Merge pull request 'feat(ai-chat): персист tool-error частей — упавшие тулы видны в истории и сохраняют текст ошибки' (#426) from feat/407-persist-tool-errors into develop
Reviewed-on: #426
This commit was merged in pull request #426.
This commit is contained in:
@@ -148,6 +148,53 @@ describe('assistantParts', () => {
|
||||
expect(toolPart).not.toHaveProperty('output');
|
||||
});
|
||||
|
||||
it('replays the REAL error text for a THROWN tool (tool-error part)', () => {
|
||||
const steps = [
|
||||
{
|
||||
text: '',
|
||||
toolCalls: [
|
||||
{ toolCallId: 'c7', toolName: 'editPageText', input: { id: 'p1' } },
|
||||
],
|
||||
// A thrown tool is a `tool-error` content part; toolResults holds only
|
||||
// successes and stays empty for this call.
|
||||
toolResults: [],
|
||||
content: [
|
||||
{
|
||||
type: 'tool-error',
|
||||
toolCallId: 'c7',
|
||||
toolName: 'editPageText',
|
||||
input: { id: 'p1' },
|
||||
error: new Error('page is locked'),
|
||||
},
|
||||
],
|
||||
},
|
||||
];
|
||||
const parts = assistantParts(steps, '') as AnyPart[];
|
||||
const toolPart = parts.find((p) => p.type === 'tool-editPageText');
|
||||
expect(toolPart).toBeDefined();
|
||||
expect(toolPart!.state).toBe('output-error');
|
||||
// The REAL error is replayed, NOT the 'Tool call did not complete.' placeholder.
|
||||
expect(toolPart!.errorText).toBe('page is locked');
|
||||
expect(toolPart).not.toHaveProperty('output');
|
||||
});
|
||||
|
||||
it('keeps the placeholder ONLY for a call with neither result nor tool-error', () => {
|
||||
const steps = [
|
||||
{
|
||||
text: '',
|
||||
toolCalls: [
|
||||
{ toolCallId: 'c8', toolName: 'insertNode', input: { node: {} } },
|
||||
],
|
||||
toolResults: [],
|
||||
content: [], // aborted mid-step: no result AND no tool-error
|
||||
},
|
||||
];
|
||||
const parts = assistantParts(steps, '') as AnyPart[];
|
||||
const toolPart = parts.find((p) => p.type === 'tool-insertNode');
|
||||
expect(toolPart!.state).toBe('output-error');
|
||||
expect(toolPart!.errorText).toBe('Tool call did not complete.');
|
||||
});
|
||||
|
||||
it('skips malformed tool-calls (missing toolName or toolCallId)', () => {
|
||||
const steps = [
|
||||
{
|
||||
@@ -195,6 +242,45 @@ describe('serializeSteps', () => {
|
||||
expect(trace[0]).toEqual({ toolName: 'getPage', input: { id: 'p1' } });
|
||||
expect(trace[1]).toEqual({ toolName: 'getPage', output: { title: 'T' } });
|
||||
});
|
||||
|
||||
it('records a THROWN tool failure (tool-error part) with its error message', () => {
|
||||
const trace = serializeSteps([
|
||||
{
|
||||
toolCalls: [{ toolName: 'editPageText', input: { id: 'p1' } }],
|
||||
toolResults: [],
|
||||
content: [
|
||||
{
|
||||
type: 'tool-error',
|
||||
toolName: 'editPageText',
|
||||
error: new Error('page is locked'),
|
||||
},
|
||||
],
|
||||
},
|
||||
]) as Array<Record<string, unknown>>;
|
||||
// The call element is followed by a paired error element (mirroring how a
|
||||
// successful result is appended), so the failure survives in the trace.
|
||||
expect(trace).toHaveLength(2);
|
||||
expect(trace[0]).toEqual({ toolName: 'editPageText', input: { id: 'p1' } });
|
||||
expect(trace[1]).toEqual({
|
||||
toolName: 'editPageText',
|
||||
error: 'page is locked',
|
||||
});
|
||||
});
|
||||
|
||||
it('truncates a very long tool-error message to the tool-output limit', () => {
|
||||
const long = 'x'.repeat(5000);
|
||||
const trace = serializeSteps([
|
||||
{
|
||||
toolCalls: [{ toolName: 'editPageText', input: {} }],
|
||||
toolResults: [],
|
||||
content: [{ type: 'tool-error', toolName: 'editPageText', error: long }],
|
||||
},
|
||||
]) as Array<Record<string, unknown>>;
|
||||
const errorText = trace[1].error as string;
|
||||
// Truncated (not the full 5000 chars) and carries the omission marker.
|
||||
expect(errorText.length).toBeLessThan(long.length);
|
||||
expect(errorText).toContain('chars omitted');
|
||||
});
|
||||
});
|
||||
|
||||
describe('rowToUiMessage', () => {
|
||||
|
||||
@@ -1637,6 +1637,17 @@ type StepLike = {
|
||||
toolName?: string;
|
||||
output?: unknown;
|
||||
}>;
|
||||
// ai@6.0.134: a tool that THREW surfaces as a `tool-error` content part
|
||||
// ({ type:'tool-error', toolCallId, toolName, input, error }), NOT as a
|
||||
// `toolResults` entry (which holds only successes). Read from here so failed
|
||||
// calls are persisted with their real error instead of being dropped.
|
||||
content?: ReadonlyArray<{
|
||||
type?: string;
|
||||
toolCallId?: string;
|
||||
toolName?: string;
|
||||
input?: unknown;
|
||||
error?: unknown;
|
||||
}>;
|
||||
};
|
||||
|
||||
/**
|
||||
@@ -1739,6 +1750,26 @@ function compactValue(value: unknown, depth: number): unknown {
|
||||
return value;
|
||||
}
|
||||
|
||||
/**
|
||||
* Extract a bounded string message from a `tool-error` part's `error` field for
|
||||
* persistence and history replay. The field may be an `Error`, a string, or an
|
||||
* arbitrary object, so pull a message robustly. The result is passed through
|
||||
* `compactValue` so a very long error honors the SAME truncation limits the file
|
||||
* already applies to tool outputs (no new limit is introduced here).
|
||||
*/
|
||||
function normalizeToolError(error: unknown): string {
|
||||
const message =
|
||||
error instanceof Error
|
||||
? error.message
|
||||
: typeof error === 'string'
|
||||
? error
|
||||
: error != null &&
|
||||
typeof (error as { message?: unknown }).message === 'string'
|
||||
? (error as { message: string }).message
|
||||
: String(error);
|
||||
return compactValue(message, 0) as string;
|
||||
}
|
||||
|
||||
/**
|
||||
* Rebuild the FULL UIMessage `parts` for an assistant turn from the SDK steps,
|
||||
* so multi-turn history replays prior tool-calls/results to the model (not just
|
||||
@@ -1771,6 +1802,14 @@ export function assistantParts(
|
||||
for (const r of step.toolResults ?? []) {
|
||||
if (r.toolCallId) resultsById.set(r.toolCallId, r.output);
|
||||
}
|
||||
// Index this step's THROWN tool failures (ai@6 `tool-error` content parts)
|
||||
// by tool call id, so a call that failed replays with its real error text.
|
||||
const errorsById = new Map<string, unknown>();
|
||||
for (const part of step.content ?? []) {
|
||||
if (part.type === 'tool-error' && part.toolCallId) {
|
||||
errorsById.set(part.toolCallId, part.error);
|
||||
}
|
||||
}
|
||||
for (const call of step.toolCalls ?? []) {
|
||||
if (!call.toolName || !call.toolCallId) continue;
|
||||
const hasResult = resultsById.has(call.toolCallId);
|
||||
@@ -1783,9 +1822,21 @@ export function assistantParts(
|
||||
input: call.input,
|
||||
output: compactToolOutput(resultsById.get(call.toolCallId)),
|
||||
});
|
||||
} else if (errorsById.has(call.toolCallId)) {
|
||||
// The tool THREW: replay the REAL error so the model on the next turn
|
||||
// knows WHY the call failed (and does not blindly repeat it). An
|
||||
// output-error round-trips through convertToModelMessages as a balanced
|
||||
// tool-call + tool-result, keeping the rebuilt history valid.
|
||||
parts.push({
|
||||
type: `tool-${call.toolName}`,
|
||||
toolCallId: call.toolCallId,
|
||||
state: 'output-error',
|
||||
input: call.input,
|
||||
errorText: normalizeToolError(errorsById.get(call.toolCallId)),
|
||||
});
|
||||
} else {
|
||||
// No paired result (e.g. aborted mid-step). Persisting a bare
|
||||
// tool-call (input-available) would replay as an unpaired call and
|
||||
// No paired result AND no tool-error (e.g. aborted mid-step). Persisting
|
||||
// a bare tool-call (input-available) would replay as an unpaired call and
|
||||
// throw MissingToolResultsError on the next turn (convertToModelMessages
|
||||
// emits no tool-result for it). Emit a SYNTHETIC paired result instead:
|
||||
// an output-error round-trips through convertToModelMessages as a
|
||||
@@ -2021,10 +2072,19 @@ export function serializeSteps(
|
||||
steps: ReadonlyArray<{
|
||||
toolCalls?: ReadonlyArray<{ toolName?: string; input?: unknown }>;
|
||||
toolResults?: ReadonlyArray<{ toolName?: string; output?: unknown }>;
|
||||
content?: ReadonlyArray<{
|
||||
type?: string;
|
||||
toolName?: string;
|
||||
error?: unknown;
|
||||
}>;
|
||||
}>,
|
||||
): unknown {
|
||||
const calls: Array<{ toolName?: string; input?: unknown; output?: unknown }> =
|
||||
[];
|
||||
const calls: Array<{
|
||||
toolName?: string;
|
||||
input?: unknown;
|
||||
output?: unknown;
|
||||
error?: string;
|
||||
}> = [];
|
||||
for (const step of steps ?? []) {
|
||||
for (const call of step.toolCalls ?? []) {
|
||||
calls.push({ toolName: call.toolName, input: call.input });
|
||||
@@ -2032,6 +2092,18 @@ export function serializeSteps(
|
||||
for (const r of step.toolResults ?? []) {
|
||||
calls.push({ toolName: r.toolName, output: compactToolOutput(r.output) });
|
||||
}
|
||||
// ai@6 surfaces a THROWN tool failure as a `tool-error` content part, NOT as
|
||||
// a `toolResults` entry. Record it as its own paired element (mirroring how a
|
||||
// successful result is appended) so the failure and its reason survive in the
|
||||
// trace instead of leaving an orphaned call with no result.
|
||||
for (const part of step.content ?? []) {
|
||||
if (part.type === 'tool-error') {
|
||||
calls.push({
|
||||
toolName: part.toolName,
|
||||
error: normalizeToolError(part.error),
|
||||
});
|
||||
}
|
||||
}
|
||||
}
|
||||
return calls.length > 0 ? calls : null;
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user