v0.2.20 6938d13

amazon_bedrock_test.go RawHistoryFolder

1package provider
2
3import (
4	"bytes"
5	"encoding/json"
6	"net/http"
7	"strings"
8	"testing"
9	"time"
10)
11
12// TestSigV4Known checks our signer against AWS's published test vector
13// from the SigV4 documentation. The expected Authorization header below
14// is taken from "get-vanilla" in the suite (adapted: we use POST with
15// an empty body since Bedrock is always POST).
16//
17// We don't run the official suite (it lives in S3 as a tarball and the
18// signing rules vary slightly between examples); instead this test
19// pins a fixed input/output so future refactors can't silently break
20// the signer.
21func TestSigV4Deterministic(t *testing.T) {
22	creds := &bedrockSigV4Creds{
23		accessKeyID:     "AKIDEXAMPLE",
24		secretAccessKey: "wJalrXUtnFEMI/K7MDENG+bPxRfiCYEXAMPLEKEY",
25	}
26	body := []byte(`{"hello":"world"}`)
27	req, err := http.NewRequest("POST", "https://bedrock-runtime.us-east-1.amazonaws.com/model/foo/converse-stream", bytes.NewReader(body))
28	if err != nil {
29		t.Fatal(err)
30	}
31	req.Header.Set("content-type", "application/json")
32	when := time.Date(2024, 1, 15, 12, 0, 0, 0, time.UTC)
33	if err := signSigV4(req, body, "bedrock", "us-east-1", creds, when); err != nil {
34		t.Fatal(err)
35	}
36	auth := req.Header.Get("Authorization")
37	if !strings.HasPrefix(auth, "AWS4-HMAC-SHA256 ") {
38		t.Fatalf("missing AWS4 prefix: %q", auth)
39	}
40	if !strings.Contains(auth, "Credential=AKIDEXAMPLE/20240115/us-east-1/bedrock/aws4_request") {
41		t.Errorf("wrong credential scope: %q", auth)
42	}
43	if !strings.Contains(auth, "SignedHeaders=") {
44		t.Errorf("missing SignedHeaders: %q", auth)
45	}
46	if !strings.Contains(auth, "Signature=") {
47		t.Errorf("missing Signature: %q", auth)
48	}
49	// Sign again with identical inputs -> identical signature (determinism).
50	req2, _ := http.NewRequest("POST", "https://bedrock-runtime.us-east-1.amazonaws.com/model/foo/converse-stream", bytes.NewReader(body))
51	req2.Header.Set("content-type", "application/json")
52	if err := signSigV4(req2, body, "bedrock", "us-east-1", creds, when); err != nil {
53		t.Fatal(err)
54	}
55	if req2.Header.Get("Authorization") != auth {
56		t.Errorf("signature not deterministic:\n a=%q\n b=%q", auth, req2.Header.Get("Authorization"))
57	}
58}
59
60func TestSigV4PathEncoding(t *testing.T) {
61	// Bedrock model IDs include colons and dots. The colon must encode
62	// to %3A in the canonical path (per SigV4 rules).
63	cases := map[string]string{
64		"/model/anthropic.claude-sonnet-4-5-20250929-v1:0/converse-stream": "/model/anthropic.claude-sonnet-4-5-20250929-v1%3A0/converse-stream",
65		"/":  "/",
66		"":   "/",
67		"/a": "/a",
68	}
69	for in, want := range cases {
70		got := canonicalSigV4Path(in)
71		if got != want {
72			t.Errorf("canonicalSigV4Path(%q) = %q, want %q", in, got, want)
73		}
74	}
75}
76
77func TestReadAWSCredentialsFile(t *testing.T) {
78	// Smoke test: parses an INI-ish file when present. We don't require
79	// the file to exist on the test host.
80	if _, err := readAWSCredentialsFile("default"); err != nil {
81		// missing file or missing profile is fine — both are non-panics
82		t.Logf("no aws creds available (expected on CI): %v", err)
83	}
84}
85
86func TestNormalizeBedrockToolResultsMovesResultsAdjacentToToolUse(t *testing.T) {
87	msgs := []Message{
88		{Role: RoleAssistant, Content: []Content{ToolCallBlock{ID: "tool-1", Name: "edit", Arguments: json.RawMessage(`{"path":"a"}`)}}},
89		{Role: RoleUser, Content: []Content{TextBlock{Text: "what's wrong"}}},
90		{Role: RoleTool, Content: []Content{ToolResultBlock{CallID: "tool-1", Content: []Content{TextBlock{Text: "edited"}}}}},
91	}
92
93	out := normalizeBedrockToolResults(msgs)
94	if len(out) != 3 {
95		t.Fatalf("got %d messages, want 3: %+v", len(out), out)
96	}
97	if out[0].Role != RoleAssistant {
98		t.Fatalf("first role = %s, want assistant", out[0].Role)
99	}
100	if out[1].Role != RoleTool {
101		t.Fatalf("second role = %s, want tool", out[1].Role)
102	}
103	tr, ok := out[1].Content[0].(ToolResultBlock)
104	if !ok {
105		t.Fatalf("second message content = %T, want ToolResultBlock", out[1].Content[0])
106	}
107	if tr.CallID != "tool-1" || tr.IsError {
108		t.Fatalf("unexpected tool result: %+v", tr)
109	}
110	if out[2].Role != RoleUser {
111		t.Fatalf("third role = %s, want user", out[2].Role)
112	}
113}
114
115func TestNormalizeBedrockToolResultsInjectsMissingResult(t *testing.T) {
116	msgs := []Message{
117		{Role: RoleAssistant, Content: []Content{ToolCallBlock{ID: "tool-1", Name: "edit", Arguments: json.RawMessage(`{}`)}}},
118		{Role: RoleUser, Content: []Content{TextBlock{Text: "what's wrong"}}},
119	}
120
121	out := normalizeBedrockToolResults(msgs)
122	if len(out) != 3 {
123		t.Fatalf("got %d messages, want 3: %+v", len(out), out)
124	}
125	tr, ok := out[1].Content[0].(ToolResultBlock)
126	if !ok {
127		t.Fatalf("second message content = %T, want ToolResultBlock", out[1].Content[0])
128	}
129	if tr.CallID != "tool-1" || !tr.IsError {
130		t.Fatalf("unexpected synthetic result: %+v", tr)
131	}
132}
133
134func TestBedrockModelSupportsCaching(t *testing.T) {
135	cases := []struct {
136		model string
137		want  bool
138	}{
139		// Bare Claude IDs (as they come from catalog_builtin)
140		{"anthropic.claude-sonnet-4-5-20250929-v1:0", true},
141		{"anthropic.claude-haiku-4-5-20251001-v1:0", true},
142		{"anthropic.claude-opus-4-5-20251101-v1:0", true},
143		// Geo-prefixed (resolved form that arrives at bedrockModelSupportsCaching)
144		{"us.anthropic.claude-sonnet-4-5-20250929-v1:0", true},
145		{"eu.anthropic.claude-haiku-4-5-20251001-v1:0", true},
146		{"global.anthropic.claude-opus-4-5-20251101-v1:0", true},
147		// Nova models have PriceCacheWrite==0 — they use automatic caching
148		{"amazon.nova-pro-v1:0", false},
149		{"amazon.nova-lite-v1:0", false},
150		{"amazon.nova-micro-v1:0", false},
151		// DeepSeek — no cache write price
152		{"deepseek.r1-v1:0", false},
153		// Unknown model with claude prefix
154		{"anthropic.claude-future-v99:0", true},
155		// Unknown non-claude model
156		{"some.unknown-model-v1:0", false},
157	}
158	for _, c := range cases {
159		got := bedrockModelSupportsCaching(c.model)
160		if got != c.want {
161			t.Errorf("bedrockModelSupportsCaching(%q) = %v, want %v", c.model, got, c.want)
162		}
163	}
164}
165
166func TestBedrockBuildRequestCachingClaudeModel(t *testing.T) {
167	// A Claude model (PriceCacheWrite > 0) should get cachePoint markers
168	// in the system array and on the last user message.
169	client := &bedrockClient{region: "us-east-1"}
170	req, err := client.buildRequest(Request{
171		Model:  "anthropic.claude-sonnet-4-5-20250929-v1:0",
172		System: "You are a helpful assistant.",
173		Messages: []Message{
174			{Role: RoleUser, Content: []Content{TextBlock{Text: "hello"}}},
175		},
176	})
177	if err != nil {
178		t.Fatal(err)
179	}
180
181	// system should be: [{text: ...}, {cachePoint: {type: default}}]
182	if len(req.System) != 2 {
183		t.Fatalf("system len = %d, want 2 (text + cachePoint)", len(req.System))
184	}
185	if _, ok := req.System[0]["text"]; !ok {
186		t.Errorf("system[0] missing text key: %v", req.System[0])
187	}
188	if _, ok := req.System[1]["cachePoint"]; !ok {
189		t.Errorf("system[1] missing cachePoint key: %v", req.System[1])
190	}
191
192	// last user message should end with a cachePoint block
193	if len(req.Messages) == 0 {
194		t.Fatal("no messages")
195	}
196	lastMsg := req.Messages[len(req.Messages)-1]
197	if lastMsg.Role != "user" {
198		t.Fatalf("last message role = %q, want user", lastMsg.Role)
199	}
200	lastBlock := lastMsg.Content[len(lastMsg.Content)-1]
201	if _, ok := lastBlock["cachePoint"]; !ok {
202		t.Errorf("last user message final block missing cachePoint: %v", lastBlock)
203	}
204}
205
206func TestBedrockBuildRequestNoCachingNovaModel(t *testing.T) {
207	// A Nova model (PriceCacheWrite == 0) should NOT get cachePoint markers.
208	client := &bedrockClient{region: "us-east-1"}
209	req, err := client.buildRequest(Request{
210		Model:  "amazon.nova-pro-v1:0",
211		System: "You are helpful.",
212		Messages: []Message{
213			{Role: RoleUser, Content: []Content{TextBlock{Text: "hi"}}},
214		},
215	})
216	if err != nil {
217		t.Fatal(err)
218	}
219
220	// system should be plain: [{text: ...}] with no cachePoint
221	if len(req.System) != 1 {
222		t.Fatalf("system len = %d, want 1 (text only)", len(req.System))
223	}
224	if _, ok := req.System[0]["cachePoint"]; ok {
225		t.Errorf("Nova system unexpectedly contains cachePoint")
226	}
227
228	// last user message should NOT end with a cachePoint block
229	if len(req.Messages) == 0 {
230		t.Fatal("no messages")
231	}
232	lastMsg := req.Messages[len(req.Messages)-1]
233	for _, block := range lastMsg.Content {
234		if _, ok := block["cachePoint"]; ok {
235			t.Errorf("Nova user message unexpectedly contains cachePoint: %v", block)
236		}
237	}
238}
239
240func TestBedrockBuildRequestCachingMultiTurn(t *testing.T) {
241	// In a multi-turn conversation the cachePoint should be on the LAST
242	// user message only, not on earlier ones.
243	client := &bedrockClient{region: "us-east-1"}
244	req, err := client.buildRequest(Request{
245		Model: "anthropic.claude-sonnet-4-5-20250929-v1:0",
246		Messages: []Message{
247			{Role: RoleUser, Content: []Content{TextBlock{Text: "turn 1"}}},
248			{Role: RoleAssistant, Content: []Content{TextBlock{Text: "response 1"}}},
249			{Role: RoleUser, Content: []Content{TextBlock{Text: "turn 2"}}},
250		},
251	})
252	if err != nil {
253		t.Fatal(err)
254	}
255
256	// Find all user messages and verify only the last has a cachePoint.
257	var userMsgs []bedrockMessage
258	for _, m := range req.Messages {
259		if m.Role == "user" {
260			userMsgs = append(userMsgs, m)
261		}
262	}
263	if len(userMsgs) != 2 {
264		t.Fatalf("expected 2 user messages, got %d", len(userMsgs))
265	}
266
267	// First user message: no cachePoint
268	for _, block := range userMsgs[0].Content {
269		if _, ok := block["cachePoint"]; ok {
270			t.Errorf("first user message should not have cachePoint: %v", block)
271		}
272	}
273
274	// Last user message: ends with cachePoint
275	lastContent := userMsgs[len(userMsgs)-1].Content
276	lastBlock := lastContent[len(lastContent)-1]
277	if _, ok := lastBlock["cachePoint"]; !ok {
278		t.Errorf("last user message should end with cachePoint, got: %v", lastBlock)
279	}
280}
281
282func TestBedrockBuildRequestCachingNoSystemNoCrash(t *testing.T) {
283	// No system prompt: system array should be nil, not a bare cachePoint.
284	client := &bedrockClient{region: "us-east-1"}
285	req, err := client.buildRequest(Request{
286		Model:    "anthropic.claude-sonnet-4-5-20250929-v1:0",
287		Messages: []Message{{Role: RoleUser, Content: []Content{TextBlock{Text: "hi"}}}},
288	})
289	if err != nil {
290		t.Fatal(err)
291	}
292	if len(req.System) != 0 {
293		t.Errorf("empty system prompt should produce nil system, got: %v", req.System)
294	}
295	// Message cachePoint still present
296	lastBlock := req.Messages[0].Content[len(req.Messages[0].Content)-1]
297	if _, ok := lastBlock["cachePoint"]; !ok {
298		t.Errorf("user message should still have cachePoint when system is empty")
299	}
300}
301
302func TestBedrockTagLastUserCache(t *testing.T) {
303	msgs := []bedrockMessage{
304		{Role: "user", Content: []map[string]interface{}{{"text": "hi"}}},
305		{Role: "assistant", Content: []map[string]interface{}{{"text": "hello"}}},
306		{Role: "user", Content: []map[string]interface{}{{"text": "followup"}}},
307	}
308	bedrockTagLastUserCache(msgs)
309
310	// Only the last user message (index 2) should have a cachePoint appended.
311	last := msgs[2].Content
312	if _, ok := last[len(last)-1]["cachePoint"]; !ok {
313		t.Errorf("last user message should end with cachePoint")
314	}
315	// The first user message should be untouched.
316	if len(msgs[0].Content) != 1 {
317		t.Errorf("first user message content len = %d, want 1", len(msgs[0].Content))
318	}
319	if _, ok := msgs[0].Content[0]["cachePoint"]; ok {
320		t.Errorf("first user message should not have cachePoint")
321	}
322}
323
324func TestBedrockTagLastUserCacheEmpty(t *testing.T) {
325	// Should not panic on empty or assistant-only history.
326	bedrockTagLastUserCache(nil)
327	bedrockTagLastUserCache([]bedrockMessage{})
328	msgs := []bedrockMessage{
329		{Role: "assistant", Content: []map[string]interface{}{{"text": "hi"}}},
330	}
331	bedrockTagLastUserCache(msgs) // should not panic, no user message to tag
332}
333
334func TestBedrockBuildRequestCachingWireJSON(t *testing.T) {
335	// Verify the JSON shape Bedrock actually receives has the right keys.
336	client := &bedrockClient{region: "us-east-1"}
337	breq, err := client.buildRequest(Request{
338		Model:  "anthropic.claude-sonnet-4-5-20250929-v1:0",
339		System: "be helpful",
340		Messages: []Message{
341			{Role: RoleUser, Content: []Content{TextBlock{Text: "hello"}}},
342		},
343	})
344	if err != nil {
345		t.Fatal(err)
346	}
347	b, err := json.Marshal(breq)
348	if err != nil {
349		t.Fatal(err)
350	}
351	s := string(b)
352	if !strings.Contains(s, `"cachePoint"`) {
353		t.Errorf("serialised request missing cachePoint key: %s", s)
354	}
355	if !strings.Contains(s, `"type":"default"`) {
356		t.Errorf("serialised request missing cachePoint type:default: %s", s)
357	}
358}
359
360// TestBedrockBuildRequestBtwToolHistory reproduces the /btw HTTP 400 bug:
361// the side-chat sends no tools but the frozen main transcript contains
362// toolUse / toolResult blocks. Bedrock rejects such a request unless
363// toolConfig is present. buildRequest must inject a stub toolConfig.
364func TestBedrockBuildRequestBtwToolHistory(t *testing.T) {
365	client := &bedrockClient{region: "us-east-1"}
366	req, err := client.buildRequest(Request{
367		Model: "amazon.nova-pro-v1:0",
368		// No tools — simulates /btw side-chat.
369		Messages: []Message{
370			// Frozen main transcript: one tool call + result already happened.
371			{Role: RoleAssistant, Content: []Content{
372				ToolCallBlock{ID: "tc-1", Name: "bash", Arguments: json.RawMessage(`{"command":"ls"}`)},
373			}},
374			{Role: RoleTool, Content: []Content{
375				ToolResultBlock{CallID: "tc-1", Content: []Content{TextBlock{Text: "file.go"}}},
376			}},
377			// Side-chat question appended by btwDialog.submit.
378			{Role: RoleUser, Content: []Content{TextBlock{Text: "what does that file do?"}}},
379		},
380	})
381	if err != nil {
382		t.Fatal(err)
383	}
384	if req.ToolConfig == nil {
385		t.Fatal("buildRequest should inject a stub toolConfig when history has tool blocks and req.Tools is empty, but ToolConfig is nil")
386	}
387	if len(req.ToolConfig.Tools) == 0 {
388		t.Fatal("stub toolConfig should have at least one tool")
389	}
390	// The stub must have a valid name and a non-nil schema so Bedrock won't
391	// reject it for a different reason.
392	stub := req.ToolConfig.Tools[0]
393	if stub.ToolSpec.Name == "" {
394		t.Error("stub tool name must not be empty")
395	}
396	if stub.ToolSpec.InputSchema.JSON == nil {
397		t.Error("stub tool schema must not be nil")
398	}
399	// Serialised wire payload must include toolConfig.
400	b, err := json.Marshal(req)
401	if err != nil {
402		t.Fatal(err)
403	}
404	if !strings.Contains(string(b), `"toolConfig"`) {
405		t.Errorf("serialised request missing toolConfig: %s", b)
406	}
407}
408
409// TestBedrockBuildRequestNoToolHistoryNoStub ensures that when there are
410// no tool blocks in the history and no tools in the request (ordinary
411// conversational call), we do NOT inject a toolConfig at all — the extra
412// field is unnecessary and some models may behave differently with it.
413func TestBedrockBuildRequestNoToolHistoryNoStub(t *testing.T) {
414	client := &bedrockClient{region: "us-east-1"}
415	req, err := client.buildRequest(Request{
416		Model: "amazon.nova-pro-v1:0",
417		Messages: []Message{
418			{Role: RoleUser, Content: []Content{TextBlock{Text: "hello"}}},
419		},
420	})
421	if err != nil {
422		t.Fatal(err)
423	}
424	if req.ToolConfig != nil {
425		t.Errorf("expected nil ToolConfig for plain conversation, got: %+v", req.ToolConfig)
426	}
427}
428
429func TestBedrockBuildRequestSkipsEmptyToolMessages(t *testing.T) {
430	client := &bedrockClient{}
431	req, err := client.buildRequest(Request{Messages: []Message{
432		{Role: RoleTool, Content: []Content{ToolResultBlock{CallID: "missing", Content: []Content{TextBlock{Text: "orphan"}}}}},
433		{Role: RoleUser, Content: []Content{TextBlock{Text: "hello"}}},
434	}})
435	if err != nil {
436		t.Fatal(err)
437	}
438	if len(req.Messages) != 1 {
439		t.Fatalf("got %d bedrock messages, want 1: %+v", len(req.Messages), req.Messages)
440	}
441	if req.Messages[0].Role != "user" {
442		t.Fatalf("role = %s, want user", req.Messages[0].Role)
443	}
444}
445
446func TestBedrockEventPayloadHelpers(t *testing.T) {
447	wrapped := []byte(`{"contentBlockDelta":{"contentBlockIndex":0,"delta":{"text":"Hello"}}}`)
448	if got := bedrockEventTypeFromPayload(wrapped); got != "contentBlockDelta" {
449		t.Fatalf("event type = %q, want contentBlockDelta", got)
450	}
451	var delta struct {
452		ContentBlockIndex int `json:"contentBlockIndex"`
453		Delta             struct {
454			Text string `json:"text"`
455		} `json:"delta"`
456	}
457	if err := unmarshalBedrockEventPayload(wrapped, "contentBlockDelta", &delta); err != nil {
458		t.Fatal(err)
459	}
460	if delta.ContentBlockIndex != 0 || delta.Delta.Text != "Hello" {
461		t.Fatalf("unexpected wrapped delta: %+v", delta)
462	}
463
464	direct := []byte(`{"contentBlockIndex":1,"delta":{"text":"world"}}`)
465	if err := unmarshalBedrockEventPayload(direct, "contentBlockDelta", &delta); err != nil {
466		t.Fatal(err)
467	}
468	if delta.ContentBlockIndex != 1 || delta.Delta.Text != "world" {
469		t.Fatalf("unexpected direct delta: %+v", delta)
470	}
471}
472
473func TestResolveBedrockInferenceProfileID(t *testing.T) {
474	cases := []struct {
475		model  string
476		region string
477		want   string
478	}{
479		// Bare Anthropic foundation IDs get the region-matched prefix.
480		{"anthropic.claude-sonnet-4-5-20250929-v1:0", "us-east-1", "us.anthropic.claude-sonnet-4-5-20250929-v1:0"},
481		{"anthropic.claude-sonnet-4-5-20250929-v1:0", "eu-central-1", "eu.anthropic.claude-sonnet-4-5-20250929-v1:0"},
482		{"anthropic.claude-opus-4-6-v1", "ap-southeast-2", "apac.anthropic.claude-opus-4-6-v1"},
483		{"anthropic.claude-opus-4-6-v1", "us-gov-west-1", "us-gov.anthropic.claude-opus-4-6-v1"},
484		{"deepseek.r1-v1:0", "eu-west-1", "eu.deepseek.r1-v1:0"},
485		// Empty region defaults to us.
486		{"anthropic.claude-opus-4-6-v1", "", "us.anthropic.claude-opus-4-6-v1"},
487		// Already-prefixed IDs are left untouched.
488		{"eu.anthropic.claude-sonnet-4-5-20250929-v1:0", "us-east-1", "eu.anthropic.claude-sonnet-4-5-20250929-v1:0"},
489		{"global.anthropic.claude-opus-4-6-v1", "us-east-1", "global.anthropic.claude-opus-4-6-v1"},
490		{"us.anthropic.claude-opus-4-6-v1", "eu-central-1", "us.anthropic.claude-opus-4-6-v1"},
491		// ARNs are passed through verbatim.
492		{"arn:aws:bedrock:us-east-1:123:inference-profile/us.anthropic.claude-opus-4-6-v1", "eu-west-1", "arn:aws:bedrock:us-east-1:123:inference-profile/us.anthropic.claude-opus-4-6-v1"},
493		// Families that don't need a profile are untouched.
494		{"amazon.nova-pro-v1:0", "us-east-1", "amazon.nova-pro-v1:0"},
495		{"", "us-east-1", ""},
496	}
497	for _, c := range cases {
498		if got := resolveBedrockInferenceProfileID(c.model, c.region); got != c.want {
499			t.Errorf("resolveBedrockInferenceProfileID(%q, %q) = %q; want %q", c.model, c.region, got, c.want)
500		}
501	}
502}

Keyboard shortcuts

Single-key shortcuts are on. ctrl k and esc work whatever this says.

ctrl k :
go to a section, branch, tag, file or repository
?
this list
j k
move the row cursor down / up
enter
open the row under the cursor; on Files, a directory opens or closes in place
↑ ↓ ← → home end
in the file tree (focus on a row): previous / next row, close / open a directory or go to its parent / first entry, first / last row. They never open a file
l h
on Files: open a directory or a file / close a directory or go up to its parent, then to the parent page. Elsewhere: into the side panel (on a merge request, its files) / back to the content; there j k move between its controls
g g G
first / last row
/
filter the list; on Files, find a file anywhere in the repository (esc clears, enter opens the best match)
b
switch branch or tag
g s g f g c g b g m g i g r g a
tabs: summary, files, commits, branches & tags, merge requests, issues, releases, actions (when available)
g h
home
w
wide mode: use the whole width of the screen, or not (remembered)
y
copy a permalink pinned to the commit
d u x
diff: split / unified / collapse or expand all
e E
diff: the whole file in view / every file, with the lines the hunks hide
n
merge request: next note in the diff
esc
close a dialog, leave a field

Branches 1

Tags 477

↑↓ moveenter openesc close