631 lines
26 KiB
Go
631 lines
26 KiB
Go
package llm
|
|
|
|
import (
|
|
"context"
|
|
"crypto/sha256"
|
|
"encoding/hex"
|
|
"encoding/json"
|
|
"io"
|
|
"net/http"
|
|
"net/http/httptest"
|
|
"reflect"
|
|
"strconv"
|
|
"strings"
|
|
"testing"
|
|
)
|
|
|
|
const (
|
|
rcModel = "fixture-model"
|
|
rcEOFReason = "tool call stream ended before [DONE]"
|
|
)
|
|
|
|
var rcTools = []ToolDecl{
|
|
{Name: "read_PRIVATE_I1", Parameters: json.RawMessage(`{"type":"object","required":["path"],"properties":{"path":{"type":"string"}}}`)},
|
|
{Name: "list_PRIVATE_I1", Parameters: json.RawMessage(`{"type":"object","properties":{"dir":{"type":"string"}}}`)},
|
|
}
|
|
|
|
func rcFrame(v any) string {
|
|
raw, err := json.Marshal(v)
|
|
if err != nil {
|
|
panic(err)
|
|
}
|
|
return "data: " + string(raw) + "\n\n"
|
|
}
|
|
|
|
func rcChunk(choices ...map[string]any) string {
|
|
return rcFrame(map[string]any{"choices": choices})
|
|
}
|
|
|
|
func rcFrag(index int, id, name, args string) map[string]any {
|
|
fn := map[string]any{}
|
|
if name != "" {
|
|
fn["name"] = name
|
|
}
|
|
if args != "" {
|
|
fn["arguments"] = args
|
|
}
|
|
frag := map[string]any{"index": index, "function": fn}
|
|
if id != "" {
|
|
frag["id"] = id
|
|
}
|
|
return frag
|
|
}
|
|
|
|
func rcToolChoice(finish string, frags ...map[string]any) map[string]any {
|
|
choice := map[string]any{"delta": map[string]any{"tool_calls": frags}}
|
|
if finish != "" {
|
|
choice["finish_reason"] = finish
|
|
}
|
|
return choice
|
|
}
|
|
|
|
func rcFinishChunk(reason string) string {
|
|
return rcChunk(map[string]any{"delta": map[string]any{}, "finish_reason": reason})
|
|
}
|
|
|
|
var (
|
|
rcText = rcChunk(map[string]any{"delta": map[string]any{"content": "PRIVATE_TEXT_I1", "reasoning_content": "PRIVATE_THINK_I1"}})
|
|
rcCall1 = rcChunk(rcToolChoice("", rcFrag(0, "call_PRIVATE_I1", "read_PRIVATE_I1", `{"path":`)))
|
|
rcCall2 = rcChunk(rcToolChoice("", rcFrag(0, "", "", `"PRIVATE_I1.txt"}`)))
|
|
rcFinish = rcFinishChunk("tool_calls")
|
|
rcUsage = rcFrame(map[string]any{"choices": []any{}, "usage": map[string]any{"prompt_tokens": 5, "completion_tokens": 2, "total_tokens": 7}})
|
|
rcDone = "data: [DONE]\n\n"
|
|
rcStream = rcText + rcCall1 + rcCall2 + rcFinish + rcUsage
|
|
rcCall = ToolCall{ID: "call_PRIVATE_I1", Name: "read_PRIVATE_I1", Arguments: `{"path":"PRIVATE_I1.txt"}`}
|
|
|
|
rcErrorFrame = rcFrame(map[string]any{"error": map[string]any{"message": "PRIVATE_PROVIDER_I1"}})
|
|
)
|
|
|
|
// rcSingleCall streams one complete call in one fragment with the given parts.
|
|
func rcSingleCall(id, name, args string) string {
|
|
return rcText + rcChunk(rcToolChoice("", rcFrag(0, id, name, args))) + rcFinish + rcUsage
|
|
}
|
|
|
|
func rcWrite(stream string) http.HandlerFunc {
|
|
return func(w http.ResponseWriter, r *http.Request) {
|
|
w.Header().Set("Content-Type", "text/event-stream")
|
|
_, _ = io.WriteString(w, stream)
|
|
w.(http.Flusher).Flush()
|
|
}
|
|
}
|
|
|
|
func rcHTTP2(t *testing.T, h http.Handler) *httptest.Server {
|
|
t.Helper()
|
|
server := httptest.NewUnstartedServer(h)
|
|
server.EnableHTTP2 = true
|
|
server.StartTLS()
|
|
t.Cleanup(server.Close)
|
|
return server
|
|
}
|
|
|
|
func rcTLS1(t *testing.T, h http.Handler) *httptest.Server {
|
|
t.Helper()
|
|
server := httptest.NewTLSServer(h)
|
|
t.Cleanup(server.Close)
|
|
return server
|
|
}
|
|
|
|
func rcPlain(t *testing.T, h http.Handler) *httptest.Server {
|
|
t.Helper()
|
|
server := httptest.NewServer(h)
|
|
t.Cleanup(server.Close)
|
|
return server
|
|
}
|
|
|
|
// rcHTTPClient trusts server like server.Client() but owns its connection
|
|
// pool, so each run starts on a fresh connection and HTTP/2 stream ids match.
|
|
func rcHTTPClient(t *testing.T, server *httptest.Server) *http.Client {
|
|
t.Helper()
|
|
base := server.Client().Transport.(*http.Transport)
|
|
transport := &http.Transport{TLSClientConfig: base.TLSClientConfig.Clone(), ForceAttemptHTTP2: true}
|
|
t.Cleanup(transport.CloseIdleConnections)
|
|
return &http.Client{Transport: transport}
|
|
}
|
|
|
|
func rcClientFor(t *testing.T, server *httptest.Server) *Client {
|
|
return &Client{BaseURL: server.URL, APIKey: "fixture", HTTPClient: rcHTTPClient(t, server)}
|
|
}
|
|
|
|
func rcContract(model, base string) *RouteContractEOF {
|
|
return &RouteContractEOF{Model: model, BaseURL: base, IdentitySHA256: RouteContractIdentity(model, base)}
|
|
}
|
|
|
|
func rcFixtureClient(resp func() *http.Response) *Client {
|
|
return &Client{BaseURL: "https://fixture.invalid", APIKey: "fixture", HTTPClient: &http.Client{
|
|
Transport: terminalRoundTripFunc(func(r *http.Request) (*http.Response, error) {
|
|
res := resp()
|
|
res.Request = r
|
|
return res, nil
|
|
}),
|
|
}}
|
|
}
|
|
|
|
type rcSetup struct {
|
|
client *Client
|
|
contract *RouteContractEOF
|
|
opts StreamOptions
|
|
tools []ToolDecl
|
|
}
|
|
|
|
type rcResult struct {
|
|
end Event
|
|
releasedMessages int
|
|
releasedCalls int
|
|
}
|
|
|
|
func rcRun(t *testing.T, ctx context.Context, base *Client, contract *RouteContractEOF, tools []ToolDecl, opts StreamOptions, onEvent func(Event)) rcResult {
|
|
t.Helper()
|
|
client := *base
|
|
client.RouteContract = contract
|
|
var res rcResult
|
|
terminals := 0
|
|
for ev := range client.Stream(ctx, rcModel, []Message{{Role: RoleUser, Content: "hello"}}, tools, opts) {
|
|
if onEvent != nil {
|
|
onEvent(ev)
|
|
}
|
|
if ev.Kind != KindMessageEnd && ev.Kind != KindError {
|
|
continue
|
|
}
|
|
terminals++
|
|
res.end = ev
|
|
if ev.Kind == KindMessageEnd && ev.Err == nil && len(ev.Message.ToolCalls) > 0 {
|
|
res.releasedMessages++
|
|
res.releasedCalls += len(ev.Message.ToolCalls)
|
|
}
|
|
}
|
|
if terminals != 1 {
|
|
t.Fatalf("terminal events = %d, want 1", terminals)
|
|
}
|
|
rcRequireSafeLabels(t, res.end)
|
|
return res
|
|
}
|
|
|
|
// rcRequireSafeLabels is the sentinel scan: the new labels hold only their
|
|
// fixed enums and the observation never carries content, tool identity,
|
|
// arguments or error text.
|
|
func rcRequireSafeLabels(t *testing.T, end Event) {
|
|
t.Helper()
|
|
obs := end.TerminalObservation
|
|
if obs == nil {
|
|
t.Fatalf("terminal without observation: %#v", end)
|
|
}
|
|
switch end.CompletionBasis {
|
|
case "", CompletionDoneMarker, CompletionRouteContractEOF:
|
|
default:
|
|
t.Fatalf("completion basis outside enum: %q", end.CompletionBasis)
|
|
}
|
|
if obs.CompletionBasis != end.CompletionBasis {
|
|
t.Fatalf("observation basis %q != event basis %q", obs.CompletionBasis, end.CompletionBasis)
|
|
}
|
|
if c := obs.ContractCheck; c != "" && (len(c) != 2 || c[0] != 'C' || c[1] < '1' || c[1] > '9') {
|
|
t.Fatalf("contract check outside enum: %q", c)
|
|
}
|
|
raw, err := json.Marshal(obs)
|
|
if err != nil {
|
|
t.Fatal(err)
|
|
}
|
|
for _, secret := range []string{"PRIVATE", "[DONE]", "tool call", "fixture-model"} {
|
|
if strings.Contains(string(raw), secret) {
|
|
t.Fatalf("observation leaked %q: %s", secret, raw)
|
|
}
|
|
}
|
|
for _, label := range []string{end.CompletionBasis, obs.CompletionBasis, obs.ContractCheck} {
|
|
if strings.Contains(label, "PRIVATE") || strings.Contains(label, " ") {
|
|
t.Fatalf("label carries content: %q", label)
|
|
}
|
|
}
|
|
}
|
|
|
|
func rcStreamCase(server func(*testing.T, http.Handler) *httptest.Server, stream string) func(*testing.T) rcSetup {
|
|
return func(t *testing.T) rcSetup {
|
|
s := server(t, rcWrite(stream))
|
|
return rcSetup{client: rcClientFor(t, s), contract: rcContract(rcModel, s.URL)}
|
|
}
|
|
}
|
|
|
|
func rcH2(stream string) func(*testing.T) rcSetup { return rcStreamCase(rcHTTP2, stream) }
|
|
|
|
func rcRedirectCase(target func(t *testing.T) string) func(*testing.T) rcSetup {
|
|
return func(t *testing.T) rcSetup {
|
|
location := target(t)
|
|
mux := http.NewServeMux()
|
|
mux.HandleFunc("/chat/completions", func(w http.ResponseWriter, r *http.Request) {
|
|
http.Redirect(w, r, location, http.StatusTemporaryRedirect)
|
|
})
|
|
mux.HandleFunc("/final/chat/completions", rcWrite(rcStream))
|
|
s := rcHTTP2(t, mux)
|
|
return rcSetup{client: rcClientFor(t, s), contract: rcContract(rcModel, s.URL)}
|
|
}
|
|
}
|
|
|
|
type rcCase struct {
|
|
name string
|
|
setup func(*testing.T) rcSetup
|
|
// want is done_marker, route_contract_eof, off, or a check id C1..C9.
|
|
want string
|
|
// reason is the contract-off failure text; empty skips the check.
|
|
reason string
|
|
calls []ToolCall
|
|
noUsage bool
|
|
}
|
|
|
|
func rcCases() []rcCase {
|
|
twoCalls := rcText + rcCall1 + rcCall2 +
|
|
rcChunk(rcToolChoice("", rcFrag(1, "call_PRIVATE_I1_b", "list_PRIVATE_I1", `{}`))) + rcFinish + rcUsage
|
|
return []rcCase{
|
|
// [DONE] keeps today's path whether or not the contract is configured.
|
|
{name: "done/contract-on", setup: rcH2(rcStream + rcDone), want: CompletionDoneMarker},
|
|
{name: "done/contract-c1-failing", setup: func(t *testing.T) rcSetup {
|
|
s := rcH2(rcStream + rcDone)(t)
|
|
s.contract = rcContract("other-model", s.client.BaseURL)
|
|
return s
|
|
}, want: CompletionDoneMarker},
|
|
{name: "done/known-length", setup: rcKnownLength(rcStream + rcDone), want: CompletionDoneMarker},
|
|
|
|
// Released only when every condition holds.
|
|
{name: "release/http2", setup: rcH2(rcStream), want: CompletionRouteContractEOF},
|
|
{name: "release/known-length-match", setup: rcKnownLength(rcStream), want: CompletionRouteContractEOF},
|
|
{name: "release/usage-before-finish", setup: rcH2(rcText + rcUsage + rcCall1 + rcCall2 + rcFinish), want: CompletionRouteContractEOF},
|
|
{name: "release/usage-after-finish", setup: rcH2(rcText + rcCall1 + rcCall2 + rcFinish + rcUsage), want: CompletionRouteContractEOF},
|
|
{name: "release/fragment-in-finish-chunk", setup: rcH2(rcText + rcCall1 +
|
|
rcChunk(rcToolChoice("tool_calls", rcFrag(0, "", "", `"PRIVATE_I1.txt"}`))) + rcUsage), want: CompletionRouteContractEOF},
|
|
{name: "release/two-calls", setup: rcH2(twoCalls), want: CompletionRouteContractEOF,
|
|
calls: []ToolCall{rcCall, {ID: "call_PRIVATE_I1_b", Name: "list_PRIVATE_I1", Arguments: `{}`}}},
|
|
{name: "release/api-base-override-to-contract", setup: func(t *testing.T) rcSetup {
|
|
s := rcH2(rcStream)(t)
|
|
s.opts.APIBase = s.client.BaseURL + "/"
|
|
s.client.BaseURL = "https://unused.invalid"
|
|
return s
|
|
}, want: CompletionRouteContractEOF},
|
|
|
|
// The zero value is off.
|
|
{name: "off/zero-contract", setup: func(t *testing.T) rcSetup {
|
|
s := rcH2(rcStream)(t)
|
|
s.contract = &RouteContractEOF{}
|
|
return s
|
|
}, want: "off", reason: rcEOFReason},
|
|
|
|
// C1 route identity.
|
|
{name: "c1/model-mismatch", setup: func(t *testing.T) rcSetup {
|
|
s := rcH2(rcStream)(t)
|
|
s.contract = rcContract("other-model", s.client.BaseURL)
|
|
return s
|
|
}, want: "C1", reason: rcEOFReason},
|
|
{name: "c1/base-http", setup: rcStreamCase(rcPlain, rcStream), want: "C1", reason: rcEOFReason},
|
|
{name: "c1/base-userinfo", setup: func(t *testing.T) rcSetup {
|
|
s := rcH2(rcStream)(t)
|
|
s.client.BaseURL = strings.Replace(s.client.BaseURL, "https://", "https://user@", 1)
|
|
return s
|
|
}, want: "C1", reason: rcEOFReason},
|
|
{name: "c1/base-query", setup: func(t *testing.T) rcSetup {
|
|
s := rcH2(rcStream)(t)
|
|
s.client.BaseURL += "?x=1"
|
|
return s
|
|
}, want: "C1", reason: rcEOFReason},
|
|
{name: "c1/digest-mismatch", setup: func(t *testing.T) rcSetup {
|
|
s := rcH2(rcStream)(t)
|
|
s.contract.IdentitySHA256 = strings.Repeat("0", 64)
|
|
return s
|
|
}, want: "C1", reason: rcEOFReason},
|
|
{name: "c1/digest-uppercase", setup: func(t *testing.T) rcSetup {
|
|
s := rcH2(rcStream)(t)
|
|
s.contract.IdentitySHA256 = strings.ToUpper(s.contract.IdentitySHA256)
|
|
return s
|
|
}, want: "C1", reason: rcEOFReason},
|
|
{name: "c1/protocol-not-openai-chat", setup: func(t *testing.T) rcSetup {
|
|
s := rcH2(rcStream)(t)
|
|
s.client.Protocol = ProtocolGoogle
|
|
return s
|
|
}, want: "C1", reason: rcEOFReason},
|
|
{name: "c1/api-base-override-elsewhere", setup: func(t *testing.T) rcSetup {
|
|
other := rcHTTP2(t, rcWrite(rcStream))
|
|
s := rcH2(rcStream)(t)
|
|
s.opts.APIBase = other.URL
|
|
s.client.HTTPClient = rcHTTPClient(t, other)
|
|
return s
|
|
}, want: "C1", reason: rcEOFReason},
|
|
{name: "c1/redirect-other-host", setup: rcRedirectCase(func(t *testing.T) string {
|
|
return rcHTTP2(t, rcWrite(rcStream)).URL + "/chat/completions"
|
|
}), want: "C1", reason: rcEOFReason},
|
|
{name: "c1/redirect-same-host-other-path", setup: rcRedirectCase(func(t *testing.T) string {
|
|
return "/final/chat/completions"
|
|
}), want: "C1", reason: rcEOFReason},
|
|
{name: "c1/redirect-https-downgrade", setup: rcRedirectCase(func(t *testing.T) string {
|
|
return rcPlain(t, rcWrite(rcStream)).URL + "/chat/completions"
|
|
}), want: "C1", reason: rcEOFReason},
|
|
|
|
// C2 body boundary.
|
|
{name: "c2/known-length-mismatch", setup: func(t *testing.T) rcSetup {
|
|
client := rcFixtureClient(func() *http.Response {
|
|
return &http.Response{StatusCode: http.StatusOK, ProtoMajor: 1, ProtoMinor: 1,
|
|
ContentLength: int64(len(rcStream) + 64), Header: http.Header{},
|
|
Body: io.NopCloser(strings.NewReader(rcStream))}
|
|
})
|
|
return rcSetup{client: client, contract: rcContract(rcModel, client.BaseURL)}
|
|
}, want: "C2", reason: rcEOFReason},
|
|
{name: "c2/unknown-framing", setup: func(t *testing.T) rcSetup {
|
|
client := rcFixtureClient(func() *http.Response {
|
|
return &http.Response{StatusCode: http.StatusOK, ContentLength: -1, Header: http.Header{},
|
|
Body: io.NopCloser(strings.NewReader(rcStream))}
|
|
})
|
|
return rcSetup{client: client, contract: rcContract(rcModel, client.BaseURL)}
|
|
}, want: "C2", reason: rcEOFReason},
|
|
{name: "c2/chunked", setup: rcStreamCase(rcTLS1, rcStream), want: "C2", reason: rcEOFReason},
|
|
{name: "c2/close-delimited", setup: func(t *testing.T) rcSetup {
|
|
s := rcTLS1(t, http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
|
|
w.Header().Set("Content-Type", "text/event-stream")
|
|
w.Header().Set("Transfer-Encoding", "identity")
|
|
w.WriteHeader(http.StatusOK)
|
|
_, _ = io.WriteString(w, rcStream)
|
|
w.(http.Flusher).Flush()
|
|
}))
|
|
return rcSetup{client: rcClientFor(t, s), contract: rcContract(rcModel, s.URL)}
|
|
}, want: "C2", reason: rcEOFReason},
|
|
// A reset or dropped stream never reaches body EOF, so the boundary
|
|
// condition fails before the read-outcome condition is evaluated.
|
|
{name: "c2/http2-rst-stream", setup: func(t *testing.T) rcSetup {
|
|
s := rcHTTP2(t, http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
|
|
rcWrite(rcStream)(w, r)
|
|
panic(http.ErrAbortHandler)
|
|
}))
|
|
return rcSetup{client: rcClientFor(t, s), contract: rcContract(rcModel, s.URL)}
|
|
}, want: "C2", reason: "stream read failed after reported usage"},
|
|
{name: "c2/mid-stream-disconnect", setup: func(t *testing.T) rcSetup {
|
|
s := rcTLS1(t, http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
|
|
rcWrite(rcStream)(w, r)
|
|
conn, _, err := w.(http.Hijacker).Hijack()
|
|
if err == nil {
|
|
_ = conn.Close()
|
|
}
|
|
}))
|
|
return rcSetup{client: rcClientFor(t, s), contract: rcContract(rcModel, s.URL)}
|
|
}, want: "C2", reason: "stream read failed after reported usage"},
|
|
|
|
// The parser stops at a bad frame; with a trailing delimiter it never
|
|
// reads to body EOF, so the boundary condition fails first.
|
|
{name: "c2/invalid-frame-before-eof", setup: rcH2(rcStream + "data: PRIVATE_not_json\n\n"), want: "C2",
|
|
reason: "malformed stream chunk after reported usage"},
|
|
{name: "c2/provider-error-before-eof", setup: rcH2(rcStream + rcErrorFrame), want: "C2",
|
|
reason: "provider error after reported usage: PRIVATE_PROVIDER_I1"},
|
|
|
|
// C3 read outcome: the bad frame is the last bytes, so body EOF is seen.
|
|
{name: "c3/half-frame", setup: rcH2(rcStream + `data: {"choices":[{"delta":{"content":"PRIVATE_HALF`), want: "C3",
|
|
reason: "malformed stream chunk after reported usage"},
|
|
{name: "c3/invalid-frame", setup: rcH2(rcStream + "data: PRIVATE_not_json"), want: "C3",
|
|
reason: "malformed stream chunk after reported usage"},
|
|
{name: "c3/provider-error-after-usage", setup: rcH2(rcStream + strings.TrimRight(rcErrorFrame, "\n")),
|
|
want: "C3", reason: "provider error after reported usage: PRIVATE_PROVIDER_I1"},
|
|
{name: "c3/provider-error-before-usage", setup: rcH2(rcText + rcCall1 + rcCall2 + rcFinish + strings.TrimRight(rcErrorFrame, "\n")),
|
|
want: "C3", reason: "PRIVATE_PROVIDER_I1", noUsage: true},
|
|
|
|
// C4 finish reason and choices.
|
|
{name: "c4/two-finish-chunks", setup: rcH2(rcText + rcCall1 + rcCall2 + rcFinish + rcFinish + rcUsage), want: "C4", reason: rcEOFReason},
|
|
{name: "c4/finish-stop-with-tools", setup: rcH2(rcText + rcCall1 + rcCall2 + rcFinishChunk("stop") + rcUsage), want: "C4", reason: rcEOFReason},
|
|
{name: "c4/finish-length-with-tools", setup: rcH2(rcText + rcCall1 + rcCall2 + rcFinishChunk("length") + rcUsage), want: "C4", reason: rcEOFReason},
|
|
{name: "c4/multi-choice-chunk", setup: rcH2(rcText +
|
|
rcChunk(map[string]any{"delta": map[string]any{"content": "a"}}, map[string]any{"delta": map[string]any{"content": "b"}}) +
|
|
rcCall1 + rcCall2 + rcFinish + rcUsage), want: "C4", reason: rcEOFReason},
|
|
{name: "c4/truncated-at-frame-boundary-without-finish", setup: rcH2(rcText + rcCall1 + rcCall2 + rcUsage), want: "C4", reason: rcEOFReason},
|
|
|
|
// C5 usage.
|
|
{name: "c5/missing-usage", setup: rcH2(rcText + rcCall1 + rcCall2 + rcFinish), want: "C5", reason: rcEOFReason, noUsage: true},
|
|
|
|
// C6 fragments after finish.
|
|
{name: "c6/fragment-after-finish-chunk", setup: rcH2(rcText + rcCall1 + rcFinish + rcCall2 + rcUsage), want: "C6", reason: rcEOFReason},
|
|
|
|
// C7 call identity and argument shape.
|
|
{name: "c7/missing-id", setup: rcH2(rcSingleCall("", "read_PRIVATE_I1", `{"path":"PRIVATE_I1.txt"}`)), want: "C7", reason: rcEOFReason},
|
|
{name: "c7/missing-name", setup: rcH2(rcSingleCall("call_PRIVATE_I1", "", `{"path":"PRIVATE_I1.txt"}`)), want: "C7", reason: rcEOFReason},
|
|
{name: "c7/array-arguments", setup: rcH2(rcSingleCall("call_PRIVATE_I1", "read_PRIVATE_I1", `["PRIVATE_I1.txt"]`)), want: "C7", reason: rcEOFReason},
|
|
{name: "c7/broken-arguments", setup: rcH2(rcSingleCall("call_PRIVATE_I1", "read_PRIVATE_I1", `{"path":"PRIVATE_I1.txt"`)), want: "C7", reason: rcEOFReason},
|
|
{name: "c7/finish-without-call", setup: rcH2(rcText + rcFinish + rcUsage), want: "C7", reason: "tool call finish reason without a tool call"},
|
|
|
|
// C8 declared tools and argument schema.
|
|
{name: "c8/unknown-tool", setup: rcH2(rcSingleCall("call_PRIVATE_I1", "write_PRIVATE_I1", `{"path":"PRIVATE_I1.txt"}`)), want: "C8", reason: rcEOFReason},
|
|
{name: "c8/no-tools-declared", setup: func(t *testing.T) rcSetup {
|
|
s := rcH2(rcStream)(t)
|
|
s.tools = []ToolDecl{}
|
|
return s
|
|
}, want: "C8", reason: rcEOFReason},
|
|
{name: "c8/missing-required", setup: rcH2(rcSingleCall("call_PRIVATE_I1", "read_PRIVATE_I1", `{}`)), want: "C8", reason: rcEOFReason},
|
|
{name: "c8/blank-required", setup: rcH2(rcSingleCall("call_PRIVATE_I1", "read_PRIVATE_I1", `{"path":" "}`)), want: "C8", reason: rcEOFReason},
|
|
{name: "c8/wrong-property-type", setup: rcH2(rcSingleCall("call_PRIVATE_I1", "read_PRIVATE_I1", `{"path":7}`)), want: "C8", reason: rcEOFReason},
|
|
{name: "c8/unknown-argument", setup: rcH2(rcSingleCall("call_PRIVATE_I1", "read_PRIVATE_I1", `{"path":"a","PRIVATE":1}`)), want: "C8", reason: rcEOFReason},
|
|
|
|
// C9 tail event.
|
|
{name: "c9/comment-tail", setup: rcH2(rcStream + ": PRIVATE_keepalive\n\n"), want: "C9", reason: rcEOFReason},
|
|
}
|
|
}
|
|
|
|
func rcKnownLength(stream string) func(*testing.T) rcSetup {
|
|
return func(t *testing.T) rcSetup {
|
|
s := rcTLS1(t, http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
|
|
w.Header().Set("Content-Type", "text/event-stream")
|
|
w.Header().Set("Content-Length", strconv.Itoa(len(stream)))
|
|
_, _ = io.WriteString(w, stream)
|
|
}))
|
|
return rcSetup{client: rcClientFor(t, s), contract: rcContract(rcModel, s.URL)}
|
|
}
|
|
}
|
|
|
|
func TestRouteContractEOFMatrix(t *testing.T) {
|
|
for _, tc := range rcCases() {
|
|
t.Run(tc.name, func(t *testing.T) {
|
|
s := tc.setup(t)
|
|
tools := s.tools
|
|
if tools == nil {
|
|
tools = rcTools
|
|
}
|
|
wantCalls := tc.calls
|
|
if wantCalls == nil {
|
|
wantCalls = []ToolCall{rcCall}
|
|
}
|
|
wantUsage := 7
|
|
if tc.noUsage {
|
|
wantUsage = 0
|
|
}
|
|
off := rcRun(t, context.Background(), s.client, nil, tools, s.opts, nil)
|
|
on := rcRun(t, context.Background(), s.client, s.contract, tools, s.opts, nil)
|
|
for _, r := range []rcResult{off, on} {
|
|
if r.end.Usage.TotalTokens != wantUsage {
|
|
t.Fatalf("usage = %+v, want total %d", r.end.Usage, wantUsage)
|
|
}
|
|
}
|
|
if off.end.TerminalObservation.ContractCheck != "" {
|
|
t.Fatalf("contract-off stream carried a contract check: %+v", off.end.TerminalObservation)
|
|
}
|
|
if tc.reason != "" && off.end.ErrorMessage != tc.reason {
|
|
t.Fatalf("contract-off failure = %q, want %q", off.end.ErrorMessage, tc.reason)
|
|
}
|
|
|
|
switch tc.want {
|
|
case CompletionDoneMarker:
|
|
if !reflect.DeepEqual(on.end, off.end) {
|
|
t.Fatalf("contract changed a [DONE] terminal:\n on=%#v\noff=%#v", on.end, off.end)
|
|
}
|
|
if on.end.Err != nil || !on.end.StreamComplete || on.end.CompletionBasis != CompletionDoneMarker ||
|
|
on.end.TerminalObservation.ContractCheck != "" || on.end.TerminalObservation.ReadOutcome != "done_marker" {
|
|
t.Fatalf("[DONE] terminal not marked done_marker: %#v", on.end)
|
|
}
|
|
rcRequireReleased(t, on, wantCalls)
|
|
case CompletionRouteContractEOF:
|
|
rcRequireRejected(t, off)
|
|
if off.end.CompletionBasis != "" || off.end.ErrorMessage != rcEOFReason {
|
|
t.Fatalf("contract-off stream was not rejected unchanged: %#v", off.end)
|
|
}
|
|
end := on.end
|
|
if end.Err != nil || end.StreamComplete || end.CompletionBasis != CompletionRouteContractEOF ||
|
|
end.TerminalObservation.ContractCheck != "" || end.TerminalObservation.ReadOutcome != "clean_eof" ||
|
|
end.Message.StopReason != StopToolCalls || end.FinishReason != "tool_calls" || end.HTTPStatus != http.StatusOK {
|
|
t.Fatalf("contract stream was not released: %#v obs=%+v", end, end.TerminalObservation)
|
|
}
|
|
if end.Message.Content != off.end.Message.Content || !reflect.DeepEqual(end.Message.Thinking, off.end.Message.Thinking) {
|
|
t.Fatalf("released message changed content: %#v vs %#v", end.Message, off.end.Message)
|
|
}
|
|
rcRequireReleased(t, on, wantCalls)
|
|
case "off":
|
|
if !reflect.DeepEqual(on.end, off.end) {
|
|
t.Fatalf("zero contract changed the terminal:\n on=%#v\noff=%#v", on.end, off.end)
|
|
}
|
|
rcRequireRejected(t, on)
|
|
default:
|
|
rcRequireRejected(t, off)
|
|
rcRequireRejected(t, on)
|
|
if got := on.end.TerminalObservation.ContractCheck; got != tc.want {
|
|
t.Fatalf("contract check = %q, want %q (obs=%+v)", got, tc.want, on.end.TerminalObservation)
|
|
}
|
|
stripped := on.end
|
|
obs := *on.end.TerminalObservation
|
|
obs.ContractCheck = ""
|
|
stripped.TerminalObservation = &obs
|
|
if !reflect.DeepEqual(stripped, off.end) {
|
|
t.Fatalf("rejection differs from contract-off beyond contractCheck:\n on=%#v %+v\noff=%#v %+v",
|
|
stripped, stripped.TerminalObservation, off.end, off.end.TerminalObservation)
|
|
}
|
|
}
|
|
})
|
|
}
|
|
}
|
|
|
|
func rcRequireReleased(t *testing.T, r rcResult, calls []ToolCall) {
|
|
t.Helper()
|
|
if r.releasedMessages != 1 || r.releasedCalls != len(calls) || !reflect.DeepEqual(r.end.Message.ToolCalls, calls) {
|
|
t.Fatalf("released %d messages / %d calls %#v, want 1 / %#v", r.releasedMessages, r.releasedCalls, r.end.Message.ToolCalls, calls)
|
|
}
|
|
}
|
|
|
|
func rcRequireRejected(t *testing.T, r rcResult) {
|
|
t.Helper()
|
|
end := r.end
|
|
if r.releasedMessages != 0 || r.releasedCalls != 0 || len(end.Message.ToolCalls) != 0 || end.Err == nil ||
|
|
end.StreamComplete || end.CompletionBasis != "" || end.TerminalObservation.CompletionBasis != "" {
|
|
t.Fatalf("stream was released: %#v", end)
|
|
}
|
|
}
|
|
|
|
func TestRouteContractEOFCancelDuringReadAborts(t *testing.T) {
|
|
server := rcHTTP2(t, http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
|
|
rcWrite(rcText+rcCall1+rcCall2+rcFinish+rcUsage)(w, r)
|
|
<-r.Context().Done()
|
|
}))
|
|
client := rcClientFor(t, server)
|
|
ctx, cancel := context.WithCancel(context.Background())
|
|
defer cancel()
|
|
res := rcRun(t, ctx, client, rcContract(rcModel, server.URL), rcTools, StreamOptions{}, func(ev Event) {
|
|
if ev.Kind == KindToolCallDelta {
|
|
cancel()
|
|
}
|
|
})
|
|
end := res.end
|
|
if end.Kind != KindError || end.Message.StopReason != StopAborted || end.TerminalObservation.ReadOutcome != "cancelled" ||
|
|
end.TerminalObservation.ContractCheck != "" || end.CompletionBasis != "" || res.releasedMessages != 0 ||
|
|
res.releasedCalls != 0 || len(end.Message.ToolCalls) != 0 {
|
|
t.Fatalf("cancelled contract stream was not aborted: %#v obs=%+v", end, end.TerminalObservation)
|
|
}
|
|
}
|
|
|
|
func TestParseSSEUnchangedWithoutDone(t *testing.T) {
|
|
var end Event
|
|
for ev := range ParseSSE(context.Background(), strings.NewReader(rcStream)) {
|
|
end = ev
|
|
}
|
|
if end.Kind != KindMessageEnd || end.ErrorMessage != rcEOFReason || len(end.Message.ToolCalls) != 0 ||
|
|
end.StreamComplete || end.CompletionBasis != "" || end.Usage.TotalTokens != 7 ||
|
|
end.TerminalObservation == nil || end.TerminalObservation.CompletionBasis != "" ||
|
|
end.TerminalObservation.ContractCheck != "" || end.TerminalObservation.BodyBoundary != nil {
|
|
t.Fatalf("ParseSSE changed without [DONE]: %#v", end)
|
|
}
|
|
raw, err := json.Marshal(end.TerminalObservation)
|
|
if err != nil || strings.Contains(string(raw), "completionBasis") || strings.Contains(string(raw), "contractCheck") {
|
|
t.Fatalf("ParseSSE observation JSON changed: %s err=%v", raw, err)
|
|
}
|
|
var done Event
|
|
for ev := range ParseSSE(context.Background(), strings.NewReader(rcStream+rcDone)) {
|
|
done = ev
|
|
}
|
|
if done.Err != nil || !done.StreamComplete || done.CompletionBasis != "" || done.TerminalObservation.CompletionBasis != "" ||
|
|
!reflect.DeepEqual(done.Message.ToolCalls, []ToolCall{rcCall}) {
|
|
t.Fatalf("ParseSSE changed with [DONE]: %#v", done)
|
|
}
|
|
}
|
|
|
|
func TestRouteContractIdentity(t *testing.T) {
|
|
sum := sha256.Sum256([]byte("openai-chat\nm\nhttps://api.example.com/v1"))
|
|
if got, want := RouteContractIdentity("m", "https://api.example.com/v1"), hex.EncodeToString(sum[:]); got != want {
|
|
t.Fatalf("identity = %s, want %s", got, want)
|
|
}
|
|
if got := RouteContractIdentity("m", "https://api.example.com/v1"); got != strings.ToLower(got) || len(got) != 64 {
|
|
t.Fatalf("identity is not lowercase hex: %s", got)
|
|
}
|
|
}
|
|
|
|
func TestNormalizeRouteBaseURL(t *testing.T) {
|
|
tests := []struct {
|
|
raw, want string
|
|
ok bool
|
|
}{
|
|
{"https://api.example.com/v1", "https://api.example.com/v1", true},
|
|
{"HTTPS://API.Example.COM/v1/", "https://api.example.com/v1", true},
|
|
{"https://api.example.com:8443/v1//", "https://api.example.com:8443/v1", true},
|
|
{"https://api.example.com", "https://api.example.com", true},
|
|
{"https://api.example.com/V1", "https://api.example.com/V1", true},
|
|
{"http://api.example.com/v1", "", false},
|
|
{"https://user@api.example.com/v1", "", false},
|
|
{"https://user:pw@api.example.com/v1", "", false},
|
|
{"https://api.example.com/v1?x=1", "", false},
|
|
{"https://api.example.com/v1?", "", false},
|
|
{"https://api.example.com/v1#frag", "", false},
|
|
{"https://api.example.com/v1#", "", false},
|
|
{"https:api.example.com", "", false},
|
|
{"https:///v1", "", false},
|
|
{"", "", false},
|
|
{"://bad", "", false},
|
|
}
|
|
for _, tc := range tests {
|
|
got, ok := NormalizeRouteBaseURL(tc.raw)
|
|
if got != tc.want || ok != tc.ok {
|
|
t.Fatalf("NormalizeRouteBaseURL(%q) = %q, %v; want %q, %v", tc.raw, got, ok, tc.want, tc.ok)
|
|
}
|
|
}
|
|
}
|