Files
new-api/relay/channel/claude/relay-claude.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

289 lines
11 KiB
Go

package claude
import (
"io"
"net/http"
"strings"
"github.com/QuantumNous/new-api/common"
"github.com/QuantumNous/new-api/constant"
"github.com/QuantumNous/new-api/logger"
relaycommon "github.com/QuantumNous/new-api/relay/common"
"github.com/QuantumNous/new-api/relay/helper"
"github.com/QuantumNous/new-api/relaykit/dto"
"github.com/QuantumNous/new-api/relaykit/relayconvert"
"github.com/QuantumNous/new-api/relaykit/types"
"github.com/QuantumNous/new-api/service"
"github.com/QuantumNous/new-api/setting/model_setting"
"github.com/gin-gonic/gin"
)
func stopReasonClaude2OpenAI(reason string) string {
return relayconvert.StopReasonClaudeToOpenAI(reason)
}
func maybeMarkClaudeRefusal(c *gin.Context, stopReason string) {
if c == nil {
return
}
if strings.EqualFold(stopReason, "refusal") {
common.SetContextKey(c, constant.ContextKeyAdminRejectReason, "claude_stop_reason=refusal")
}
}
func StreamResponseClaude2OpenAI(claudeResponse *dto.ClaudeResponse) *dto.ChatCompletionsStreamResponse {
return relayconvert.StreamResponseClaude2OpenAI(claudeResponse)
}
func ResponseClaude2OpenAI(claudeResponse *dto.ClaudeResponse) *dto.OpenAITextResponse {
return relayconvert.ResponseClaude2OpenAI(claudeResponse)
}
type ClaudeResponseInfo = relayconvert.ClaudeResponseInfo
func cacheCreationTokensForOpenAIUsage(usage *dto.Usage) int {
if usage == nil {
return 0
}
openAIUsage := relayconvert.UsageFromClaudeUsage(usage)
if openAIUsage == nil {
return 0
}
return openAIUsage.PromptTokens - usage.PromptTokens - usage.PromptTokensDetails.CachedTokens
}
func buildOpenAIStyleUsageFromClaudeUsage(usage *dto.Usage) dto.Usage {
mapped := relayconvert.UsageFromClaudeUsage(usage)
if mapped == nil {
return dto.Usage{}
}
return *mapped
}
func buildMessageDeltaPatchUsage(claudeResponse *dto.ClaudeResponse, claudeInfo *ClaudeResponseInfo) *dto.ClaudeUsage {
return relayconvert.BuildMessageDeltaPatchUsage(claudeResponse, claudeInfo)
}
func shouldSkipClaudeMessageDeltaUsagePatch(info *relaycommon.RelayInfo) bool {
if model_setting.GetGlobalSettings().PassThroughRequestEnabled {
return true
}
if info == nil {
return false
}
return info.ChannelSetting.PassThroughBodyEnabled
}
func patchClaudeMessageDeltaUsageData(data string, usage *dto.ClaudeUsage) string {
return relayconvert.PatchClaudeMessageDeltaUsageData(data, usage)
}
func FormatClaudeResponseInfo(claudeResponse *dto.ClaudeResponse, oaiResponse *dto.ChatCompletionsStreamResponse, claudeInfo *ClaudeResponseInfo) bool {
return relayconvert.FormatClaudeResponseInfo(claudeResponse, oaiResponse, claudeInfo)
}
func HandleStreamResponseData(c *gin.Context, info *relaycommon.RelayInfo, claudeInfo *ClaudeResponseInfo, data string) *types.NewAPIError {
var claudeResponse dto.ClaudeResponse
err := common.UnmarshalJsonStr(data, &claudeResponse)
if err != nil {
common.SysLog("error unmarshalling stream response: " + err.Error())
return types.NewError(err, types.ErrorCodeBadResponseBody)
}
if claudeError := claudeResponse.GetClaudeError(); claudeError != nil && claudeError.Type != "" {
return types.WithClaudeError(*claudeError, http.StatusInternalServerError)
}
if claudeResponse.StopReason != "" {
maybeMarkClaudeRefusal(c, claudeResponse.StopReason)
}
if claudeResponse.Delta != nil && claudeResponse.Delta.StopReason != nil {
maybeMarkClaudeRefusal(c, *claudeResponse.Delta.StopReason)
}
if info.RelayFormat == types.RelayFormatClaude {
FormatClaudeResponseInfo(&claudeResponse, nil, claudeInfo)
if claudeResponse.Type == "message_start" {
// message_start, 获取usage
if claudeResponse.Message != nil {
info.UpstreamModelName = claudeResponse.Message.Model
}
} else if claudeResponse.Type == "message_delta" {
// 确保 message_delta 的 usage 包含完整的 input_tokens 和 cache 相关字段
// 解决 AWS Bedrock 等上游返回的 message_delta 缺少这些字段的问题
if !shouldSkipClaudeMessageDeltaUsagePatch(info) {
data = patchClaudeMessageDeltaUsageData(data, buildMessageDeltaPatchUsage(&claudeResponse, claudeInfo))
}
}
countClaudeStreamBillableTools(c, info, &claudeResponse)
helper.ClaudeChunkData(c, claudeResponse, data)
} else if info.RelayFormat == types.RelayFormatOpenAI {
response := StreamResponseClaude2OpenAI(&claudeResponse)
if !FormatClaudeResponseInfo(&claudeResponse, response, claudeInfo) {
return nil
}
countClaudeStreamBillableTools(c, info, &claudeResponse)
err = helper.ObjectData(c, response)
if err != nil {
logger.LogError(c, "send_stream_response_failed: "+err.Error())
}
}
return nil
}
func countClaudeStreamBillableTools(c *gin.Context, info *relaycommon.RelayInfo, claudeResponse *dto.ClaudeResponse) {
if claudeResponse == nil {
return
}
if claudeResponse.Type == "content_block_start" &&
claudeResponse.ContentBlock != nil &&
claudeResponse.ContentBlock.Type == "tool_use" {
info.CountBillableToolCall(dto.BuildInCallToolUse, claudeResponse.ContentBlock.Name)
}
if claudeResponse.Type == "message_delta" &&
claudeResponse.Usage != nil &&
claudeResponse.Usage.ServerToolUse != nil &&
claudeResponse.Usage.ServerToolUse.WebSearchRequests > 0 {
c.Set("claude_web_search_requests", claudeResponse.Usage.ServerToolUse.WebSearchRequests)
}
}
func HandleStreamFinalResponse(c *gin.Context, info *relaycommon.RelayInfo, claudeInfo *ClaudeResponseInfo) {
if claudeInfo.Usage.PromptTokens == 0 {
//上游出错
}
if claudeInfo.Usage.CompletionTokens == 0 || !claudeInfo.Done {
if common.DebugEnabled {
common.SysLog("claude response usage is not complete, maybe upstream error")
}
// 只补缺失字段,不整份覆盖——保留 message_start 已拿到的 cache 字段
fallback := service.ResponseText2Usage(c, claudeInfo.ResponseText.String(), info.UpstreamModelName, info.GetEstimatePromptTokens())
if claudeInfo.Usage.CompletionTokens == 0 ||
(!claudeInfo.Done && fallback.CompletionTokens > claudeInfo.Usage.CompletionTokens) {
claudeInfo.Usage.CompletionTokens = fallback.CompletionTokens
}
if claudeInfo.Usage.PromptTokens == 0 {
claudeInfo.Usage.PromptTokens = fallback.PromptTokens
}
claudeInfo.Usage.TotalTokens = claudeInfo.Usage.PromptTokens + claudeInfo.Usage.CompletionTokens
}
if claudeInfo.Usage != nil {
claudeInfo.Usage.UsageSemantic = "anthropic"
}
if claudeInfo.Usage != nil && claudeInfo.Usage.BillingUsage == nil {
claudeInfo.Usage.BillingUsage = dto.NewClaudeMessagesBillingUsage(buildMessageDeltaPatchUsage(nil, claudeInfo))
}
if info.RelayFormat == types.RelayFormatClaude {
//
} else if info.RelayFormat == types.RelayFormatOpenAI {
if info.ShouldIncludeUsage {
openAIUsage := buildOpenAIStyleUsageFromClaudeUsage(claudeInfo.Usage)
response := helper.GenerateFinalUsageResponse(claudeInfo.ResponseId, claudeInfo.Created, info.UpstreamModelName, openAIUsage)
err := helper.ObjectData(c, response)
if err != nil {
common.SysLog("send final response failed: " + err.Error())
}
}
helper.Done(c)
}
}
func ClaudeStreamHandler(c *gin.Context, resp *http.Response, info *relaycommon.RelayInfo) (*dto.Usage, *types.NewAPIError) {
claudeInfo := &ClaudeResponseInfo{
ResponseId: helper.GetResponseID(c),
Created: common.GetTimestamp(),
Model: info.UpstreamModelName,
ResponseText: strings.Builder{},
Usage: &dto.Usage{},
}
var err *types.NewAPIError
helper.StreamScannerHandler(c, resp, info, func(data string, sr *helper.StreamResult) {
err = HandleStreamResponseData(c, info, claudeInfo, data)
if err != nil {
sr.Stop(err)
}
})
if err != nil {
return nil, err
}
HandleStreamFinalResponse(c, info, claudeInfo)
return claudeInfo.Usage, nil
}
func HandleClaudeResponseData(c *gin.Context, info *relaycommon.RelayInfo, claudeInfo *ClaudeResponseInfo, httpResp *http.Response, data []byte) *types.NewAPIError {
var claudeResponse dto.ClaudeResponse
err := common.Unmarshal(data, &claudeResponse)
if err != nil {
return types.NewError(err, types.ErrorCodeBadResponseBody)
}
if claudeError := claudeResponse.GetClaudeError(); claudeError != nil && claudeError.Type != "" {
return types.WithClaudeError(*claudeError, http.StatusInternalServerError)
}
maybeMarkClaudeRefusal(c, claudeResponse.StopReason)
if claudeInfo.Usage == nil {
claudeInfo.Usage = &dto.Usage{}
}
if claudeResponse.Usage != nil {
claudeInfo.Usage.PromptTokens = claudeResponse.Usage.InputTokens
claudeInfo.Usage.CompletionTokens = claudeResponse.Usage.OutputTokens
claudeInfo.Usage.TotalTokens = claudeResponse.Usage.InputTokens + claudeResponse.Usage.OutputTokens
claudeInfo.Usage.UsageSemantic = "anthropic"
claudeInfo.Usage.BillingUsage = dto.NewClaudeMessagesBillingUsage(claudeResponse.Usage)
claudeInfo.Usage.PromptTokensDetails.CachedTokens = claudeResponse.Usage.CacheReadInputTokens
claudeInfo.Usage.PromptTokensDetails.CachedCreationTokens = claudeResponse.Usage.CacheCreationInputTokens
claudeInfo.Usage.ClaudeCacheCreation5mTokens = claudeResponse.Usage.GetCacheCreation5mTokens()
claudeInfo.Usage.ClaudeCacheCreation1hTokens = claudeResponse.Usage.GetCacheCreation1hTokens()
}
var responseData []byte
switch info.RelayFormat {
case types.RelayFormatOpenAI:
openaiResponse := ResponseClaude2OpenAI(&claudeResponse)
openaiResponse.Usage = buildOpenAIStyleUsageFromClaudeUsage(claudeInfo.Usage)
responseData, err = common.Marshal(openaiResponse)
if err != nil {
return types.NewError(err, types.ErrorCodeBadResponseBody)
}
case types.RelayFormatClaude:
responseData = data
}
if claudeResponse.Usage != nil && claudeResponse.Usage.ServerToolUse != nil && claudeResponse.Usage.ServerToolUse.WebSearchRequests > 0 {
c.Set("claude_web_search_requests", claudeResponse.Usage.ServerToolUse.WebSearchRequests)
}
for _, block := range claudeResponse.Content {
if block.Type == "tool_use" {
info.CountBillableToolCall(dto.BuildInCallToolUse, block.Name)
}
}
service.IOCopyBytesGracefully(c, httpResp, responseData)
return nil
}
func ClaudeHandler(c *gin.Context, resp *http.Response, info *relaycommon.RelayInfo) (*dto.Usage, *types.NewAPIError) {
defer service.CloseResponseBodyGracefully(resp)
claudeInfo := &ClaudeResponseInfo{
ResponseId: helper.GetResponseID(c),
Created: common.GetTimestamp(),
Model: info.UpstreamModelName,
ResponseText: strings.Builder{},
Usage: &dto.Usage{},
}
responseBody, err := io.ReadAll(resp.Body)
if err != nil {
return nil, types.NewError(err, types.ErrorCodeBadResponseBody)
}
logger.LogDebug(c, "responseBody: %s", responseBody)
handleErr := HandleClaudeResponseData(c, info, claudeInfo, resp, responseBody)
if handleErr != nil {
return nil, handleErr
}
return claudeInfo.Usage, nil
}