opencode-thinking-fix 2.0.1 → 3.1.4

package/CHANGELOG.md

@@ -1,106 +0,0 @@
-# Changelog
-
-All notable changes to `opencode-thinking-fix`.
-
-## [2.0.0] - 2026-06-24
-
-### Changed (breaking)
-- **Proxy memory leak fixed**: replaced unbounded `Map` with `LRUCache(500)`, O(1) eviction, zero timer overhead.
-- **SIGTERM/SIGINT cleanup**: `process.on` handlers drain active sessions and close the HTTP server cleanly. 5-second force-exit safety net.
-- **SSE parser replaced**: hand-rolled state machine → `eventsource-parser` (368k dependents). Built-in `maxBufferSize: 1MB` guard.
-- **Plugin `export default` removed**: named export `{ ThinkingFixPlugin }` only. Bare default caused TUI double-load at 71ms+73ms.
-- **`engines` field added**: `node >=18` required (node:http, structuredClone).
-- **`oc-plugin: ["server"]` field added**: explicit server-only target for OpenCode 1.3.x loader.
-
-### Added
-- `eventsource-parser` runtime dependency.
-- Provider docs evidence: all 5 providers (DeepSeek, Z.AI, Kimi, MiniMax, Xiaomi MiMo) officially confirm the `reasoning_content` bug in their published documentation.
-
-### Removed
-- `export default` bare export from plugin.
-- Hand-rolled SSE parser (~100 lines replaced by 18-line `eventsource-parser` integration).
-
-### Fixed
-- Plugin 175→92 lines (-47%). Proxy 409→422 lines (net: +13, but SSE parser is now battle-tested and memory-safe).
-
-## [1.1.9] - 2026-06-23
-
-### Added
-- **File-based logging**: JSON-lines log at `~/.local/share/opencode/thinking-fix.log`, captures `plugin_loaded`, `inspect`, `patched`, and `error` events with timestamps.
-- **`inspect` event**: logs field coverage on every request, `isReasoningModel`, `assistantTurns`, `missingContent`, `missingReasoningContent`, `missingReasoning`, proving plugin activity even when zero patches needed.
-- **Console fallback**: `writeLog` catches file write failures and emits `console.error` so failures are visible somewhere.
-
-### Changed
-- `client.app.log()` calls now use `await` for proper async handling.
-- Removal of `hook_fired` debug artifact, replaced by structured `inspect` event.
-
-### Fixed
-- **TUI install flow**: Ctrl+P "install plugin" → type `opencode-thinking-fix` no longer errors. Server-only plugin with bare `export default` properly installs via TUI and CLI without TUI target warnings.
-
-## [1.1.8] - 2026-06-23
-
-### Fixed
-- **package.json aligned with project Documents**: description shortened, `homepage` added, `peerDependenciesMeta.optional` set to `false`, `repository.url` normalized, keywords added (`mimo`, `minimax`), duplicate keywords removed.
-- **Stale `node_modules/` deleted** from repo (was committed accidentally).
-- **Stale npm cache entry** (`npm i opencode-thinking-fix`) removed from `~/.cache/opencode/packages/`.
-
-## [1.1.7] - 2026-06-23
-
-### Fixed
-- **TUI plugin loading error**: removed broken `opencode-thinking-fix-tui.ts`. OpenCode tried to load it as a TUI plugin at startup and failed because it didn't export `{ tui() }`. Package is now server-only.
-- **Plugin default export**: switched from v1 format (`{ server: Plugin }`) to v0 legacy format (bare `export default Plugin`), matching the proven opencode-wakatime pattern.
-- **package.json**: removed `oc-plugin` field, removed `dependencies`, moved `@opencode-ai/plugin` to `peerDependencies` (type-only import, no runtime dep needed), whitelisted individual files instead of entire directories.
-
-## [1.1.6] - 2026-06-23
-
-### Fixed
-- Exposed a no-op TUI plugin target so installation via `Ctrl+P` / `opencode plugin opencode-thinking-fix` completes without the "Package has no TUI target" warning.
-
-## [1.1.5] - 2026-06-23
-
-### Fixed
-- **MiniMax-M3 JSON parse errors in OpenCode Go mode**: model name was not parsed in fixed-upstream mode, so `reasoning_split: true` was never injected for MiniMax models routed through port 3458. Now model name is parsed regardless of `UPSTREAM_URL`.
-- **`<think>` tags leaked into `delta.content`**: added a fallback extractor that strips `<think>...</think>` blocks from content, moves them into the reasoning buffer, and forwards a sanitized SSE chunk.
-- **Kimi K2.6/K2.7 parameter rejection**: strip all hardcoded sampling parameters (`temperature`, `top_p`, `top_k`, `presence_penalty`, `frequency_penalty`, `n`, and `thinking`/`reasoning_effort` for K2.7) before sending to Moonshot.
-
-### Changed
-- SSE stream parser now re-serializes and forwards modified chunks instead of forwarding the raw upstream bytes unchanged.
-
-## [1.1.4] - 2026-06-23
-
-### Fixed
-- Strip all Moonshot-hardcoded sampling parameters for Kimi K2.6/K2.7 thinking mode, not just `top_p`/`top_k`.
-
-## [1.1.3] - 2026-06-23
-
-### Fixed
-- Strip `top_p` and `top_k` from request body for Kimi/Moonshot models.
-
-## [1.1.2] - 2026-06-23
-
-### Fixed
-- Interleaved thinking support: reasoning can now appear after content in the same streaming turn (GLM-5+, MiniMax-M3). Removed premature flush on content appearance.
-- Inject `reasoning_split: true` for MiniMax models to keep reasoning out of `content`.
-
-## [1.1.1] - 2026-06-23
-
-### Fixed
-- Parse MiniMax-M3 `reasoning_details[]` array format and replay it on subsequent turns.
-
-## [1.1.0] - 2026-06-23
-
-### Fixed
-- Restored OpenCode 1.3.x plugin loader compatibility by exporting `{ server: ThinkingFixPlugin }` and adding `oc-plugin: ["server"]`.
-
-## [1.0.1] - 2026-06-22
-
-### Added
-- MIT LICENSE.
-- `files` whitelist in `package.json`.
-
-## [1.0.0] - 2026-06-22
-
-### Added
-- Initial release: plugin + proxy + watchdog.
-- Support for DeepSeek, Kimi, GLM, and MiMo `reasoning_content` replay.
-- Two-proxy architecture (direct providers on port 3457, OpenCode Go on port 3458).

package/tests/test-plugin.js

@@ -1,228 +0,0 @@
-'use strict'
-
-// ─────────────────────────────────────────────────────────────────────────────
-// patchMessages, ported from opencode-thinking-fix-universal.ts (lines 31-77)
-// ─────────────────────────────────────────────────────────────────────────────
-function patchMessages(messages) {
-  let count = 0
-
-  // Step 1: Self-detect reasoning model
-  const isReasoningModel = messages.some((wrapper) => {
-    const msg = wrapper?.info ?? wrapper
-    return msg?.role === 'assistant' && ('reasoning_content' in msg || 'reasoning' in msg)
-  })
-
-  // Step 2: Non-reasoning model, exit immediately
-  if (!isReasoningModel) return 0
-
-  // Step 3: Patch all assistant turns missing required fields
-  for (const wrapper of messages) {
-    const msg = wrapper?.info ?? wrapper
-    if (!msg || msg.role !== 'assistant') continue
-
-    if (!msg.content && msg.content !== '') {
-      msg.content = msg.tool_calls?.length ? 'call tool' : ''
-      count++
-    }
-
-    if (!('reasoning_content' in msg)) {
-      msg.reasoning_content = ''
-      count++
-    }
-
-    if (!('reasoning' in msg)) {
-      msg.reasoning = ''
-      count++
-    }
-  }
-
-  return count
-}
-
-// ─────────────────────────────────────────────────────────────────────────────
-// Helpers
-// ─────────────────────────────────────────────────────────────────────────────
-function deepClone(o) { return JSON.parse(JSON.stringify(o)) }
-
-let passed = 0
-let failed = 0
-const failures = []
-
-function test(name, fn) {
-  try {
-    fn()
-    passed++
-    console.log(`  PASS: ${name}`)
-  } catch (e) {
-    failed++
-    const msg = `${name}, ${e.message}`
-    failures.push(msg)
-    console.log(`  FAIL: ${msg}`)
-  }
-}
-
-function assert(cond, msg) {
-  if (!cond) throw new Error(msg || 'assertion failed')
-}
-
-// ─────────────────────────────────────────────────────────────────────────────
-// Test Cases
-// ─────────────────────────────────────────────────────────────────────────────
-console.log('\n=== patchMessages Regression Tests ===\n')
-
-// ── TC1: Native reasoning model (reasoning_content) ─────────────────────────
-test('TC1: Native reasoning model detects via reasoning_content', () => {
-  const msgs = deepClone([
-    { role: 'user', content: 'hello' },
-    { role: 'assistant', content: 'hi there', reasoning_content: 'step 1...' },
-  ])
-  const result = patchMessages(msgs)
-  assert(result === 1, `expected 1 patch, got ${result}`)
-  assert(msgs[1].reasoning === '', 'reasoning should be added as empty string')
-  assert(msgs[1].reasoning_content === 'step 1...', 'should not overwrite existing reasoning_content')
-  assert(msgs[0].role === 'user' && msgs[0].content === 'hello', 'user message must not be touched')
-})
-
-// ── TC2: OpenCode Go model (reasoning, NO reasoning_content) ─────────────────
-test('TC2: OpenCode Go model detected via reasoning field', () => {
-  const msgs = deepClone([
-    { role: 'assistant', content: 'ok', reasoning: 'step 1...' },
-  ])
-  const result = patchMessages(msgs)
-  assert(result === 1, `expected 1 patch, got ${result}`)
-  assert(msgs[0].reasoning_content === '', 'reasoning_content should be added as empty string')
-  assert(msgs[0].reasoning === 'step 1...', 'should not overwrite existing reasoning')
-})
-
-// ── TC3: Non-reasoning model (Qwen/GPT/Claude) ──────────────────────────────
-test('TC3: Non-reasoning model returns 0 with zero modifications', () => {
-  const msgs = deepClone([
-    { role: 'user', content: 'hello' },
-    { role: 'assistant', content: 'hi' },
-  ])
-  const result = patchMessages(msgs)
-  assert(result === 0, `expected 0 patches, got ${result}`)
-  assert(!('reasoning_content' in msgs[1]), 'reasoning_content must not be injected')
-  assert(!('reasoning' in msgs[1]), 'reasoning must not be injected')
-})
-
-// ── TC4: Mixed messages (user + assistant + user) ───────────────────────────
-test('TC4: Only assistant messages are patched, user messages untouched', () => {
-  const msgs = deepClone([
-    { role: 'user', content: 'q1' },
-    { role: 'assistant', content: 'a1', reasoning_content: 'thinking' },
-    { role: 'user', content: 'q2' },
-  ])
-  const result = patchMessages(msgs)
-  assert(result === 1, `expected 1 patch, got ${result}`)
-  assert(msgs[1].reasoning === '', 'assistant reasoning should be added')
-  assert(!('reasoning_content' in msgs[0]), 'user message must not get reasoning_content')
-  assert(!('reasoning' in msgs[0]), 'user message must not get reasoning')
-  assert(!('reasoning_content' in msgs[2]), 'second user message must not get reasoning_content')
-  assert(msgs[2].content === 'q2', 'second user message content unchanged')
-})
-
-// ── TC5: Multiple assistant turns ───────────────────────────────────────────
-test('TC5: Both assistant messages patched when at least one has reasoning', () => {
-  const msgs = deepClone([
-    { role: 'assistant', content: 'first', tool_calls: [{ function: { name: 'f' } }] },
-    { role: 'assistant', content: 'second', reasoning: 'thinking' },
-  ])
-  const result = patchMessages(msgs)
-  // msgs[0]: +reasoning_content, +reasoning (2). msgs[1]: +reasoning_content (1). Total = 3.
-  // msgs[0].content is 'first' (truthy) so no content patch.
-  assert(result === 3, `expected 3 patches, got ${result}`)
-  assert(msgs[0].reasoning_content === '', 'first assistant should get reasoning_content')
-  assert(msgs[0].reasoning === '', 'first assistant should get reasoning')
-  // first assistant content is 'first' (truthy) so NOT patched
-  assert(msgs[0].content === 'first', 'first assistant content unchanged')
-  assert(!('content' in Object.getOwnPropertyDescriptor(msgs[1], 'content') || true) || msgs[1].content === 'second', 'second assistant content unchanged')
-  assert(msgs[1].reasoning_content === '', 'second assistant should get reasoning_content')
-  assert(msgs[1].reasoning === 'thinking', 'second assistant reasoning unchanged')
-})
-
-// ── TC6: Already complete message ───────────────────────────────────────────
-test('TC6: Already complete message, no changes', () => {
-  const msgs = deepClone([
-    { role: 'assistant', content: 'done', reasoning_content: 'done', reasoning: 'done' },
-  ])
-  const result = patchMessages(msgs)
-  assert(result === 0, `expected 0 patches, got ${result}`)
-  assert(msgs[0].reasoning_content === 'done', 'reasoning_content unchanged')
-  assert(msgs[0].reasoning === 'done', 'reasoning unchanged')
-})
-
-// ── TC7: Wrapper format ({ info: Message }) ─────────────────────────────────
-test('TC7: Wrapper format correctly handled', () => {
-  const msgs = deepClone([
-    { info: { role: 'user', content: 'hello' } },
-    { info: { role: 'assistant', content: 'reply', reasoning: 'thinking' } },
-  ])
-  const result = patchMessages(msgs)
-  assert(result === 1, `expected 1 patch, got ${result}`)
-  assert(msgs[1].info.reasoning_content === '', 'wrapped assistant should get reasoning_content')
-  assert(msgs[1].info.reasoning === 'thinking', 'wrapped assistant reasoning unchanged')
-  assert(!('reasoning_content' in msgs[0].info), 'wrapped user must not get reasoning_content')
-  assert(msgs[0].info.content === 'hello', 'wrapped user content unchanged')
-})
-
-// ── TC8: Empty reasoning_content string ─────────────────────────────────────
-test('TC8: Empty reasoning_content still triggers detection', () => {
-  const msgs = deepClone([
-    { role: 'assistant', content: 'ok', reasoning_content: '' },
-  ])
-  const result = patchMessages(msgs)
-  assert(result === 1, `expected 1 patch (reasoning), got ${result}`)
-  assert(msgs[0].reasoning === '', 'reasoning should be added')
-  assert(msgs[0].reasoning_content === '', 'reasoning_content stays empty')
-})
-
-// ── TC9: Empty messages array ───────────────────────────────────────────────
-test('TC9: Empty messages array returns 0', () => {
-  const result = patchMessages([])
-  assert(result === 0, `expected 0, got ${result}`)
-})
-
-// ── TC10: Non-reasoning model with tool_calls ────────────────────────────────
-test('TC10: Non-reasoning model with tool_calls, no injection', () => {
-  const msgs = deepClone([
-    { role: 'assistant', content: '', tool_calls: [{ function: { name: 'f' } }] },
-  ])
-  const result = patchMessages(msgs)
-  assert(result === 0, `expected 0, got ${result}`)
-  assert(!('reasoning_content' in msgs[0]), 'must not inject reasoning_content')
-})
-
-// ── TC11: Reasoning model with empty content + tool_calls ───────────────────
-test('TC11: Reasoning model patches content to "call tool"', () => {
-  const msgs = deepClone([
-    { role: 'assistant', content: '', tool_calls: [{ function: { name: 'f' } }], reasoning_content: 'thinking' },
-  ])
-  const result = patchMessages(msgs)
-  // content is '' (empty string), guard `!msg.content && msg.content !== ''` skips it
-  // because '' !== '' is false. The original code treats '' as valid content.
-  // Only reasoning is patched (reasoning_content and content are already present/invalid).
-  assert(result === 1, `expected 1 patch (reasoning), got ${result}`)
-  assert(msgs[0].content === '', 'content stays empty string')
-  assert(msgs[0].reasoning === '', 'reasoning added')
-  assert(msgs[0].reasoning_content === 'thinking', 'reasoning_content preserved')
-})
-
-// ── TC12: Null wrapper ──────────────────────────────────────────────────────
-test('TC12: Null/undefined wrapper in array is skipped gracefully', () => {
-  const msgs = deepClone([
-    null,
-    { role: 'assistant', content: 'hi', reasoning_content: 'think' },
-    undefined,
-  ])
-  const result = patchMessages(msgs)
-  assert(result >= 1, `expected at least 1 patch, got ${result}`)
-})
-
-// ─────────────────────────────────────────────────────────────────────────────
-// Summary
-// ─────────────────────────────────────────────────────────────────────────────
-const total = passed + failed
-console.log(`\n=== Results: ${passed}/${total} passed, ${failed} failed ===\n`)
-
-process.exit(failed > 0 ? 1 : 0)