Files
new-api/relaykit/relayconvert/internal/oai_responses/to_claude_messages_req.go
T
Calcium-Ion 86ac0f7745 refactor: extract protocol conversion layer into standalone relaykit module (#6369)
* test(relayconvert): add golden snapshot matrix and relaykit boundary guard

Phase 0 of the relaykit extraction plan: pin byte-level output of every
registered (from,to) request/response/stream conversion route, and
forbid kit-bound packages from growing host-only imports.

* wip(relayconvert): drop gin.Context from converter signatures; add convmeta draft

Phase 1 in progress: relayconvert now takes context.Context; host media
resolver adapts gin.Context back at the service boundary.

* refactor(relayconvert): decouple converters from RelayInfo, gin, and settings

Phase 1 of the relaykit extraction plan:
- converters now depend on convmeta.Meta (implemented by RelayInfo) instead
  of *relaycommon.RelayInfo; ClaudeConvertInfo and the format guesser move
  to convmeta with aliases left behind
- host settings reach converters via a convmeta.Options snapshot built in
  RelayInfo.ConvOptions; no more model_setting/reasoning global reads inside
  the conversion layer
- effort-suffix helpers move to service/relayconvert/reasoning (old package
  forwards); chat-to-responses upgrade policy moves to service (host routing
  logic, not conversion)
- golden conversion matrix unchanged

* test(relayconvert): tighten boundary — kit packages now free of gin/setting imports

* refactor(dto): drop gin and logger dependencies

Phase 2 (part 1): dto.Request.IsStream now takes *http.Request instead of
*gin.Context (Gemini's impl reads query/path off the std request); dto's
three logger calls become common.SysError. Boundary test allowlist is now
empty — kit-bound packages import no gin/setting/logger/model.

* refactor(kit): extract dependency-free kitutil; dto/types/relayconvert stop importing common

Phase 2 of the relaykit extraction plan:
- new service/relayconvert/kitutil holds the pure helpers the kit needs
  (JSON wrappers, pointer/string/uuid/timestamp utils, MaskSensitiveInfo,
  pluggable LogInfo/LogError hooks, Debug flag)
- dto, types, and all relayconvert packages now use kitutil; their only
  remaining internal deps are dto/types/constant
- common keeps every original symbol (MaskSensitiveInfo delegates to
  kitutil) so host code is untouched; main.go routes kit logging into
  common.SysLog/SysError and mirrors DebugEnabled
- golden conversion matrix unchanged

* refactor(kit): move EndpointType/FinishReason to types; OpenRouter dialect via Options

Kit packages (dto/types/relayconvert/reasonmap) no longer import constant:
- EndpointType and finish-reason values live in types; constant re-exports
- the OpenRouter special-case in claude->openai request conversion reads
  Options.OpenRouterDialect, set by the host from the channel type;
  InitChannelMeta invalidates the cached snapshot on channel switch

* refactor: extract relaykit submodule (dto/types/relayconvert/reasonmap)

Phase 3 of the relaykit extraction plan:
- new go module github.com/QuantumNous/new-api/relaykit containing dto
  (minus task family), types, relayconvert (with convmeta/kitutil/reasoning),
  and reasonmap; host consumes it via require + replace, go.work for dev
- task-family dto (task/suno/midjourney/video) stays in the host dto
  package; dual-consumer host files alias it as taskdto
- relaykit builds and tests standalone (GOWORK=off): no host imports,
  no gin, no DB, no settings
- golden conversion matrix unchanged

* build(docker): copy relaykit/go.mod before go mod download

The local-replace submodule's go.mod must exist inside the build context
for the main module graph to resolve.

* fix: address relaykit extraction regressions

* fix: address relaykit review regressions

* docs: document Meta nil receiver contract

* fix(relaykit): fail OpenAI→Claude conversion without max_tokens; reject negative default_max_tokens

The Claude Messages API requires max_tokens (omitting it is a 400
"Field required"), but with a nil Options.Claude.DefaultMaxTokens hook
the converters silently emitted a request the upstream is guaranteed to
reject. Both OpenAI Chat and Responses → Claude conversions now return
sharedclaude.ErrMissingMaxTokens when no path (client value, default
hook, thinking-adapter floor) supplied one. Unreachable in the host,
which always configures the hook.

Host side, claude.default_max_tokens now rejects negative values at the
option API before persisting — they would wrap into huge unsigned values
during conversion. Zero stays allowed: the current API treats
max_tokens: 0 as cache pre-warming.

* fix: make Gemini safety settings read path race-free
2026-07-27 15:56:21 +08:00

330 lines
9.5 KiB
Go

package oairesponses
import (
"fmt"
"strings"
"context"
"github.com/QuantumNous/new-api/relaykit/dto"
"github.com/QuantumNous/new-api/relaykit/relayconvert/convmeta"
relaymedia "github.com/QuantumNous/new-api/relaykit/relayconvert/internal/media"
sharedclaude "github.com/QuantumNous/new-api/relaykit/relayconvert/internal/shared/claude"
kitutil "github.com/QuantumNous/new-api/relaykit/relayconvert/kitutil"
)
func convertOpenAIResponsesRequestToClaudeMessages(c context.Context, info convmeta.Meta, request any) (any, error) {
responsesRequest, err := OpenAIResponsesRequestFromAny(request)
if err != nil {
return nil, err
}
return OpenAIResponsesRequestToClaudeMessages(c, info, responsesRequest)
}
func OpenAIResponsesRequestToClaudeMessages(c context.Context, info convmeta.Meta, req *dto.OpenAIResponsesRequest) (*dto.ClaudeRequest, error) {
if req == nil {
return nil, fmt.Errorf("request is nil")
}
if req.Model == "" {
return nil, fmt.Errorf("model is required")
}
if err := ValidateRequestChatUnsupportedFields(req); err != nil {
return nil, err
}
claudeRequest := &dto.ClaudeRequest{
Model: req.Model,
Temperature: req.Temperature,
TopP: req.TopP,
Stream: req.Stream,
}
if req.MaxOutputTokens != nil && *req.MaxOutputTokens > 0 {
claudeRequest.MaxTokens = kitutil.GetPointer(*req.MaxOutputTokens)
}
if claudeRequest.MaxTokens == nil || *claudeRequest.MaxTokens == 0 {
if defaultMaxTokens, configured := convmeta.OptionsOf(info).Claude.DefaultMaxTokensFor(req.Model); configured {
value := uint(defaultMaxTokens)
claudeRequest.MaxTokens = &value
}
}
functions, err := RequestFunctionDeclarations(req.Tools)
if err != nil {
return nil, err
}
if len(functions) > 0 {
claudeRequest.Tools = responsesFunctionDeclarationsToClaudeTools(functions)
}
toolChoice, err := RequestToolChoiceToChat(req.ToolChoice)
if err != nil {
return nil, err
}
if toolChoice != nil || RawJSONPresent(req.ParallelToolCalls) {
claudeRequest.ToolChoice = sharedclaude.MapOpenAIToolChoice(toolChoice, ParallelToolCalls(req.ParallelToolCalls))
}
applyResponsesReasoningToClaude(req, claudeRequest)
systemMessages := make([]dto.ClaudeMediaMessage, 0)
if RawJSONPresent(req.Instructions) {
instructions, err := JSONString(req.Instructions)
if err != nil {
return nil, fmt.Errorf("invalid instructions: %w", err)
}
if strings.TrimSpace(instructions) != "" {
systemMessages = append(systemMessages, dto.ClaudeMediaMessage{
Type: "text",
Text: kitutil.GetPointer(instructions),
})
}
}
inputItems, err := InputItems(req.Input)
if err != nil {
return nil, err
}
for _, item := range inputItems {
itemType := strings.TrimSpace(kitutil.Interface2String(item["type"]))
switch itemType {
case ResponsesInputTypeFunctionCall:
claudeRequest.Messages = appendClaudeToolUse(claudeRequest.Messages, responsesFunctionCallItemToClaudeToolUse(item, "arguments"))
case ResponsesInputTypeCustomToolCall:
claudeRequest.Messages = appendClaudeToolUse(claudeRequest.Messages, responsesFunctionCallItemToClaudeToolUse(item, "input"))
case ResponsesInputTypeFunctionCallOutput, ResponsesInputTypeCustomToolOutput:
claudeRequest.Messages = appendClaudeToolResult(claudeRequest.Messages, responsesFunctionOutputItemToClaudeToolResult(item))
default:
role := responsesClaudeRole(item)
parts, err := responsesInputContentToClaudeMediaMessages(c, item["content"])
if err != nil {
return nil, err
}
if role == "system" {
systemMessages = append(systemMessages, parts...)
continue
}
if len(parts) == 0 {
parts = []dto.ClaudeMediaMessage{
{
Type: "text",
Text: kitutil.GetPointer("..."),
},
}
}
claudeRequest.Messages = append(claudeRequest.Messages, dto.ClaudeMessage{
Role: role,
Content: parts,
})
}
}
if len(systemMessages) > 0 {
claudeRequest.System = systemMessages
}
claudeRequest.Messages = ensureClaudeMessagesStartWithUser(claudeRequest.Messages)
// Checked last so every injection path has had its chance to satisfy the
// required field.
if claudeRequest.MaxTokens == nil {
return nil, sharedclaude.ErrMissingMaxTokens
}
return claudeRequest, nil
}
func responsesFunctionDeclarationsToClaudeTools(functions []dto.FunctionRequest) []any {
tools := make([]any, 0, len(functions))
for _, function := range functions {
tools = append(tools, &dto.Tool{
Name: function.Name,
Description: function.Description,
InputSchema: responsesFunctionParametersToClaudeInputSchema(function.Parameters),
})
}
return tools
}
func responsesFunctionParametersToClaudeInputSchema(parameters any) map[string]interface{} {
if params, ok := parameters.(map[string]any); ok {
schema := make(map[string]interface{}, len(params))
for key, value := range params {
schema[key] = value
}
if schema["type"] == nil {
schema["type"] = "object"
}
if schema["properties"] == nil {
schema["properties"] = map[string]interface{}{}
}
return schema
}
return map[string]interface{}{
"type": "object",
"properties": map[string]interface{}{},
}
}
func applyResponsesReasoningToClaude(req *dto.OpenAIResponsesRequest, claudeRequest *dto.ClaudeRequest) {
effort := ReasoningEffort(req)
switch effort {
case "low":
claudeRequest.Thinking = &dto.Thinking{
Type: "enabled",
BudgetTokens: kitutil.GetPointer(1280),
}
case "medium":
claudeRequest.Thinking = &dto.Thinking{
Type: "enabled",
BudgetTokens: kitutil.GetPointer(2048),
}
case "high":
claudeRequest.Thinking = &dto.Thinking{
Type: "enabled",
BudgetTokens: kitutil.GetPointer(4096),
}
}
}
func responsesInputContentToClaudeMediaMessages(c context.Context, content any) ([]dto.ClaudeMediaMessage, error) {
contentParts, err := ContentParts(content)
if err != nil {
return nil, err
}
parts := make([]dto.ClaudeMediaMessage, 0, len(contentParts))
for _, contentPart := range contentParts {
partType := strings.TrimSpace(kitutil.Interface2String(contentPart["type"]))
switch partType {
case "input_text", "output_text", "text":
text := kitutil.Interface2String(contentPart["text"])
if text != "" {
parts = append(parts, dto.ClaudeMediaMessage{
Type: "text",
Text: kitutil.GetPointer(text),
})
}
case "input_image", "input_file", "input_audio", "input_video":
source := ContentPartToFileSource(contentPart)
if source == nil {
continue
}
base64Data, mimeType, err := relaymedia.ResolveBase64Data(c, source, "formatting Responses input for Claude")
if err != nil {
return nil, fmt.Errorf("get file data failed: %s", err.Error())
}
claudePart := dto.ClaudeMediaMessage{
Source: &dto.ClaudeMessageSource{
Type: "base64",
MediaType: mimeType,
Data: base64Data,
},
}
if strings.HasPrefix(mimeType, "application/pdf") {
claudePart.Type = "document"
} else {
claudePart.Type = "image"
}
parts = append(parts, claudePart)
}
}
return parts, nil
}
func responsesFunctionCallItemToClaudeToolUse(item map[string]any, inputKey string) dto.ClaudeMediaMessage {
return dto.ClaudeMediaMessage{
Type: "tool_use",
Id: CallID(item),
Name: strings.TrimSpace(kitutil.Interface2String(item["name"])),
Input: ObjectValue(item[inputKey], inputKey),
}
}
func responsesFunctionOutputItemToClaudeToolResult(item map[string]any) dto.ClaudeMediaMessage {
return dto.ClaudeMediaMessage{
Type: "tool_result",
ToolUseId: CallID(item),
Content: responsesToolOutputValue(item["output"]),
}
}
func responsesToolOutputValue(value any) any {
if value == nil {
return ""
}
return value
}
func appendClaudeToolUse(messages []dto.ClaudeMessage, toolUse dto.ClaudeMediaMessage) []dto.ClaudeMessage {
if len(messages) > 0 && messages[len(messages)-1].Role == "assistant" {
last := messages[len(messages)-1]
parts := claudeMessageContentParts(last.Content)
parts = append(parts, toolUse)
last.Content = parts
messages[len(messages)-1] = last
return messages
}
return append(messages, dto.ClaudeMessage{
Role: "assistant",
Content: []dto.ClaudeMediaMessage{toolUse},
})
}
func appendClaudeToolResult(messages []dto.ClaudeMessage, toolResult dto.ClaudeMediaMessage) []dto.ClaudeMessage {
if len(messages) > 0 && messages[len(messages)-1].Role == "user" {
last := messages[len(messages)-1]
parts := claudeMessageContentParts(last.Content)
parts = append(parts, toolResult)
last.Content = parts
messages[len(messages)-1] = last
return messages
}
return append(messages, dto.ClaudeMessage{
Role: "user",
Content: []dto.ClaudeMediaMessage{toolResult},
})
}
func claudeMessageContentParts(content any) []dto.ClaudeMediaMessage {
switch typed := content.(type) {
case []dto.ClaudeMediaMessage:
return typed
case string:
if typed == "" {
return nil
}
return []dto.ClaudeMediaMessage{
{
Type: "text",
Text: kitutil.GetPointer(typed),
},
}
default:
parts, _ := kitutil.Any2Type[[]dto.ClaudeMediaMessage](content)
return parts
}
}
func responsesClaudeRole(item map[string]any) string {
switch strings.TrimSpace(kitutil.Interface2String(item["role"])) {
case "assistant":
return "assistant"
case "system", "developer":
return "system"
default:
return "user"
}
}
func ensureClaudeMessagesStartWithUser(messages []dto.ClaudeMessage) []dto.ClaudeMessage {
if len(messages) == 0 || messages[0].Role == "user" {
return messages
}
return append([]dto.ClaudeMessage{
{
Role: "user",
Content: []dto.ClaudeMediaMessage{
{
Type: "text",
Text: kitutil.GetPointer("..."),
},
},
},
}, messages...)
}