mirror of
https://github.com/router-for-me/CLIProxyAPI.git
synced 2026-10-20 22:51:24 +08:00
470 lines
22 KiB
Go
470 lines
22 KiB
Go
package thinking_test
|
|
|
|
import (
|
|
"bytes"
|
|
"testing"
|
|
|
|
"github.com/router-for-me/CLIProxyAPI/v8/internal/registry"
|
|
"github.com/router-for-me/CLIProxyAPI/v8/internal/thinking"
|
|
_ "github.com/router-for-me/CLIProxyAPI/v8/internal/thinking/provider/codex"
|
|
log "github.com/sirupsen/logrus"
|
|
logtest "github.com/sirupsen/logrus/hooks/test"
|
|
"github.com/tidwall/gjson"
|
|
)
|
|
|
|
func TestExtractCodexReasoningEffortWithConfigurationUpdate(t *testing.T) {
|
|
tests := []struct {
|
|
name string
|
|
provider string
|
|
model string
|
|
body string
|
|
wantRequestEffort string
|
|
wantTranslatedEffort string
|
|
}{
|
|
{
|
|
name: "codex extracts configuration_update effort over top-level",
|
|
provider: "codex",
|
|
model: "gpt-6-astra",
|
|
body: `{"model":"gpt-6-astra","reasoning":{"effort":"xhigh","summary":"auto"},"input":[{"type":"configuration_update","reasoning":{"effort":"low"}}]}`,
|
|
wantRequestEffort: "low",
|
|
wantTranslatedEffort: "low",
|
|
},
|
|
{
|
|
name: "openai-response extracts configuration_update effort over top-level",
|
|
provider: "openai-response",
|
|
model: "gpt-6-astra",
|
|
body: `{"model":"gpt-6-astra","reasoning":{"effort":"xhigh","summary":"auto"},"input":[{"type":"configuration_update","reasoning":{"effort":"low"}}]}`,
|
|
wantRequestEffort: "low",
|
|
wantTranslatedEffort: "low",
|
|
},
|
|
{
|
|
name: "codex picks latest configuration_update when multiple are present",
|
|
provider: "codex",
|
|
model: "gpt-6-astra",
|
|
body: `{
|
|
"model":"gpt-6-astra",
|
|
"reasoning":{"effort":"xhigh","summary":"auto"},
|
|
"input":[
|
|
{"type":"configuration_update","reasoning":{"effort":"low"}},
|
|
{"role":"user","content":"second turn"},
|
|
{"type":"configuration_update","reasoning":{"effort":"medium"}}
|
|
]
|
|
}`,
|
|
wantRequestEffort: "medium",
|
|
wantTranslatedEffort: "medium",
|
|
},
|
|
{
|
|
name: "codex falls back to top-level when configuration_update has no reasoning effort",
|
|
provider: "codex",
|
|
model: "gpt-6-astra",
|
|
body: `{"model":"gpt-6-astra","reasoning":{"effort":"xhigh","summary":"auto"},"input":[{"type":"configuration_update","tools":[]}]}`,
|
|
wantRequestEffort: "xhigh",
|
|
wantTranslatedEffort: "xhigh",
|
|
},
|
|
{
|
|
name: "codex handles configuration_update effort none",
|
|
provider: "codex",
|
|
model: "gpt-6-astra",
|
|
body: `{"model":"gpt-6-astra","reasoning":{"effort":"xhigh"},"input":[{"type":"configuration_update","reasoning":{"effort":"none"}}]}`,
|
|
wantRequestEffort: "none",
|
|
wantTranslatedEffort: "none",
|
|
},
|
|
{
|
|
name: "codex handles configuration_update effort auto",
|
|
provider: "codex",
|
|
model: "gpt-6-astra",
|
|
body: `{"model":"gpt-6-astra","reasoning":{"effort":"xhigh"},"input":[{"type":"configuration_update","reasoning":{"effort":"auto"}}]}`,
|
|
wantRequestEffort: "auto",
|
|
wantTranslatedEffort: "auto",
|
|
},
|
|
{
|
|
name: "trailing configuration_update without reasoning effort preserves earlier effort",
|
|
provider: "codex",
|
|
model: "gpt-6-astra",
|
|
body: `{
|
|
"model":"gpt-6-astra",
|
|
"reasoning":{"effort":"xhigh"},
|
|
"input":[
|
|
{"type":"configuration_update","reasoning":{"effort":"low"}},
|
|
{"type":"configuration_update","tools":[]}
|
|
]
|
|
}`,
|
|
wantRequestEffort: "low",
|
|
wantTranslatedEffort: "low",
|
|
},
|
|
{
|
|
name: "invalid json returns empty effort",
|
|
provider: "codex",
|
|
model: "gpt-6-astra",
|
|
body: `{"model":"gpt-6-astra","reasoning":{"effort":"xhigh"},"input":`,
|
|
wantRequestEffort: "",
|
|
wantTranslatedEffort: "",
|
|
},
|
|
{
|
|
name: "codex falls back to top-level when input has no configuration_update",
|
|
provider: "codex",
|
|
model: "gpt-6-astra",
|
|
body: `{"model":"gpt-6-astra","reasoning":{"effort":"xhigh"},"input":[{"role":"user","content":"hello"}]}`,
|
|
wantRequestEffort: "xhigh",
|
|
wantTranslatedEffort: "xhigh",
|
|
},
|
|
{
|
|
name: "source configuration_update takes precedence over suffix for request effort",
|
|
provider: "codex",
|
|
model: "gpt-6-astra(high)",
|
|
body: `{"model":"gpt-6-astra","reasoning":{"effort":"xhigh"},"input":[{"type":"configuration_update","reasoning":{"effort":"low"}}]}`,
|
|
wantRequestEffort: "low",
|
|
wantTranslatedEffort: "low",
|
|
},
|
|
{
|
|
name: "suffix takes precedence over top-level without updates",
|
|
provider: "openai-response",
|
|
model: "gpt-6-astra(high)",
|
|
body: `{"reasoning":{"effort":"xhigh"},"input":[{"type":"configuration_update","tools":[]}]}`,
|
|
wantRequestEffort: "high",
|
|
wantTranslatedEffort: "xhigh",
|
|
},
|
|
}
|
|
|
|
for _, tc := range tests {
|
|
t.Run(tc.name, func(t *testing.T) {
|
|
gotRequest := thinking.ExtractReasoningEffort([]byte(tc.body), tc.provider, tc.model)
|
|
if gotRequest != tc.wantRequestEffort {
|
|
t.Fatalf("ExtractReasoningEffort() = %q, want %q", gotRequest, tc.wantRequestEffort)
|
|
}
|
|
|
|
gotTranslated := thinking.ExtractTranslatedReasoningEffort([]byte(tc.body), tc.provider)
|
|
if gotTranslated != tc.wantTranslatedEffort {
|
|
t.Fatalf("ExtractTranslatedReasoningEffort() = %q, want %q", gotTranslated, tc.wantTranslatedEffort)
|
|
}
|
|
})
|
|
}
|
|
}
|
|
|
|
func TestExtractCodexReasoningEffortWithConfigurationUpdateTargetRouting(t *testing.T) {
|
|
const source = `{"reasoning":{"effort":"xhigh"},"input":[{"type":"configuration_update","reasoning":{"effort":"medium"}},{"type":"configuration_update","reasoning":{"effort":"low"}},{"type":"configuration_update","reasoning":{"effort":null}},{"role":"user","content":"ok"}]}`
|
|
for _, supported := range []bool{true, false} {
|
|
name := "unsupported"
|
|
if supported {
|
|
name = "supported"
|
|
}
|
|
t.Run(name, func(t *testing.T) {
|
|
info := ®istry.ModelInfo{ID: "opaque-route", Type: "codex", SupportConfigurationUpdate: supported, Thinking: ®istry.ThinkingSupport{Levels: []string{"low", "high", "xhigh"}}}
|
|
body, err := thinking.ApplyThinkingWithModelInfoAndSummary([]byte(source), []byte(source), "opaque-route(high)", "openai-response", "codex", "codex", info, thinking.SummaryConfig{})
|
|
if err != nil {
|
|
t.Fatalf("ApplyThinkingWithModelInfoAndSummary() error = %v", err)
|
|
}
|
|
if got := thinking.ExtractReasoningEffort([]byte(source), "openai-response", "opaque-route(high)"); got != "low" {
|
|
t.Fatalf("source effort = %q, want low", got)
|
|
}
|
|
want := "high"
|
|
if supported {
|
|
want = "low"
|
|
}
|
|
if got := thinking.ExtractTranslatedReasoningEffort(body, "codex"); got != want {
|
|
t.Fatalf("final effort = %q, want %q; body=%s", got, want, body)
|
|
}
|
|
if got := gjson.GetBytes(body, "reasoning.effort").String(); got != "high" {
|
|
t.Fatalf("final top-level effort = %q, want high; body=%s", got, body)
|
|
}
|
|
if hasUpdate := gjson.GetBytes(body, "input.0.type").String() == "configuration_update"; hasUpdate != supported {
|
|
t.Fatalf("final update present = %t, want %t; body=%s", hasUpdate, supported, body)
|
|
}
|
|
})
|
|
}
|
|
}
|
|
|
|
func TestApplyConfigurationUpdateRouting(t *testing.T) {
|
|
const model = "configured-responses"
|
|
tests := []struct {
|
|
name string
|
|
body string
|
|
format string
|
|
suffix string
|
|
supported bool
|
|
noThinking bool
|
|
skipEffort bool
|
|
wantEffort string
|
|
wantInput string
|
|
wantSame bool
|
|
}{
|
|
{
|
|
name: "supported native request preserves baseline and updates byte for byte",
|
|
body: `{"reasoning":{"effort":"xhigh","summary":"auto"},"input":[{"type":"configuration_update","reasoning":{"effort":"low"}},{"role":"user","content":"ok"}]}`,
|
|
supported: true, wantEffort: "xhigh", wantInput: `[{"type":"configuration_update","reasoning":{"effort":"low"}},{"role":"user","content":"ok"}]`, wantSame: true,
|
|
},
|
|
{
|
|
name: "supported no-effort request remains unchanged",
|
|
body: `{"reasoning":{"summary":"auto"},"input":[{"type":"configuration_update","tools":[]},{"role":"user","content":"ok"}]}`,
|
|
supported: true, wantInput: `[{"type":"configuration_update","tools":[]},{"role":"user","content":"ok"}]`, wantSame: true,
|
|
},
|
|
{
|
|
name: "supported native request without thinking metadata preserves baseline summary and updates",
|
|
body: `{"reasoning":{"effort":"xhigh","summary":"auto"},"input":[{"type":"configuration_update","reasoning":{"effort":"low"}},{"role":"user","content":"ok"}]}`,
|
|
supported: true,
|
|
noThinking: true,
|
|
wantEffort: "xhigh",
|
|
wantInput: `[{"type":"configuration_update","reasoning":{"effort":"low"}},{"role":"user","content":"ok"}]`,
|
|
wantSame: true,
|
|
},
|
|
{
|
|
name: "supported native suffix still strips effort without thinking metadata",
|
|
body: `{"reasoning":{"effort":"xhigh","summary":"auto"},"input":[{"type":"configuration_update","reasoning":{"effort":"low"}},{"role":"user","content":"ok"}]}`,
|
|
supported: true,
|
|
noThinking: true,
|
|
suffix: "high",
|
|
wantInput: `[{"type":"configuration_update","reasoning":{"effort":"low"}},{"role":"user","content":"ok"}]`,
|
|
},
|
|
{
|
|
name: "supported suffix only rewrites the top-level effort",
|
|
body: `{"reasoning":{"effort":"xhigh","summary":"auto","other":7},"input":[{"type":"configuration_update","reasoning":{"effort":"low"}},{"role":"user","content":"ok"}]}`,
|
|
supported: true, suffix: "high", wantEffort: "high", wantInput: `[{"type":"configuration_update","reasoning":{"effort":"low"}},{"role":"user","content":"ok"}]`,
|
|
},
|
|
{
|
|
name: "supported suffix keeps effective input effort for current turn",
|
|
body: `{"reasoning":{"summary":"auto"},"input":[{"type":"configuration_update","reasoning":{"effort":"low"}},{"role":"user","content":"ok"}]}`,
|
|
supported: true, suffix: "high", wantEffort: "high", wantInput: `[{"type":"configuration_update","reasoning":{"effort":"low"}},{"role":"user","content":"ok"}]`,
|
|
},
|
|
{
|
|
name: "supported invalid suffix leaves native payload untouched",
|
|
body: `{"reasoning":{"generate_summary":"auto"},"input":[{"type":"configuration_update","reasoning":{"effort":"low"}},{"role":"user","content":"ok"}]}`,
|
|
supported: true, suffix: "invalid", wantInput: `[{"type":"configuration_update","reasoning":{"effort":"low"}},{"role":"user","content":"ok"}]`, wantSame: true,
|
|
},
|
|
{
|
|
name: "unsupported latest nonempty update wins and other input order stays unchanged",
|
|
body: `{"reasoning":{"effort":"xhigh","summary":"auto","other":7},"input":[{"role":"user","content":"first"},{"type":"configuration_update","reasoning":{"effort":"low"}},{"type":"configuration_update","reasoning":{"effort":" "}},{"role":"assistant","content":"reply"},{"type":"configuration_update","reasoning":{"effort":"medium"}},{"type":"configuration_update","tools":[]},{"role":"user","content":"last"}]}`,
|
|
wantEffort: "medium", wantInput: `[{"role":"user","content":"first"},{"role":"assistant","content":"reply"},{"role":"user","content":"last"}]`,
|
|
},
|
|
{
|
|
name: "unsupported ignores empty and nonstring efforts",
|
|
body: `{"reasoning":{"summary":"auto"},"input":[{"type":"configuration_update","reasoning":{"effort":"low"}},{"type":"configuration_update","reasoning":{"effort":42}},{"type":"configuration_update","reasoning":{"effort":null}},{"type":"configuration_update","reasoning":{"effort":" "}},{"role":"user","content":"ok"}]}`,
|
|
wantEffort: "low", wantInput: `[{"role":"user","content":"ok"}]`,
|
|
},
|
|
{
|
|
name: "unsupported suffix takes precedence and still removes updates",
|
|
body: `{"reasoning":{"effort":"xhigh","summary":"auto"},"input":[{"type":"configuration_update","reasoning":{"effort":"low"}},{"role":"user","content":"ok"}]}`,
|
|
suffix: "high", wantEffort: "high", wantInput: `[{"role":"user","content":"ok"}]`,
|
|
},
|
|
{
|
|
name: "unsupported without top-level reasoning promotes the last update",
|
|
body: `{"input":[{"type":"configuration_update","reasoning":{"effort":"low"}},{"role":"user","content":"ok"}]}`,
|
|
wantEffort: "low", wantInput: `[{"role":"user","content":"ok"}]`,
|
|
},
|
|
{
|
|
name: "unsupported removes updates with no effort without inventing one",
|
|
body: `{"reasoning":{"summary":"auto"},"input":[{"type":"configuration_update","tools":[]},{"role":"user","content":"ok"}]}`,
|
|
wantInput: `[{"role":"user","content":"ok"}]`,
|
|
},
|
|
{
|
|
name: "unsupported without thinking support still removes updates",
|
|
body: `{"reasoning":{"summary":"auto"},"input":[{"type":"configuration_update","tools":[]},{"role":"user","content":"ok"}]}`,
|
|
noThinking: true, wantInput: `[{"role":"user","content":"ok"}]`,
|
|
},
|
|
{
|
|
name: "unsupported without thinking support strips effort but keeps summary",
|
|
body: `{"reasoning":{"effort":"xhigh","summary":"auto","other":7},"input":[{"type":"configuration_update","reasoning":{"effort":"low"}},{"role":"user","content":"ok"}]}`,
|
|
noThinking: true, wantInput: `[{"role":"user","content":"ok"}]`,
|
|
},
|
|
{
|
|
name: "unsupported openai-response alias removes updates",
|
|
body: `{"reasoning":{"effort":"xhigh","summary":"auto"},"input":[{"type":"configuration_update","reasoning":{"effort":"low"}},{"role":"user","content":"ok"}]}`,
|
|
format: "openai-response", wantEffort: "low", wantInput: `[{"role":"user","content":"ok"}]`,
|
|
},
|
|
{
|
|
name: "invalid JSON is untouched",
|
|
body: `{"reasoning":{"effort":"xhigh"},"input":[`,
|
|
wantSame: true, skipEffort: true,
|
|
},
|
|
{
|
|
name: "nonarray input is untouched",
|
|
body: `{"reasoning":{"summary":"auto"},"input":{"type":"configuration_update","reasoning":{"effort":"low"}}}`,
|
|
wantInput: `{"type":"configuration_update","reasoning":{"effort":"low"}}`, wantSame: true,
|
|
},
|
|
}
|
|
for _, tc := range tests {
|
|
t.Run(tc.name, func(t *testing.T) {
|
|
format := tc.format
|
|
if format == "" {
|
|
format = "codex"
|
|
}
|
|
thinkingSupport := ®istry.ThinkingSupport{Levels: []string{"low", "medium", "high", "xhigh"}}
|
|
if tc.noThinking {
|
|
thinkingSupport = nil
|
|
}
|
|
info := ®istry.ModelInfo{ID: model, Type: "codex", SupportConfigurationUpdate: tc.supported, Thinking: thinkingSupport}
|
|
body := []byte(tc.body)
|
|
applied, err := thinking.ApplyThinkingWithModelInfo(body, body, model+suffixForTest(tc.suffix), format, format, "codex", info)
|
|
if err != nil {
|
|
t.Fatalf("ApplyThinkingWithModelInfo() error = %v", err)
|
|
}
|
|
if tc.wantSame && !bytes.Equal(applied, body) {
|
|
t.Fatalf("body changed: got %s, want %s", applied, body)
|
|
}
|
|
if !tc.skipEffort {
|
|
if got := gjson.GetBytes(applied, "reasoning.effort").String(); got != tc.wantEffort {
|
|
t.Errorf("reasoning.effort = %q, want %q; body=%s", got, tc.wantEffort, applied)
|
|
}
|
|
}
|
|
if tc.wantInput != "" && gjson.GetBytes(applied, "input").Raw != tc.wantInput {
|
|
t.Errorf("input = %s, want %s; body=%s", gjson.GetBytes(applied, "input").Raw, tc.wantInput, applied)
|
|
}
|
|
if gjson.GetBytes(body, "reasoning.summary").Exists() {
|
|
if got := gjson.GetBytes(applied, "reasoning.summary").String(); got != "auto" {
|
|
t.Errorf("reasoning.summary = %q, want auto; body=%s", got, applied)
|
|
}
|
|
}
|
|
if gjson.GetBytes(body, "reasoning.other").Exists() && gjson.GetBytes(applied, "reasoning.other").Int() != 7 {
|
|
t.Errorf("reasoning.other changed: %s", applied)
|
|
}
|
|
if tc.supported && tc.suffix != "" {
|
|
if got := thinking.ExtractTranslatedReasoningEffort(applied, format); got != "low" {
|
|
t.Errorf("effective effort = %q, want low; body=%s", got, applied)
|
|
}
|
|
}
|
|
})
|
|
}
|
|
}
|
|
|
|
func suffixForTest(suffix string) string {
|
|
if suffix == "" {
|
|
return ""
|
|
}
|
|
return "(" + suffix + ")"
|
|
}
|
|
|
|
func TestApplyThinkingLogsNativeResponsesEffectiveEffort(t *testing.T) {
|
|
logger := log.StandardLogger()
|
|
previousLevel := logger.GetLevel()
|
|
previousHooks := logger.ReplaceHooks(make(log.LevelHooks))
|
|
hook := logtest.NewLocal(logger)
|
|
logger.SetLevel(log.DebugLevel)
|
|
t.Cleanup(func() {
|
|
logger.ReplaceHooks(previousHooks)
|
|
logger.SetLevel(previousLevel)
|
|
})
|
|
|
|
tests := []struct {
|
|
name string
|
|
body string
|
|
wantEffort string
|
|
wantLevel thinking.ThinkingLevel
|
|
wantMode thinking.ThinkingMode
|
|
wantBaseline thinking.ThinkingLevel
|
|
}{
|
|
{
|
|
name: "update overrides top-level baseline",
|
|
body: `{"model":"gpt-6-sol","reasoning":{"effort":"high","summary":"auto"},"input":[{"type":"configuration_update","reasoning":{"effort":"xhigh"}},{"role":"user","content":"ok"}]}`,
|
|
wantEffort: "xhigh", wantLevel: thinking.LevelXHigh, wantMode: thinking.ModeLevel, wantBaseline: thinking.LevelHigh,
|
|
},
|
|
{
|
|
name: "latest update wins",
|
|
body: `{"model":"gpt-6-sol","reasoning":{"effort":"high"},"input":[{"type":"configuration_update","reasoning":{"effort":"xhigh"}},{"type":"configuration_update","reasoning":{"effort":"max"}}]}`,
|
|
wantEffort: "max", wantLevel: thinking.LevelMax, wantMode: thinking.ModeLevel, wantBaseline: thinking.LevelHigh,
|
|
},
|
|
{
|
|
name: "trailing empty and nonstring updates keep last effective effort",
|
|
body: `{"model":"gpt-6-sol","reasoning":{"effort":"high"},"input":[{"type":"configuration_update","reasoning":{"effort":"xhigh"}},{"type":"configuration_update","reasoning":{"effort":null}},{"type":"configuration_update","reasoning":{"effort":42}},{"type":"configuration_update","reasoning":{"effort":" "}}]}`,
|
|
wantEffort: "xhigh", wantLevel: thinking.LevelXHigh, wantMode: thinking.ModeLevel, wantBaseline: thinking.LevelHigh,
|
|
},
|
|
{
|
|
name: "top-level fallback without update",
|
|
body: `{"model":"gpt-6-sol","reasoning":{"effort":"high"},"input":[{"role":"user","content":"ok"}]}`,
|
|
wantEffort: "high", wantLevel: thinking.LevelHigh, wantMode: thinking.ModeLevel, wantBaseline: thinking.LevelHigh,
|
|
},
|
|
{
|
|
name: "update without top-level effort",
|
|
body: `{"model":"gpt-6-sol","input":[{"type":"configuration_update","reasoning":{"effort":"xhigh"}}]}`,
|
|
wantEffort: "xhigh", wantLevel: thinking.LevelXHigh, wantMode: thinking.ModeLevel,
|
|
},
|
|
{
|
|
name: "update disables thinking",
|
|
body: `{"model":"gpt-6-sol","reasoning":{"effort":"high"},"input":[{"type":"configuration_update","reasoning":{"effort":"none"}}]}`,
|
|
wantEffort: "none", wantMode: thinking.ModeNone, wantBaseline: thinking.LevelHigh,
|
|
},
|
|
}
|
|
for _, tc := range tests {
|
|
t.Run(tc.name, func(t *testing.T) {
|
|
for _, bound := range []bool{false, true} {
|
|
name := "registry lookup"
|
|
if bound {
|
|
name = "resolved model"
|
|
}
|
|
t.Run(name, func(t *testing.T) {
|
|
hook.Reset()
|
|
body := []byte(tc.body)
|
|
var applied []byte
|
|
var err error
|
|
if bound {
|
|
modelInfo := registry.LookupModelInfo("gpt-6-sol", "codex")
|
|
if modelInfo == nil || !modelInfo.SupportConfigurationUpdate {
|
|
t.Fatal("gpt-6-sol must support configuration updates")
|
|
}
|
|
applied, err = thinking.ApplyThinkingWithModelInfo(body, body, "gpt-6-sol", "codex", "codex", "codex", modelInfo)
|
|
} else {
|
|
applied, err = thinking.ApplyThinking(body, "gpt-6-sol", "codex", "codex", "codex")
|
|
}
|
|
if err != nil {
|
|
t.Fatalf("ApplyThinking() error = %v", err)
|
|
}
|
|
if !bytes.Equal(applied, body) {
|
|
t.Fatalf("native Responses body changed: got %s, want %s", applied, body)
|
|
}
|
|
if got := thinking.ExtractTranslatedReasoningEffort(applied, "codex"); got != tc.wantEffort {
|
|
t.Errorf("effective effort = %q, want %q", got, tc.wantEffort)
|
|
}
|
|
|
|
for _, message := range []string{
|
|
"thinking: original config from request |",
|
|
"thinking: processed config to apply |",
|
|
} {
|
|
found := false
|
|
for _, entry := range hook.AllEntries() {
|
|
if entry.Message != message {
|
|
continue
|
|
}
|
|
found = true
|
|
if entry.Level != log.DebugLevel || entry.Data["provider"] != "codex" || entry.Data["model"] != "gpt-6-sol" ||
|
|
entry.Data["mode"] != tc.wantMode || entry.Data["budget"] != 0 || entry.Data["level"] != tc.wantLevel {
|
|
t.Errorf("unexpected native Responses log %q fields: %v", message, entry.Data)
|
|
}
|
|
baseline, hasBaseline := entry.Data["baseline_level"]
|
|
if tc.wantBaseline == "" && hasBaseline || tc.wantBaseline != "" && baseline != tc.wantBaseline {
|
|
t.Errorf("baseline_level = %v (present: %t), want %q", baseline, hasBaseline, tc.wantBaseline)
|
|
}
|
|
}
|
|
if !found {
|
|
t.Errorf("missing native Responses debug log %q", message)
|
|
}
|
|
}
|
|
})
|
|
}
|
|
})
|
|
}
|
|
}
|
|
|
|
func TestApplyThinkingPreservesCodexTopLevelReasoningEffortBaseline(t *testing.T) {
|
|
// Preserving request-level baseline sent upstream ensures prompt prefix cache preservation.
|
|
body := []byte(`{"model":"gpt-6-astra","reasoning":{"effort":"xhigh","summary":"auto"},"input":[{"type":"configuration_update","reasoning":{"effort":"low"}}]}`)
|
|
applied, err := thinking.ApplyThinking(body, "gpt-6-astra", "codex", "codex", "codex")
|
|
if err != nil {
|
|
t.Fatalf("ApplyThinking() error = %v", err)
|
|
}
|
|
|
|
// Verify top-level request baseline is preserved for upstream prompt caching.
|
|
if gotTopEffort := gjson.GetBytes(applied, "reasoning.effort").String(); gotTopEffort != "xhigh" {
|
|
t.Fatalf("ApplyThinking() top-level reasoning.effort = %q, want %q", gotTopEffort, "xhigh")
|
|
}
|
|
if gotSummary := gjson.GetBytes(applied, "reasoning.summary").String(); gotSummary != "auto" {
|
|
t.Fatalf("ApplyThinking() reasoning.summary = %q, want %q", gotSummary, "auto")
|
|
}
|
|
if gotInputEffort := gjson.GetBytes(applied, "input.0.reasoning.effort").String(); gotInputEffort != "low" {
|
|
t.Fatalf("ApplyThinking() input[0].reasoning.effort = %q, want %q", gotInputEffort, "low")
|
|
}
|
|
|
|
// Verify usage reporting extracts the effective in-turn effort.
|
|
effort := thinking.ExtractTranslatedReasoningEffort(applied, "codex")
|
|
if effort != "low" {
|
|
t.Fatalf("ExtractTranslatedReasoningEffort(applied) = %q, want %q", effort, "low")
|
|
}
|
|
}
|