dsh-tap 0.20.0

package/scripts/verify-providers.mjs

@@ -0,0 +1,255 @@
+#!/usr/bin/env node
+/**
+ * Offline regression for providers/openai-compat.js (G6/G7 shared skeleton):
+ *
+ *   1. primary path: GET /models works → real catalog returned
+ *   2. /models 401 (bad key) → throws, NOT tagged MODELS_ENDPOINT_404
+ *   3. /models 404 without fallbackModels → error propagates (strict custom)
+ *   4. /models 404 + fallbackModels + good key → chat probe ok → fallback list
+ *   5. /models 404 + fallbackModels + bad key (std 401 invalid_api_key) → throws
+ *   6. iFlow dialect: 200 + {"status":"434"} → auth fail; 200 + choices → ok
+ *   7. providerBlock shape (apiKeyEnv convention)
+ *   8. shipped presets: baseURL 钉选；qwen 兜底、openrouter staticCatalog
+ *   9. staticCatalog 形态：公开 /models 不调不验，chat 探针验 key，吃内置清单
+ *   10. refresh 重建接缝（0.19.0 修）：重建 adapter 重铺块后档位声明仍在
+ *       （mock 上游走 refresh 实际序列）+ index.js test/refresh 调用点源级断言
+ *
+ * Usage: node scripts/verify-providers.mjs   (no network, no credentials)
+ */
+
+import { createServer } from 'node:http'
+import { readFileSync } from 'node:fs'
+
+import { PROVIDER_ID_RE, keyRefFor, fetchOpenAIModels, probeChatKey, providerBlock, createOpenAICompatProvider, rebuildAdapterForEntry } from '../providers/openai-compat.js'
+import ark from '../providers/ark/index.js'
+import bailian from '../providers/bailian/index.js'
+import deepseek from '../providers/deepseek/index.js'
+import bigmodel from '../providers/bigmodel/index.js'
+import moonshot from '../providers/moonshot/index.js'
+import openrouter from '../providers/openrouter/index.js'
+import qwen from '../providers/qwen/index.js'
+
+let failures = 0
+function check(label, cond, detail = '') {
+  if (cond) console.log(`  ok   ${label}`)
+  else { failures++; console.log(`  FAIL ${label}${detail ? ` — ${detail}` : ''}`) }
+}
+
+/** Spin up a mock upstream; routes: { models: [status, body], chat: (req, body) => [status, body] }. */
+function mockUpstream(routes) {
+  const server = createServer((req, res) => {
+    let raw = ''
+    req.on('data', (c) => { raw += c })
+    req.on('end', () => {
+      const hit = req.url.endsWith('/models') ? routes.models
+        : req.url.endsWith('/chat/completions') ? (typeof routes.chat === 'function' ? routes.chat(req, raw) : routes.chat)
+        : null
+      if (!hit) { res.writeHead(404); res.end('nf'); return }
+      res.writeHead(hit[0], { 'content-type': 'application/json' })
+      res.end(typeof hit[1] === 'string' ? hit[1] : JSON.stringify(hit[1]))
+    })
+  })
+  return new Promise((resolve) => server.listen(0, '127.0.0.1', () => {
+    resolve({ server, base: `http://127.0.0.1:${server.address().port}/v1` })
+  }))
+}
+
+const GOOD = 'sk-good-key'
+
+// 1–3: /models behavior variants
+{
+  const { server, base } = await mockUpstream({
+    models: [200, { object: 'list', data: [{ id: 'm-a' }, { id: 'm-b', name: 'B' }] }],
+    chat: [200, { choices: [] }],
+  })
+  const models = await fetchOpenAIModels(base, GOOD)
+  check('1 GET /models 主路径返回真实目录', models.length === 2 && models[0].id === 'm-a' && models[1].id === 'm-b', JSON.stringify(models))
+  server.close()
+}
+{
+  const { server, base } = await mockUpstream({ models: [401, { error: { message: 'bad key' } }], chat: [200, {}] })
+  const err = await fetchOpenAIModels(base, 'sk-bad').then(() => null, (e) => e)
+  check('2 /models 401 抛错且不带 404 标记', !!err && err.code !== 'MODELS_ENDPOINT_404' && /401/.test(err.message), String(err))
+  server.close()
+}
+{
+  const { server, base } = await mockUpstream({ models: [404, 'nf'], chat: [200, { choices: [] }] })
+  const strict = createOpenAICompatProvider({ id: 'strict', displayName: 'Strict', baseURL: base })
+  const err = await strict.fetchModels(GOOD).then(() => null, (e) => e)
+  check('3 无 fallbackModels 时 /models 404 原样抛出（custom 严格）', err?.code === 'MODELS_ENDPOINT_404', String(err))
+  server.close()
+}
+
+// 4–5: fallback path, standard auth dialect
+{
+  const { server, base } = await mockUpstream({
+    models: [404, 'nf'],
+    chat: (req) => (req.headers.authorization === `Bearer ${GOOD}`
+      ? [200, { choices: [{ message: { content: 'p' } }], usage: { total_tokens: 1 } }]
+      : [401, { error: { code: 'invalid_api_key', message: 'invalid access token or token expired' } }]),
+  })
+  const adapter = createOpenAICompatProvider({ id: 'fb', displayName: 'FB', baseURL: base, fallbackModels: ['fb-pro', 'fb-lite'] })
+  const models = await adapter.fetchModels(GOOD)
+  check('4 /models 404 + 好 key：探针通过，吃兜底清单', models.length === 2 && models[0].id === 'fb-pro', JSON.stringify(models))
+  const err = await adapter.fetchModels('sk-bad').then(() => null, (e) => e)
+  check('5 /models 404 + 坏 key（401 invalid_api_key）：拒绝', !!err && /认证拒绝|401/.test(err.message), String(err))
+  server.close()
+}
+
+// 6: iFlow dialect (HTTP 200 + status 434)
+{
+  const { server, base } = await mockUpstream({
+    models: [404, 'nf'],
+    chat: (req) => (req.headers.authorization === `Bearer ${GOOD}`
+      ? [200, { status: '200', body: { choices: [{ message: { content: 'p' } }] } }]
+      : [200, { status: '434', msg: 'Invalid apiKey，get your apiKey', body: null }]),
+  })
+  const adapter = createOpenAICompatProvider({ id: 'iflowish', displayName: 'IF', baseURL: base, fallbackModels: ['if-a'] })
+  const err = await adapter.fetchModels('sk-bad').then(() => null, (e) => e)
+  check('6a iFlow 方言坏 key（200+status 434）：拒绝', !!err && /认证拒绝/.test(err.message), String(err))
+  const models = await adapter.fetchModels(GOOD)
+  check('6b iFlow 方言好 key（200+choices）：通过', models.length === 1 && models[0].id === 'if-a', JSON.stringify(models))
+  // probeChatKey 直连形态也锁一下（网络错误之外的判定都在这层）。
+  await probeChatKey(base, GOOD, 'if-a')
+  check('6c probeChatKey 好 key 不抛', true)
+  server.close()
+}
+
+// 7: providerBlock shape
+{
+  const block = providerBlock({ id: 'mock-up', displayName: 'Mock', baseURL: 'http://x/v1' }, [{ id: 'm1' }, { id: 'm2', name: 'N' }])
+  check('7 providerBlock：apiKeyEnv 惯例 + 模型形状',
+    block.api === 'openai-completions' && block.apiKeyEnv === 'MOCK_UP_API_KEY'
+      && block.models.length === 2 && block.models[1].name === 'N' && !('name' in block.models[0]),
+    JSON.stringify(block))
+  check('7b keyRefFor / PROVIDER_ID_RE 不变', keyRefFor('ark') === 'ARK_API_KEY' && PROVIDER_ID_RE.test('a-1') && !PROVIDER_ID_RE.test('Bad'))
+  // 7c 思考档位声明透传（0.19.0）：preset.modelEfforts 按模型 id 合并进条目，
+  // 未声明的模型保持原形状；reasoningCompat 才写 compat（缺它 pi-ai 出站不写
+  // reasoning_effort）。refresh 走同一函数 ⇒ 声明不丢（下面 7e 用缩减清单复验）。
+  const preset7 = {
+    id: 'mock-up', displayName: 'Mock', baseURL: 'http://x/v1',
+    modelEfforts: { m1: { low: 'low', high: 'high' } },
+    reasoningCompat: true,
+  }
+  const block7c = providerBlock(preset7, [{ id: 'm1' }, { id: 'm2', name: 'N' }])
+  check('7c providerBlock：声明模型带 reasoningEfforts、未声明的不带',
+    JSON.stringify(block7c.models[0].reasoningEfforts) === JSON.stringify({ low: 'low', high: 'high' })
+      && block7c.models[1].reasoningEfforts === undefined,
+    JSON.stringify(block7c.models))
+  check('7d providerBlock：reasoningCompat → compat.supportsReasoningEffort',
+    JSON.stringify(block7c.compat) === JSON.stringify({ thinkingFormat: 'openai', supportsReasoningEffort: true })
+      && providerBlock({ id: 'mock-up', displayName: 'Mock', baseURL: 'http://x/v1' }, [{ id: 'm1' }]).compat === undefined,
+    JSON.stringify(block7c.compat))
+  const block7e = providerBlock(preset7, [{ id: 'm1' }])
+  check('7e 刷新（缩减/重建清单）后声明按 id 保留',
+    JSON.stringify(block7e.models[0].reasoningEfforts) === JSON.stringify({ low: 'low', high: 'high' })
+      && block7e.models.length === 1,
+    JSON.stringify(block7e.models))
+}
+
+// 8: shipped presets
+{
+  check('8a ark/bailian/deepseek/bigmodel/moonshot 无兜底（有真 /models）',
+    [ark, bailian, deepseek, bigmodel, moonshot].every((p) => p.fallbackModels.length === 0 && !p.staticCatalog))
+  check('8b qwen 带兜底清单且首项可作探针模型',
+    qwen.fallbackModels.length >= 1 && typeof qwen.fallbackModels[0] === 'string')
+  check('8c preset baseURL 不被意外改动',
+    ark.baseURL === 'https://ark.cn-beijing.volces.com/api/v3'
+      && bailian.baseURL === 'https://dashscope.aliyuncs.com/compatible-mode/v1'
+      && deepseek.baseURL === 'https://api.deepseek.com/v1'
+      && bigmodel.baseURL === 'https://open.bigmodel.cn/api/paas/v4'
+      && moonshot.baseURL === 'https://api.moonshot.cn/v1'
+      && openrouter.baseURL === 'https://openrouter.ai/api/v1'
+      && qwen.baseURL === 'https://portal.qwen.ai/v1')
+  check('8d openrouter 静态目录：staticCatalog + 内置精选清单（id 含 / 且去重）',
+    openrouter.staticCatalog === true && openrouter.fallbackModels.length >= 5
+      && openrouter.fallbackModels.every((m) => m.includes('/'))
+      && new Set(openrouter.fallbackModels).size === openrouter.fallbackModels.length)
+}
+
+// 9: staticCatalog（公开目录型上游，OpenRouter 形态）：/models 对任意 key 都 200，
+//    目录不能验 key —— 必须走 chat 探针，且清单恒为内置表（不吃 /models 全量）。
+{
+  let modelsHits = 0
+  const server = createServer((req, res) => {
+    if (req.url.endsWith('/models')) {
+      modelsHits++
+      res.writeHead(200, { 'content-type': 'application/json' })
+      res.end(JSON.stringify({ object: 'list', data: [{ id: 'pub-a' }, { id: 'pub-b' }] }))
+      return
+    }
+    let raw = ''
+    req.on('data', (c) => { raw += c })
+    req.on('end', () => {
+      const ok = req.headers.authorization === `Bearer ${GOOD}`
+      res.writeHead(ok ? 200 : 401, { 'content-type': 'application/json' })
+      res.end(ok ? JSON.stringify({ choices: [] }) : JSON.stringify({ error: { message: 'User not found.', code: 401 } }))
+    })
+  })
+  await new Promise((resolve) => server.listen(0, '127.0.0.1', resolve))
+  const base = `http://127.0.0.1:${server.address().port}/v1`
+  const adapter = createOpenAICompatProvider({ id: 'pubcat', displayName: 'Pub', baseURL: base, staticCatalog: true, fallbackModels: ['curated-a', 'curated-b'] })
+  const models = await adapter.fetchModels(GOOD)
+  check('9a staticCatalog 好 key：清单 = 内置表（不吃公开 /models）',
+    models.length === 2 && models[0].id === 'curated-a' && models[1].id === 'curated-b', JSON.stringify(models))
+  check('9b staticCatalog 不调用 /models', modelsHits === 0, `modelsHits=${modelsHits}`)
+  const err = await adapter.fetchModels('sk-bad').then(() => null, (e) => e)
+  check('9c staticCatalog 坏 key（chat 401）：拒绝', !!err && /认证拒绝|401/.test(err.message), String(err))
+  server.close()
+}
+
+// 10: refresh 重建接缝（0.19.0 修）：refreshExtraProviderModels 用重建的 adapter
+//     重铺 provider 块，preset 的档位声明（modelEfforts/reasoningCompat）必须
+//     随行。refresh 在组合根内且 PROVIDER_PRESETS 模块私有——shipped preset 也
+//     无一实际声明档位（踩坑 #42 纪律不臆造）⇒ 全链路 mock 造不出带声明的条目，
+//     故在此按 refresh 的实际序列（entry+preset 重建 adapter → fetchModels →
+//     modelBlock）打 mock 上游做功能断言；index.js 调用点用源级断言锁住
+//     （verify-desktop-acceptance §4a 同口径）。
+{
+  const preset10 = {
+    id: 'mock-up', displayName: 'Mock', baseURL: 'http://unused.example/v1',
+    modelEfforts: { 'm-a': { low: 'low', high: 'high' } },
+    reasoningCompat: true,
+  }
+  const { server, base } = await mockUpstream({
+    models: [200, { object: 'list', data: [{ id: 'm-a' }, { id: 'm-b' }] }],
+    chat: [200, { choices: [] }],
+  })
+  // refresh 的序列：registry entry（添加时快照打底）+ preset 先验 → 重建
+  // adapter → 实拉目录 → 重铺 provider 块。baseURL 由 entry 供给（merged 对象），
+  // preset10 的 baseURL 不参与（与线上 refresh 同构）。
+  const entry10 = { id: 'mock-up', displayName: 'Mock', baseURL: base, preset: 'mock-up', keyRef: 'MOCK_UP_API_KEY', addedAt: 1 }
+  const adapter10 = rebuildAdapterForEntry(entry10, preset10)
+  const models10 = await adapter10.fetchModels(GOOD)
+  const block10 = adapter10.modelBlock(models10)
+  check('10a refresh 序列（重建 adapter → fetchModels → 重铺块）后声明仍在',
+    models10.length === 2
+      && JSON.stringify(block10.models[0].reasoningEfforts) === JSON.stringify({ low: 'low', high: 'high' })
+      && block10.models[1].reasoningEfforts === undefined
+      && JSON.stringify(block10.compat) === JSON.stringify({ thinkingFormat: 'openai', supportsReasoningEffort: true }),
+    JSON.stringify(block10))
+  server.close()
+  const block10b = rebuildAdapterForEntry({ ...entry10, baseURL: 'http://x/v1' }, null).modelBlock([{ id: 'm-a' }])
+  check('10b 重建接缝：custom 条目（preset 未命中）不投毒声明',
+    block10b.models[0].reasoningEfforts === undefined && block10b.compat === undefined,
+    JSON.stringify(block10b))
+
+  const tapSrc = readFileSync(new URL('../index.js', import.meta.url), 'utf8')
+  const fnSrc = (name) => {
+    const at = tapSrc.indexOf(`async function ${name}`)
+    return at < 0 ? '' : tapSrc.slice(at, at + tapSrc.slice(at).indexOf('\n}'))
+  }
+  const refreshSrc = fnSrc('refreshExtraProviderModels')
+  const callLine = 'const adapter = rebuildAdapterForEntry(entry, presetHit)'
+  check('10c index.js refresh 调用点：走共用接缝并重铺 provider 块（防漏传回归）',
+    refreshSrc.includes(callLine)
+      && refreshSrc.includes('writeProviderBlock(id, adapter.modelBlock(models))')
+      && !refreshSrc.includes('createOpenAICompatProvider({ ...entry'),
+    'refreshExtraProviderModels 未走接缝或未重铺块')
+  check('10d index.js test 调用点：与 refresh 同参（接缝唯一出处）',
+    fnSrc('testExtraProvider').includes(callLine),
+    'testExtraProvider 未走接缝')
+}
+
+console.log(failures ? `\n${failures} FAIL` : '\nall green')
+process.exit(failures ? 1 : 0)