free-coding-models 0.5.34 → 0.5.36

package/src/tui/key-handler.js

@@ -10,22 +10,11 @@
  *
  *   📖 Key I opens the changelog overlay.
  *
- *   It also owns the "test key" model selection used by the Settings overlay.
- *   Anonymous telemetry hooks for model launches and a few high-signal settings
- *   actions live here too, because this module already sees the final effective
- *   tool mode, provider, and selected model right before the app hands control
- *   to an external CLI.
- *   Some providers expose models in `/v1/models` that are not actually callable
- *   on the chat-completions endpoint. To avoid false negatives when a user
- *   presses `T` in Settings, the helpers below discover candidate model IDs,
- *   merge them with repo defaults, then probe several until one is accepted.
+ *   It also owns the "test key" model selection used by the Settings overlay,
+ *   delegated to the shared `provider-key-tester.js` module for consistency
+ *   with the Web Dashboard's `/api/key/:provider/test` endpoint.
  *
  *   → Functions:
- *   - `buildProviderModelsUrl` — derive the matching `/models` endpoint when available
- *   - `parseProviderModelIds` — extract model ids from an OpenAI-style `/models` payload
- *   - `listProviderTestModels` — build an ordered candidate list for provider key verification
- *   - `classifyProviderTestOutcome` — convert attempted HTTP codes into a settings badge state
- *   - `buildProviderTestDetail` — turn probe attempts into a readable failure explanation
  *   - `createKeyHandler` — returns the async keypress handler
  *
  * @exports { buildProviderModelsUrl, parseProviderModelIds, listProviderTestModels, classifyProviderTestOutcome, buildProviderTestDetail, createKeyHandler }
@@ -39,7 +28,6 @@ import { join, dirname } from 'node:path'
 import { fileURLToPath } from 'node:url'
 import { spawn } from 'node:child_process'
 import { cleanupLegacyProxyArtifacts } from '../core/legacy-proxy-cleanup.js'
-import { sleep } from '../core/shared-helpers.js'
 import { getLastLayout, COLUMN_SORT_MAP } from './render-table.js'
 import { cycleThemeSetting, detectActiveTheme } from './theme.js'
 import { syncShellEnv, ensureShellRcSource, removeShellEnv } from '../core/shell-env.js'
@@ -64,20 +52,15 @@ import {
 } from '../core/playground.js'
 import { benchmarkModel } from '../core/benchmark.js'
 import { isPackageDevMode } from '../core/updater.js'
-
-// 📖 Some providers need an explicit probe model because the first catalog entry
-// 📖 is not guaranteed to be accepted by their chat endpoint.
-const PROVIDER_TEST_MODEL_OVERRIDES = {
-  sambanova: ['MiniMax-M2.5', 'DeepSeek-V3.1', 'DeepSeek-V3.2'],
-  nvidia: ['deepseek-ai/deepseek-v4-flash', 'openai/gpt-oss-120b'],
-  'github-models': ['openai/gpt-4.1-mini'],
-  mistral: ['mistral-small-latest', 'devstral-small-latest'],
-}
-
-// 📖 Settings key tests retry retryable failures across several models so a
-// 📖 single stale catalog entry or transient timeout does not mark a valid key as dead.
-const SETTINGS_TEST_MAX_ATTEMPTS = 10
-const SETTINGS_TEST_RETRY_DELAY_MS = 4000
+import {
+  runProviderKeyTest,
+  testProviderKeyDirect,
+  buildProviderModelsUrl,
+  parseProviderModelIds,
+  listProviderTestModels,
+  classifyProviderTestOutcome,
+  buildProviderTestDetail,
+} from '../core/provider-key-tester.js'
 
 // 📖 spawnDaemonCommand — spawns `--daemon-bg` or `--daemon-stop` and captures
 // 📖 the JSON result from stdout so we can surface startup errors to the user.
@@ -125,185 +108,6 @@ function spawnDaemonCommand(state, args) {
   })
 }
 
-// 📖 PROVIDER_AUTH_ENDPOINTS maps provider keys to their auth-check URL + method.
-// 📖 For most providers this is the /models endpoint (returns 200=valid, 401=invalid).
-// 📖 Providers without an auth-check endpoint use null (falls back to chat completion ping).
-// 📖 Special cases:
-// 📖   - replicate: uses /v1/predictions (not /models) but needs a different payload
-// 📖   - cloudflare: no auth endpoint — only has chat completions, always uses ping fallback
-const PROVIDER_AUTH_ENDPOINTS = {
-  nvidia:       { url: 'https://api.nvidia.com/v1/account',           method: 'GET' },
-  groq:         { url: 'https://api.groq.com/v1/models',             method: 'GET' },
-  cerebras:     { url: 'https://api.cerebras.ai/v1/models',          method: 'GET' },
-  sambanova:    { url: 'https://api.sambanova.ai/v1/models',         method: 'GET' },
-  openrouter:   { url: 'https://openrouter.ai/api/v1/key',             method: 'GET' },
-  mistral:      { url: 'https://api.mistral.ai/v1/models',           method: 'GET' },
-  huggingface:  { url: 'https://router.huggingface.co/v1/models',    method: 'GET' },
-  deepinfra:    { url: 'https://api.deepinfra.com/v1/models',        method: 'GET' },
-  fireworks:   { url: 'https://api.fireworks.ai/v1/models',         method: 'GET' },
-  hyperbolic:   { url: 'https://api.hyperbolic.xyz/v1/models',       method: 'GET' },
-  scaleway:     { url: 'https://api.scaleway.ai/v1/models',          method: 'GET' },
-  siliconflow:  { url: 'https://api.siliconflow.com/v1/models',     method: 'GET' },
-  together:     { url: 'https://api.together.xyz/v1/models',        method: 'GET' },
-  perplexity:   { url: 'https://api.perplexity.ai/v1/models',       method: 'GET' },
-  chutes:       { url: 'https://chutes.ai/v1/models',               method: 'GET' },
-  ovhcloud:     { url: 'https://oai.endpoints.kepler.ai.cloud.ovh.net/v1/models', method: 'GET' },
-  qwen:         { url: 'https://dashscope-intl.aliyuncs.com/compatible-mode/v1/models', method: 'GET' },
-  iflow:        { url: 'https://apis.iflow.cn/v1/models',            method: 'GET' },
-  'github-models': null, // 📖 GitHub Models catalog is public; use chat ping to validate the token.
-  replicate:    null, // 📖 Replicate has no /models endpoint; use chat completions ping
-  cloudflare:   null, // 📖 Workers AI has no auth-check endpoint; use ping only
-  zai:          null, // 📖 ZAI undocumented; use ping only
-  googleai:     null, // 📖 Google AI Studio has no OpenAI-compatible /models; use ping
-  'opencode-zen': null, // 📖 OpenCode Zen uses OpenCode auth only; use ping
-  kilo:         { url: 'https://api.kilo.ai/api/gateway/models', method: 'GET' },
-  llm7:         { url: 'https://api.llm7.io/v1/models', method: 'GET' },
-  routeway:     { url: 'https://api.routeway.ai/v1/models', method: 'GET' },
-  novita:       { url: 'https://api.novita.ai/openai/v1/models', method: 'GET' },
-  'ollama-cloud': { url: 'https://ollama.com/v1/models', method: 'GET' },
-}
-
-// 📖 Sleep imported from shared-helpers.js
-
-// 📖 testProviderKeyDirect: Fast auth-only check using /v1/account or /v1/models.
-// 📖 Fires 3 parallel probes to get a fast decisive result (auth error vs timeout vs 200).
-// 📖 Returns { code, ms } from the first non-timeout response, or the best available.
-async function testProviderKeyDirect(apiKey, providerKey) {
-  const authConfig = PROVIDER_AUTH_ENDPOINTS[providerKey]
-  if (!authConfig) return null
-
-  const { url, method } = authConfig
-  const headers = { Authorization: `Bearer ${apiKey}` }
-  if (providerKey === 'openrouter') {
-    headers['HTTP-Referer'] = 'https://github.com/vava-nessa/free-coding-models'
-    headers['X-Title'] = 'free-coding-models'
-  }
-
-  const parallel = 3
-  const promises = Array.from({ length: parallel }, async () => {
-    const ctrl = new AbortController()
-    const timer = setTimeout(() => ctrl.abort(), 8000)
-    const t0 = performance.now()
-    try {
-      const resp = await fetch(url, { method, headers, signal: ctrl.signal })
-      return { code: resp.status, ms: Math.round(performance.now() - t0) }
-    } catch (err) {
-      const isTimeout = err.name === 'AbortError'
-      return { code: isTimeout ? '000' : 'ERR', ms: isTimeout ? 'TIMEOUT' : Math.round(performance.now() - t0) }
-    } finally {
-      clearTimeout(timer)
-    }
-  })
-
-  const results = await Promise.all(promises)
-  const success = results.find(r => r.code === 200)
-  if (success) return success
-  const authFailure = results.find(r => r.code === 401 || r.code === 403)
-  if (authFailure) return authFailure
-  return results[0]
-}
-
-/**
- * 📖 buildProviderModelsUrl derives the matching `/models` endpoint for providers
- * 📖 that expose an OpenAI-compatible model list next to `/chat/completions`.
- * @param {string} url
- * @returns {string|null}
- */
-export function buildProviderModelsUrl(url) {
-  if (typeof url !== 'string' || !url.includes('/chat/completions')) return null
-  return url.replace(/\/chat\/completions$/, '/models')
-}
-
-/**
- * 📖 parseProviderModelIds extracts ids from a standard OpenAI-style `/models` response.
- * 📖 Invalid payloads return an empty list so the key-test flow can safely fall back.
- * @param {unknown} data
- * @returns {string[]}
- */
-export function parseProviderModelIds(data) {
-  if (!data || typeof data !== 'object' || !Array.isArray(data.data)) return []
-  return data.data
-    .map(entry => (entry && typeof entry.id === 'string') ? entry.id.trim() : '')
-    .filter(Boolean)
-}
-
-/**
- * 📖 listProviderTestModels builds the ordered probe list used by the Settings `T` key.
- * 📖 Order matters:
- * 📖 1. provider-specific known-good overrides
- * 📖 2. discovered `/models` ids that also exist in this repo
- * 📖 3. all discovered `/models` ids
- * 📖 4. repo static model ids as final fallback
- * @param {string} providerKey
- * @param {{ models?: Array<[string, string, string, string, string]> } | undefined} src
- * @param {string[]} [discoveredModelIds=[]]
- * @returns {string[]}
- */
-export function listProviderTestModels(providerKey, src, discoveredModelIds = []) {
-  const staticModelIds = Array.isArray(src?.models) ? src.models.map(model => model[0]).filter(Boolean) : []
-  const staticModelSet = new Set(staticModelIds)
-  const preferredDiscoveredIds = discoveredModelIds.filter(modelId => staticModelSet.has(modelId))
-  const orderedCandidates = [
-    ...(PROVIDER_TEST_MODEL_OVERRIDES[providerKey] ?? []),
-    ...preferredDiscoveredIds,
-    ...discoveredModelIds,
-    ...staticModelIds,
-  ]
-  return [...new Set(orderedCandidates)]
-}
-
-/**
- * 📖 classifyProviderTestOutcome maps attempted probe codes to a user-facing test result.
- * 📖 This keeps Settings more honest than a binary success/fail badge:
- * 📖 - `rate_limited` means the key is valid but the provider is currently throttling
- * 📖 - `no_callable_model` means the provider responded, but none of the attempted models were callable
- * @param {string[]} codes
- * @returns {'ok'|'auth_error'|'rate_limited'|'no_callable_model'|'fail'}
- */
-export function classifyProviderTestOutcome(codes) {
-  if (codes.includes('200')) return 'ok'
-  if (codes.includes('401') || codes.includes('403')) return 'auth_error'
-  if (codes.length > 0 && codes.every(code => code === '429')) return 'rate_limited'
-  if (codes.length > 0 && codes.every(code => code === '404' || code === '410')) return 'no_callable_model'
-  return 'fail'
-}
-
-// 📖 buildProviderTestDetail explains why the Settings `T` probe failed, with
-// 📖 enough context for the user to know whether the key, model list, or provider
-// 📖 quota is the problem.
-export function buildProviderTestDetail(providerLabel, outcome, attempts = [], discoveryNote = '') {
-  const introByOutcome = {
-    missing_key: `${providerLabel} has no saved API key right now, so no authenticated test could be sent.`,
-    ok: `${providerLabel} accepted the key.`,
-    auth_error: `${providerLabel} rejected the configured key with an authentication error.`,
-    rate_limited: `${providerLabel} throttled every probe, so the key may still be valid but is currently rate-limited.`,
-    no_callable_model: `${providerLabel} answered the requests, but none of the probed models were callable on its chat endpoint.`,
-    fail: `${providerLabel} never returned a successful probe during the retry window.`,
-  }
-
-  const hintsByOutcome = {
-    missing_key: 'Save the key with Enter in Settings, then rerun T.',
-    ok: attempts.length > 0 ? `Validated on ${attempts[attempts.length - 1].model}.` : 'The provider returned a success response.',
-    auth_error: 'This usually means the saved key is invalid, expired, revoked, or truncated before it reached disk.',
-    rate_limited: 'Wait for the provider quota window to reset, then rerun T.',
-    no_callable_model: 'The provider catalog or repo defaults likely drifted; try another model family or refresh the catalog.',
-    fail: 'This can be caused by timeouts, 5xx responses, or a provider-side outage.',
-  }
-
-  const attemptSummary = attempts.length > 0
-    ? `Attempts: ${attempts.map(({ attempt, model, code }) => `#${attempt} ${model} -> ${code}`).join(' | ')}`
-    : 'Attempts: none'
-
-  const segments = [
-    introByOutcome[outcome] || introByOutcome.fail,
-    hintsByOutcome[outcome] || hintsByOutcome.fail,
-    discoveryNote,
-    attemptSummary,
-  ].filter(Boolean)
-
-  return segments.join(' ')
-}
-
 export function createKeyHandler(ctx) {
     const {
     state,
@@ -634,10 +438,8 @@ export function createKeyHandler(ctx) {
   }
 
   // ─── Settings key test helper ───────────────────────────────────────────────
-  // 📖 Verifies an API key by first doing a fast parallel auth-only probe (3×8s)
-  // 📖 to /v1/account or /v1/models, then falling back to chat completion pings.
-  // 📖 Auth-only result is decisive (200=valid, 401/403=invalid); only timeouts or
-  // 📖 providers without auth endpoints fall through to the ping-based approach.
+  // 📖 Verifies an API key by delegating to the shared runProviderKeyTest()
+  // 📖 pipeline. Wraps the result into TUI state for display.
   async function testProviderKey(providerKey) {
     const src = sources[providerKey]
     if (!src) return
@@ -650,98 +452,17 @@ export function createKeyHandler(ctx) {
       return
     }
 
-    // 📖 Fast path: parallel auth-only probes (3×8s) to /v1/account or /v1/models.
-    // 📖 200 = key valid and accepted. 401/403 = key rejected. null = no auth endpoint.
-    const authResult = await testProviderKeyDirect(testKey, providerKey)
-    if (authResult) {
-      if (authResult.code === 200) {
-        state.settingsTestResults[providerKey] = 'ok'
-        state.settingsTestDetails[providerKey] = buildProviderTestDetail(providerLabel, 'ok', [], `Auth-only probe returned HTTP 200.`)
-        return
-      }
-      if (authResult.code === 401 || authResult.code === 403) {
-        state.settingsTestResults[providerKey] = 'auth_error'
-        state.settingsTestDetails[providerKey] = buildProviderTestDetail(providerLabel, 'auth_error', [], `Auth probe returned HTTP ${authResult.code}.`)
-        return
-      }
-      // 📖 Timeout or ERR — fall through to ping-based approach below.
-    }
-
-    // 📖 Slow path: ping-based verification (providers without auth endpoint or timeouts).
     state.settingsTestResults[providerKey] = 'pending'
-    state.settingsTestDetails[providerKey] = `Testing ${providerLabel} across up to ${SETTINGS_TEST_MAX_ATTEMPTS} probes...`
-    const discoveredModelIds = []
-    const modelsUrl = buildProviderModelsUrl(src.url)
-    let discoveryNote = ''
-
-    if (modelsUrl) {
-      try {
-        const headers = { Authorization: `Bearer ${testKey}` }
-        if (providerKey === 'openrouter') {
-          headers['HTTP-Referer'] = 'https://github.com/vava-nessa/free-coding-models'
-          headers['X-Title'] = 'free-coding-models'
-        }
-        const modelsResp = await fetch(modelsUrl, { headers })
-        if (modelsResp.ok) {
-          const data = await modelsResp.json()
-          discoveredModelIds.push(...parseProviderModelIds(data))
-          discoveryNote = discoveredModelIds.length > 0
-            ? `Live model discovery returned ${discoveredModelIds.length} ids.`
-            : 'Live model discovery succeeded but returned no callable ids.'
-        } else {
-          discoveryNote = `Live model discovery returned HTTP ${modelsResp.status}; falling back to the repo catalog.`
-        }
-      } catch (err) {
-        discoveryNote = `Live model discovery failed (${err?.name || 'error'}); falling back to the repo catalog.`
-      }
-    }
+    state.settingsTestDetails[providerKey] = `Testing ${providerLabel}...`
 
-    const candidateModels = listProviderTestModels(providerKey, src, discoveredModelIds)
-    if (candidateModels.length === 0) {
-      state.settingsTestResults[providerKey] = 'fail'
-      state.settingsTestDetails[providerKey] = buildProviderTestDetail(providerLabel, 'fail', [], discoveryNote || 'No candidate model was available for probing.')
-      return
-    }
-
-    // 📖 Parallel ping burst: fire up to 5 probes simultaneously to get fast feedback.
-    const PARALLEL_PROBES = 5
-    const attempts = []
-    let settled = false
-
-    while (!settled) {
-      const batch = []
-      for (let i = 0; i < PARALLEL_PROBES && attempts.length + batch.length < SETTINGS_TEST_MAX_ATTEMPTS; i++) {
-        const testModel = candidateModels[(attempts.length + batch.length) % candidateModels.length]
-        batch.push(ping(testKey, testModel, providerKey, src.url).then(({ code }) => ({ attempt: attempts.length + batch.length + 1, model: testModel, code })))
-      }
-      const batchResults = await Promise.all(batch)
-      attempts.push(...batchResults)
-
-      // 📖 Check outcome after each parallel batch.
-      const outcome = classifyProviderTestOutcome(attempts.map(({ code }) => code))
-      if (outcome === 'ok') {
-        state.settingsTestResults[providerKey] = 'ok'
-        state.settingsTestDetails[providerKey] = buildProviderTestDetail(providerLabel, 'ok', attempts, discoveryNote)
-        settled = true
-        continue
-      }
-      if (outcome === 'auth_error') {
-        state.settingsTestResults[providerKey] = 'auth_error'
-        state.settingsTestDetails[providerKey] = buildProviderTestDetail(providerLabel, 'auth_error', attempts, discoveryNote)
-        settled = true
-        continue
-      }
-      if (attempts.length >= SETTINGS_TEST_MAX_ATTEMPTS) {
-        state.settingsTestResults[providerKey] = outcome
-        state.settingsTestDetails[providerKey] = buildProviderTestDetail(providerLabel, outcome, attempts, discoveryNote)
-        settled = true
-        continue
-      }
+    const result = await runProviderKeyTest(testKey, providerKey, src, {
+      onProgress: ({ attempts, maxAttempts }) => {
+        state.settingsTestDetails[providerKey] = `Testing ${providerLabel}... ${attempts}/${maxAttempts} probes tried.`
+      },
+    })
 
-      // 📖 Show progress between batches, then pause before next round.
-      state.settingsTestDetails[providerKey] = `Testing ${providerLabel}... ${attempts.length}/${SETTINGS_TEST_MAX_ATTEMPTS} probes tried. Retrying in ${SETTINGS_TEST_RETRY_DELAY_MS / 1000}s.`
-      await sleep(SETTINGS_TEST_RETRY_DELAY_MS)
-    }
+    state.settingsTestResults[providerKey] = result.outcome
+    state.settingsTestDetails[providerKey] = result.detail
   }
 
   // 📖 Manual update checker from settings; keeps status visible in maintenance row.
@@ -3763,3 +3484,11 @@ export function createMouseEventHandler(ctx) {
     // 📖 Clicks outside any recognized zone are silently ignored.
   }
 }
+
+export {
+  buildProviderModelsUrl,
+  parseProviderModelIds,
+  listProviderTestModels,
+  classifyProviderTestOutcome,
+  buildProviderTestDetail,
+};