mirror of
https://gitcode.com/JianFeeeee/ModelRouter.git
synced 2026-11-07 09:47:23 +00:00
711ccdeea8
生产实测逐渠道排查（2026-10-19）：
* qoder：PKCE 设备码流无 register 第二跳，落库凭据是裸 token 字符串；
qoderClaim 硬要求结构化 table 导致「缺少凭据对象」（而 account_test
走 token 路径正常 —— 自相矛盾状态）。改为非空裸 tok 就地包
{access_token=tok}；补 TestQoderClaimAcceptsBareTokenCredential。
* trae（SOLO 自装适配器）：上游对非流式请求也回 SSE（event: 前缀帧），
transform_response 不识别 → 整个 body 原样透传 → Go unmarshal 报
invalid character 'e'。补 aggregate_solo_sse 聚合。
* buddy（CodeBuddy）：上游 400 11101 只支持流式。transform_request 强制
stream=true；transform_response 补 SSE 聚合（tool_calls 按 index 拼分片）。
* provider 流式：真实 tool-call 片段之后出现 finish_reason=stop 时归一为
tool_calls（trae 实测如此发）；buddy/trae 不再序列化空 tool_calls 表。
* opencode：远端目录端点改为 /zen/v1/models（裸 /v1/models 回 HTML）；
兜底目录同步 2026-10-19 实测（space-bunny-free 410 下架等 5 项变化）。
opencodezen 适配器 v1.1.0：会话亲和（client_session / 首 user 消息指纹
分桶），保留免费层门禁四要素。
* 内核：适配器可声明 adapter.hidden = true 退出用户可见列表；被源引用的
隐藏适配器保持可见（源编辑器要回显当前值）。opencodezen 默认隐藏。
* channels UI：切渠道立即清空账号/模型区 + 响应代际校验，修复 qoder 账号
闪现在 opencode 卡片下的渲染竞态。
* channels request_end hook：挂接丢失的 isModelGone/recordDeadModel ——
真实 404（如 codearts GLM-5.3 "The model is not registered"）现在会进
deadModels，不再每轮 404 仍可选中。
(cherry picked from commit 2113bfc125)
293 lines
14 KiB
Lua
293 lines
14 KiB
Lua
local adapter = {}
|
||
|
||
adapter.name = "buddy"
|
||
adapter.version = "1.0.0"
|
||
adapter.endpoint = "/chat/completions"
|
||
adapter.headers = {}
|
||
|
||
-- Claude-behind-OpenAI upstreams (tabitoken, 扇贝, …) convert /chat/completions
|
||
-- to the Anthropic Messages API internally, so they inherit Anthropic's
|
||
-- tool id rule: ^[a-zA-Z0-9_-]{1,64}$, enforced by rejecting the WHOLE request
|
||
-- with "Invalid tool use format" / "tool_use.id: String should match pattern".
|
||
-- Plain OpenAI has no such rule, so an OpenAI-compatible model can mint an id
|
||
-- like "bash:0" (observed from moonshotai/kimi-k3). In a fan-out router that id
|
||
-- is replayed to every other source, so one such id kills every Claude slot at
|
||
-- once and an AUTO request falls through all tiers.
|
||
--
|
||
-- safe_tool_id is pure and deterministic, so a tool_calls entry and its
|
||
-- matching tool_call_id are rewritten identically within one request. A
|
||
-- rewritten id keeps an 8-hex digest of the ORIGINAL id, without which two
|
||
-- distinct ids could collapse into one ("a:b" and "a_b") and become an
|
||
-- unpaired/duplicate tool call. Already-legal ids are returned untouched, so
|
||
-- well-behaved traffic is byte-identical to before.
|
||
-- (anthropic.lua carries the same helper; Lua adapters have no shared prelude.)
|
||
local TOOL_ID_MAX = 64
|
||
|
||
local function safe_tool_id(id)
|
||
if type(id) ~= "string" or id == "" then return id end
|
||
local clean = string.gsub(id, "[^A-Za-z0-9_-]", "_")
|
||
if clean == id and #clean <= TOOL_ID_MAX then
|
||
return clean
|
||
end
|
||
local digest = string.sub(sha256_hex(id), 1, 8)
|
||
local keep = TOOL_ID_MAX - #digest - 1
|
||
if #clean > keep then clean = string.sub(clean, 1, keep) end
|
||
return clean .. "_" .. digest
|
||
end
|
||
|
||
-- 需要补 reasoning_effort 的模型 → 补哪个档位。
|
||
--
|
||
-- 唯一出处：参考 product.ts:237 声明 glm-5.1 的 reasoningEfforts 只有 ['medium']。
|
||
-- 但实测**任何非空档位都被接受**，故这里取 medium（与参考声明一致）。
|
||
local NEED_EFFORT = {
|
||
["glm-5.1"] = "medium",
|
||
}
|
||
|
||
-- OpenAI /chat/completions format (pass-through, strip provider-specific fields)
|
||
function adapter.transform_request(raw_body)
|
||
local ok, req = pcall(json.decode, raw_body)
|
||
if not ok then return raw_body end
|
||
-- The gateway's private control fields are GONE from the body (v2): the
|
||
-- neutral `thinking` field carries that intent and each adapter maps it in
|
||
-- its own protocol terms. These upstreams have no thinking switch at all,
|
||
-- so the field is simply dropped here — which is what the nil-ing below
|
||
-- used to do, except the field being dropped is no longer the kernel's
|
||
-- private one.
|
||
req.thinking = nil
|
||
req.extra_body = nil
|
||
-- ★★ buddy（腾讯 CodeBuddy）上游**只支持流式**：body.stream=false 会被
|
||
-- 400 拒绝（实测 2026-10-19）：
|
||
-- 11101: Non-stream chat request is currently not supported
|
||
-- 与 Accept: text/event-stream 头无关（头早已在发），判据是 body。
|
||
-- 故强制 stream=true，非流式请求由内核的"客户端要非流式、上游回流式"
|
||
-- 路径消化（transform_response 需能聚合 SSE，见下）。
|
||
req.stream = true
|
||
-- ★★ buddy 的 glm-5.1 **要求请求体里有非空的 reasoning_effort**，
|
||
-- 否则上游回 400：
|
||
-- {"code":11150,"msg":"the reasoning effort value is not supported
|
||
-- by the current model"}
|
||
-- 实测（2026-10-16，逐字对照实验）：
|
||
-- glm-5.1 裸（无该字段） → 400 11150 ✗
|
||
-- glm-5.1 + reasoning_effort:medium → 200 ✓
|
||
-- glm-5.1 空串 → 400 11150 ✗
|
||
-- 任何**非空**档位均可（low/medium/high/xhigh/max/minimal 逐个 200）
|
||
-- 而其余 15 个 buddy 模型裸跑全 200（多带该字段也无害）—— 即这是
|
||
-- glm-5.1 **独有**的要求。
|
||
-- ⚠️ 只对**已实测需要**的模型补（而不是一律补）：一律补会把"某个模型的
|
||
-- 特殊要求"扩成"全平台行为"，将来上游若对某些模型收紧就会一次性弄坏
|
||
-- 所有模型。
|
||
if NEED_EFFORT[req.model or ""] and
|
||
not (type(req.reasoning_effort) == "string" and req.reasoning_effort ~= "") and
|
||
req.reasoning_effort == nil then
|
||
req.reasoning_effort = NEED_EFFORT[req.model]
|
||
end
|
||
if req.messages then
|
||
for _, msg in ipairs(req.messages) do
|
||
msg.reasoning_content = nil
|
||
if msg.tool_call_id ~= nil then
|
||
msg.tool_call_id = safe_tool_id(msg.tool_call_id)
|
||
end
|
||
if type(msg.tool_calls) == "table" then
|
||
for _, tc in ipairs(msg.tool_calls) do
|
||
if type(tc) == "table" and tc.id ~= nil then
|
||
tc.id = safe_tool_id(tc.id)
|
||
end
|
||
end
|
||
end
|
||
end
|
||
end
|
||
return json.encode(req)
|
||
end
|
||
|
||
-- aggregate_sse(raw)：把上游 SSE（data: 行流）聚合成一个非流式 OpenAI 响应。
|
||
-- 强制 stream=true 后，内核的非流式路径会在这里收到完整 SSE 原文——必须聚合，
|
||
-- 否则 json.decode 失败原样透传，Go 侧 unmarshal 报 invalid character 'd'。
|
||
-- 聚合 content / reasoning_content / tool_calls（按 index 拼接分片）/
|
||
-- finish_reason / usage；无可解析帧时返回 nil（调用方回退原样）。
|
||
local function aggregate_sse(raw)
|
||
local content, reasoning = {}, {}
|
||
local finish, usage, id, model, created = nil, nil, nil, nil, nil
|
||
local tools_by_index = {}
|
||
local seen = false
|
||
for line in (raw .. "\n"):gmatch("([^\n]*)\n") do
|
||
local s = line:gsub("^%s+", ""):gsub("%s+$", "")
|
||
if s:sub(1, 5) == "data:" then
|
||
local payload = s:sub(6):gsub("^%s+", "")
|
||
if payload ~= "" and payload ~= "[DONE]" then
|
||
local ok, d = pcall(json.decode, payload)
|
||
if ok and type(d) == "table" then
|
||
seen = true
|
||
id = id or d.id
|
||
model = model or d.model
|
||
created = created or d.created
|
||
if type(d.usage) == "table" then usage = d.usage end
|
||
local ch = d.choices and d.choices[1]
|
||
if type(ch) == "table" then
|
||
if type(ch.finish_reason) == "string" and ch.finish_reason ~= "" then
|
||
finish = ch.finish_reason
|
||
end
|
||
local dl = ch.delta
|
||
if type(dl) == "table" then
|
||
if type(dl.content) == "string" then content[#content + 1] = dl.content end
|
||
if type(dl.reasoning_content) == "string" then
|
||
reasoning[#reasoning + 1] = dl.reasoning_content
|
||
end
|
||
if type(dl.tool_calls) == "table" then
|
||
for _, tc in ipairs(dl.tool_calls) do
|
||
if type(tc) == "table" and tc.index ~= nil then
|
||
local slot = tools_by_index[tc.index]
|
||
if slot == nil then
|
||
slot = { id = tc.id, type = tc.type or "function",
|
||
["function"] = { name = "", arguments = "" } }
|
||
tools_by_index[tc.index] = slot
|
||
end
|
||
if tc.id then slot.id = tc.id end
|
||
local fn = tc["function"]
|
||
if type(fn) == "table" then
|
||
if fn.name then slot["function"].name = slot["function"].name .. fn.name end
|
||
if fn.arguments then slot["function"].arguments = slot["function"].arguments .. fn.arguments end
|
||
end
|
||
end
|
||
end
|
||
end
|
||
end
|
||
end
|
||
end
|
||
end
|
||
end
|
||
end
|
||
if not seen then return nil end
|
||
local msg = { role = "assistant", content = table.concat(content) }
|
||
local rc = table.concat(reasoning)
|
||
if rc ~= "" then msg.reasoning_content = rc end
|
||
local ordered = {}
|
||
for i = 0, 63 do
|
||
if tools_by_index[i] then ordered[#ordered + 1] = tools_by_index[i] end
|
||
end
|
||
if #ordered > 0 then msg.tool_calls = ordered end
|
||
local out = {
|
||
id = id or "", object = "chat.completion", created = created or 0,
|
||
model = model or "",
|
||
choices = { { index = 0, message = msg,
|
||
finish_reason = (finish and finish ~= "") and finish or "stop" } },
|
||
}
|
||
if usage then out.usage = usage end
|
||
local ok2, enc = pcall(json.encode, out)
|
||
if not ok2 then return nil end
|
||
return enc
|
||
end
|
||
|
||
function adapter.transform_response(raw_body)
|
||
-- 上游只会回流式（transform_request 强制 stream=true）。客户端要非流式时
|
||
-- 内核把完整 SSE 原文交到这里，先聚合再走统一转换。
|
||
if type(raw_body) == "string"
|
||
and (raw_body:match("^%s*data:") or raw_body:match("[\r\n]data:")) then
|
||
local aggregated = aggregate_sse(raw_body)
|
||
if aggregated ~= nil then raw_body = aggregated end
|
||
end
|
||
-- 非流式 OpenAI 响应 → 统一信封。此前在 openai/groq/kimicode/mistral/github
|
||
-- 五个适配器里**逐字节相同**（各 2206B，diff 为空），已抽到共享前导。
|
||
return __prelude.openai_response(raw_body)
|
||
end
|
||
|
||
function adapter.transform_stream_chunk(raw_chunk)
|
||
local ok, chunk = pcall(json.decode, raw_chunk)
|
||
if not ok then return "" end
|
||
|
||
-- OpenAI-style streams may attach usage to a chunk with empty choices
|
||
-- (the final usage chunk). Keys must match Go's TokenUsage json tags
|
||
-- (prompt/completion/total); the gateway re-emits standard *_tokens.
|
||
local uses = nil
|
||
if type(chunk.usage) == "table" then
|
||
uses = {
|
||
prompt = chunk.usage.prompt_tokens or chunk.usage.prompt or 0,
|
||
completion = chunk.usage.completion_tokens or chunk.usage.completion or 0,
|
||
total = chunk.usage.total_tokens or chunk.usage.total or 0,
|
||
}
|
||
if type(chunk.usage.prompt_tokens_details) == "table" and chunk.usage.prompt_tokens_details.cached_tokens ~= nil then
|
||
uses.prompt_tokens_details = { cached_tokens = chunk.usage.prompt_tokens_details.cached_tokens }
|
||
elseif (chunk.usage.prompt_cache_hit_tokens or 0) > 0 then
|
||
uses.prompt_cache_hit_tokens = chunk.usage.prompt_cache_hit_tokens
|
||
uses.prompt_cache_miss_tokens = chunk.usage.prompt_cache_miss_tokens or 0
|
||
uses.prompt_tokens_details = { cached_tokens = chunk.usage.prompt_cache_hit_tokens }
|
||
end
|
||
end
|
||
|
||
if not chunk.choices or #chunk.choices == 0 then
|
||
if uses ~= nil then
|
||
-- usage-only chunk is not a content/finish signal; the gateway
|
||
-- emits its own terminal stop chunk and merges this usage.
|
||
return json.encode({ usage = uses, done = false })
|
||
end
|
||
return ""
|
||
end
|
||
local delta = chunk.choices[1].delta or {}
|
||
local fr = chunk.choices[1].finish_reason
|
||
|
||
local finish = (type(fr) == "string" and fr ~= "") and fr or nil
|
||
|
||
local unified = {
|
||
content = delta.content or "",
|
||
done = (finish ~= nil)
|
||
}
|
||
if finish then
|
||
unified.finish_reason = finish
|
||
end
|
||
if uses ~= nil then
|
||
unified.usage = uses
|
||
end
|
||
if delta.reasoning_content then
|
||
unified.reasoning_content = delta.reasoning_content
|
||
end
|
||
if type(delta.tool_calls) == "table" and #delta.tool_calls > 0 then
|
||
-- Do not serialize an empty Lua table as JSON `{}`. That shape is
|
||
-- neither an OpenAI tool-call array nor an omitted field, and several
|
||
-- agent clients reject it on the final chunk.
|
||
for _, tc in ipairs(delta.tool_calls) do
|
||
if type(tc) == "table" and tc.id ~= nil then
|
||
tc.id = safe_tool_id(tc.id)
|
||
end
|
||
end
|
||
unified.tool_calls = delta.tool_calls
|
||
end
|
||
return json.encode(unified)
|
||
end
|
||
|
||
-- 错误收敛：标准 OpenAI 信封 {error:{message,...}}
|
||
function adapter.transform_error(status, body)
|
||
-- OpenAI 风格错误体抽消息。此前在 deepseek/github/groq/kimicode/
|
||
-- openai/sensenova 六个适配器里逐字节相同（各 293B），已抽到共享前导。
|
||
return __prelude.openai_error(status, body)
|
||
end
|
||
|
||
|
||
-- ★ build_headers：buddy 要求客户端身份头 + 流式的 Accept。
|
||
--
|
||
-- 实测（2026-10-16）：不发 `Accept: text/event-stream` 时上游接受请求（200）
|
||
-- 但只回一帧空内容（completion_tokens=0），既不报错也不产出正文 —— 表现是
|
||
-- "模型不回答"，极易误判成模型坏了或额度不足。
|
||
-- 参考 buddy-adapter.ts 的 send() 固定发该头。
|
||
--
|
||
-- 身份头照参考 credentialRequestHeaders + CODEBUDDY 产品配置：
|
||
-- X-Domain 用产品域（不用凭据里的 domain 快照，那是登录时的区域，
|
||
-- 跨产品迁移后会指错区）；X-Product 是**部署类型**（SaaS）；X-Product-Code
|
||
-- 是产品码（codebuddy）。
|
||
function adapter.build_headers(meta)
|
||
local h = {
|
||
["Accept"] = "text/event-stream",
|
||
["Content-Type"] = "application/json",
|
||
["X-Domain"] = "copilot.tencent.com",
|
||
["X-Product"] = "SaaS",
|
||
["X-Product-Code"] = "codebuddy",
|
||
["User-Agent"] = "CodeBuddyIDE/1.106.1",
|
||
}
|
||
-- Authorization 由内核在 buildHeaders 之后补（值为源的 api_key）——
|
||
-- 这里显式给出同一份，保证与其它 buddy 请求形态一致。
|
||
if type(meta) == "table" and type(meta.api_key) == "string" and meta.api_key ~= "" then
|
||
h["Authorization"] = "Bearer " .. meta.api_key
|
||
end
|
||
return h
|
||
end
|
||
|
||
return adapter
|