detail: detail,
hint: l.errInternalHint,
);
case 14: // UNAVAILABLE
return FriendlyError(
headline: l.errUnavailable,
detail: detail,
hint: l.errUnavailableHint,
);
case 16: // UNAUTHENTICATED
return FriendlyError(
headline: l.errUnauthenticated,
detail: detail,
hint: l.errUnauthenticatedHint,
);
default:
final codeName = _stringField(error, 'codeName') ?? 'gRPC $code';
return FriendlyError(
headline: '$codeName: ${detail.isEmpty ? l.errGeneric : detail}',
detail: detail,
hint: null,
);
}
}
// Non-gRPC error (e.g. FormatException from a bad URL). Show
// its toString — but as the headline, not buried in a wall.
return FriendlyError(
headline: error.toString(),
detail: error.toString(),
hint: null,
);
}
/// Pattern-match the gRPC error message for hub-specific
/// failure shapes. The hub's [FlowExecutionError] uses
/// well-shaped Display formats (e.g.
/// `step 'ap1' rejected by alice: ...`) — we read the prefix
/// to return a more specific [FriendlyError] than the
/// gRPC-code default. Returns null when nothing matches.
FriendlyError? _matchHubPattern(String detail, AppLocalizations l) {
if (detail.isEmpty) return null;
// Module bundle download failed (bad / unhosted URL). The hub wraps
// these as "download failed: ..." and (older builds) mapped them to
// gRPC Unavailable — which the code-default would render as the
// misleading "hub not reachable". The hub is fine; show the real URL
// + status (carried in `detail`) under a clear headline.
if (detail.contains('download failed') || detail.contains('download error')) {
return FriendlyError(
headline: l.errModuleDownload,
detail: detail,
hint: l.errModuleDownloadHint,
);
}
// Approval rejected: "step 'X' rejected by Y: reason"
if (detail.contains('rejected by')) {
return FriendlyError(
headline: l.errApprovalRejected,
detail: detail,
hint: l.errApprovalRejectedHint,
);
}
// Approval timeout: "step 'X' timed out waiting for approval after Ns"
if (detail.contains('timed out waiting for approval')) {
return FriendlyError(
headline: l.errApprovalTimedOut,
detail: detail,
hint: l.errApprovalTimedOutHint,
);
}
// Output too large: "step 'X' produced N bytes of output, exceeding the M MB cap"
if (detail.contains('exceeding the') &&
detail.contains('MB cap')) {
return FriendlyError(
headline: l.errOutputTooLarge,
detail: detail,
hint: l.errOutputTooLargeHint,
);
}
// Host service unavailable: "step 'X' requires service 'Y' which is not declared"
if (detail.contains('requires service') &&
detail.contains('not declared')) {
return FriendlyError(
headline: l.errServiceUnavailableForStep,
detail: detail,
hint: l.errServiceUnavailableForStepHint,
);
}
// Missing value: "step 'X' references missing value 'Y'"
if (detail.contains('references missing value')) {
return FriendlyError(
headline: l.errMissingValue,
detail: detail,
hint: l.errMissingValueHint,
);
}
// MCP server unreachable — hub surfaces these as Unavailable
// with the MCP transport name in the message.
if (detail.toLowerCase().contains('mcp') &&
(detail.contains('unreachable') ||
detail.contains('connection refused') ||
detail.contains('timed out'))) {
return FriendlyError(
headline: l.errMcpUnreachable,
detail: detail,
hint: l.errMcpUnreachableHint,
);
}
// Capability not in registry — hub returns NotFound with this
// shape from the flow engine.
if (detail.toLowerCase().contains('no capability provider') ||
detail.toLowerCase().contains('capability not installed')) {
return FriendlyError(
headline: l.errCapabilityNotInstalled,
detail: detail,
hint: l.errCapabilityNotInstalledHint,
);
}
return null;
}
/// Duck-typed `GrpcError.code` reader — public so pages that
/// classify errors themselves (e.g. the runs monitor separating
/// "hub down" from "hub too old") share one accessor instead of
/// re-implementing the duck-typing.
int? grpcCodeOf(Object error) => _intField(error, 'code');
/// Duck-typed `GrpcError.message` reader — companion to
/// [grpcCodeOf].
String? grpcMessageOf(Object error) => _stringField(error, 'message');
/// Try to read an `int` field by name off an arbitrary object.
/// Returns `null` when the field doesn't exist or has another
/// runtime type. Used to duck-type `GrpcError.code` without
/// pulling package:grpc as a Studio dependency.
int? _intField(Object obj, String field) {
try {
final dyn = obj as dynamic;
// ignore: avoid_dynamic_calls
final v = (() {
switch (field) {
case 'code':
return dyn.code;
}
return null;
})();
return v is int ? v : null;
} catch (_) {
return null;
}
}
String? _stringField(Object obj, String field) {
try {
final dyn = obj as dynamic;
// ignore: avoid_dynamic_calls
final v = (() {
switch (field) {
case 'message':
return dyn.message;
case 'codeName':
return dyn.codeName;
}
return null;
})();
return v is String ? v : null;
} catch (_) {
return null;
}
}