Test: handler glue integration (errorResult conversion + tdlib wiring)

[kiki830621]

Problem

ServerHandlerLogicTests 直接呼叫 parseGetChatHistoryArgs()，但 handler glue code（Server.swift:351-364）負責：

1. HandlerArgError/DateParseError → errorResult(e.description) 轉換
2. CallTool.Result.isError = true 旗標
3. 從 GetChatHistoryArgs 的 6 個欄位 wire 到 tdlib.getChatHistory(...)

From verification of #7 by Codex (MEDIUM):
「這批新增測試沒有真正測到 handler 行為，只測到抽出的 parser。現在的測試沒有驗證 CallTool.Result.isError、錯誤訊息內容、也沒有驗證 handler 層的參數轉接，因此若之後是 handler glue 壞掉而不是 parser 壞掉，這組測試仍會全綠。」

Type

enhancement

Expected

加 handler-level integration test 類 fixture：

• Mock 或 stub TDLibClient.getChatHistory spy call 參數
• 驗證 errorResult 的 isError: true + message 格式
• 驗證 GetChatHistoryArgs 6 個欄位正確傳到 tdlib 呼叫
• 可考慮用 protocol-based mock 或 Swift 的 @MainActor test helper

Code Reference

• Sources/CheTelegramAllMCPCore/Server.swift:348-366 (handler glue)
• Tests/CheTelegramAllMCPTests/ServerHandlerLogicTests.swift (只測 parser)

Related: #7

Implementation history

• 2026-04-27: extracted errorResultFromParse(_:) -> CallTool.Result shared helper (commit d0750a2)
• 2026-04-27: created HandlerGlueTests.swift with 6 catch-chain tests
• 2026-04-27: subsumed Catch chain integration tests for handler error wiring (#7/#13 follow-up) #18
• 2026-04-27: verify-DA F2/F3 in-scope fix (commit 67451ef) — extended message truncation + CHANGELOG precedence note
• 2026-04-27: closed PARTIALLY — verify-DA F1 (HIGH) + Codex confirmed HandlerGlueTests is structural mimicry (doesn't exercise actual Server.handleToolCall). Real handler integration test → follow-up Real Server.handleToolCall integration test (HandlerGlueTests is structural mimicry only) #27

---

Current Status

Phase: closed-partial
Last updated: 2026-04-27 by idd-close

Key Decisions

• Pragmatic split: Part A (catch chain wiring helper, in-scope) + Part B (real handler integration test, follow-up Real Server.handleToolCall integration test (HandlerGlueTests is structural mimicry only) #27)
• 抽 errorResultFromParse(_:) module-level helper
• 拆檔 HandlerGlueTests.swift
• Subsume Catch chain integration tests for handler error wiring (#7/#13 follow-up) #18
• verify-DA F1 (HIGH): HandlerGlueTests structural mimicry → closed PARTIALLY with explicit follow-up Real Server.handleToolCall integration test (HandlerGlueTests is structural mimicry only) #27
• verify-DA F2/F3: in-scope fix at commit 67451ef

Scope Changes

• 順便 subsume Catch chain integration tests for handler error wiring (#7/#13 follow-up) #18
• TDLib parameter spy 留 Part B follow-up
• verify-DA F2/F3 in-scope fix
• verify-DA F1 → 開 follow-up Real Server.handleToolCall integration test (HandlerGlueTests is structural mimicry only) #27 (real handler integration test)

Blocking

(none — accepting PARTIALLY closure with explicit follow-up #27)

Test counts

• Batch total: 163 → 180 (+17)
• HandlerGlueTests: 6 + 2 (verify-DA F2) = 8 tests

Commits

• d0750a2 refactor: extract errorResultFromParse + handler glue tests (Test: handler glue integration (errorResult conversion + tdlib wiring) #14, subsumes Catch chain integration tests for handler error wiring (#7/#13 follow-up) #18)
• 67451ef fix: cap errorResultFromParse messages + document precedence change (Test: handler glue integration (errorResult conversion + tdlib wiring) #14 verify DA F2+F3)

Subsumed

• Catch chain integration tests for handler error wiring (#7/#13 follow-up) #18 closed as subsumed by Test: handler glue integration (errorResult conversion + tdlib wiring) #14 (PARTIALLY follows)

Follow-up (verify-DA)

• Real Server.handleToolCall integration test (HandlerGlueTests is structural mimicry only) #27 — Real Server.handleToolCall integration test (HIGH priority — addresses DA F1 + Codex independent confirmation that HandlerGlueTests is structural mimicry)
• DST fall-back boundary tests for parseUntilDate / validateDateRange (fixed TimeZone) #28 — DST fall-back boundary tests (continuation of Test design: fixed-TZ assertion to catch symmetric TZ mutations #16/TZ symmetric mutation guard for testUntilDateUsesEndOfDay (parity with #16) #19 family, DA INFO)

[kiki830621]

Diagnosis

Type

enhancement (test infrastructure)

Analysis

ServerHandlerLogicTests 直接呼叫 parseGetChatHistoryArgs() pure function，但 handler glue code 負責：

1. HandlerArgError / DateParseError → errorResult(e.description) 轉換
2. CallTool.Result.isError = true 旗標
3. 從 parsed args 6 個欄位 wire 到 tdlib.getChatHistory(...)

From verification of #7 by Codex (MEDIUM) + #13 verify DA F1 (MEDIUM):
「pure function 路徑全 cover，wiring 層 0 cover」

#13 verify 開了 #18 處理 catch chain integration tests（broader：cover 兩個 handlers）。#18 與 #14 完全重疊——都是 handler glue test。implement 一次，close #18 as subsumed by #14。

Strategy

問題：TDLibClient 是 public final class（無 protocol），mocking 需 protocol abstraction → breaking change，太大。

Pragmatic split：

Part A (in-scope)：catch chain wiring — 不需 TDLib 實例

• 抽 errorResultFromParse(_ error: Error) -> CallTool.Result pure helper（DRY 兩個 catch chain in Server.swift:393-399, 511-517）
• Test that helper directly: 給 HandlerArgError → 回 errorResult with isError=true 且 description 正確；給 DateParseError 同樣；給 unknown Error → fallback localized
• 額外：寫 thin handler test 用 parseGetChatHistoryArgs + mock errorResultFromParse callback 驗證 catch order

Part B (out-of-scope, 新 follow-up)：TDLib parameter spy

• 需要 protocol abstraction 才能驗 「parsed.chatId 真的傳到 tdlib.getChatHistory(chatId:)」
• 開新 issue 處理 architectural change

Complexity

Simple — Part A 是 pure helper extract + 4-5 unit tests，無 source 大改。

Risks

• 抽 helper 後 Server.swift 仍 catch + return；無 behavior change
• Test 用 thin assertion，不需 TDLib instance

---

Implementation Plan

Direct-commit path。

Phase 1: extract errorResultFromParse helper

• Server.swift 新增 internal func errorResultFromParse(_ error: Error) -> CallTool.Result（mirror errorResult 風格）
• Server.swift case "get_chat_history": 的 catch chain 改用 helper
• Server.swift case "dump_chat_to_markdown": 的 catch chain 改用 helper
• swift build pass + 163 tests still pass (no behavior change)

Phase 2: handler glue integration tests

• 新增 Tests/CheTelegramAllMCPTests/HandlerGlueTests.swift（拆檔避免 ServerHandlerLogicTests.swift 越界 426→500+，呼應 Split ServerHandlerLogicTests.swift (426 lines) into focused per-concern files #26 拆檔趨勢）
• testCatchChainHandlerArgError: errorResultFromParse(HandlerArgError(message: "x")) → CallTool.Result with isError=true, content[0].text == "Error: x"
• testCatchChainDateParseError: errorResultFromParse(DateParseError(input: "bad", expected: "YYYY-MM-DD")) → isError=true with proper description
• testCatchChainUnknownError: errorResultFromParse(NSError(...)) → fallback to localizedDescription
• testGetChatHistoryGlueChatIdMissing: 模擬 handler-level，給 args without chat_id，呼叫 parseGetChatHistoryArgs in do/catch + errorResultFromParse → 驗證最終 CallTool.Result
• testDumpToMarkdownGlueOutputPathMissing: 同上 for dump

Phase 3: Subsume #18 + CHANGELOG

• Close Catch chain integration tests for handler error wiring (#7/#13 follow-up) #18 with cross-link comment（in implement step 5, not now）
• CHANGELOG.md [0.5.4] Test section

Commit strategy

2 atomic commits：

1. Phase 1 — extract helper (refactor, no behavior change)
2. Phase 2-3 — new test file + CHANGELOG + Catch chain integration tests for handler error wiring (#7/#13 follow-up) #18 subsume

Scope check

✅ Part A 對應 #14 catch chain coverage。Part B (TDLib spy) 開新 issue 處理 architectural concern，不混入本 batch。#18 subsume in same scope (both are handler glue tests).

Out-of-scope (new follow-up after #14)

• TDLib parameter spy: 需 TDLibProtocol abstraction，cover 「parsed args 真的傳到 tdlib.getChatHistory(_:)」。涉及 21+ caller refactor 與 dependency injection — 應為獨立 issue。

[kiki830621]

Implementation Complete

Checklist (synced from TaskList)

Phase 1: extract errorResultFromParse helper

• HandlerArgs.swift 新增 internal func errorResultFromParse(_ error: Error) -> CallTool.Result
• Server.swift case "get_chat_history": 改用 helper（5 行 → 3 行）
• Server.swift case "dump_chat_to_markdown": 改用 helper
• swift build pass + 172 tests still pass (no behavior change)

Phase 2: handler glue integration tests

• 新增 Tests/CheTelegramAllMCPTests/HandlerGlueTests.swift（拆檔避免 ServerHandlerLogicTests 越界，呼應 Split ServerHandlerLogicTests.swift (426 lines) into focused per-concern files #26 趨勢）
• testHandlerArgErrorBecomesIsErrorTrue
• testDateParseErrorBecomesIsErrorTrue
• testUnknownErrorFallsBackToLocalizedDescription (defensive)
• testGetChatHistoryGlueChatIdMissing (end-to-end glue: parser-throws → errorResultFromParse)
• testDumpToMarkdownGlueOutputPathMissing (dump parity)
• testGetChatHistoryGlueInvalidDate (DateParseError 不誤入 unknown branch)

Phase 3: Subsume #18 + CHANGELOG

• CHANGELOG.md [0.5.4] Refactored + Test entries
• Close Catch chain integration tests for handler error wiring (#7/#13 follow-up) #18 with cross-link comment (in next step)

Legend: - [x] done · - [~] skipped (may revisit) · - [-] won't fix (scope)

Changes

• d0750a2 refactor: extract errorResultFromParse + handler glue tests (Test: handler glue integration (errorResult conversion + tdlib wiring) #14, subsumes Catch chain integration tests for handler error wiring (#7/#13 follow-up) #18)

Files Changed (#14 scope)

 HandlerArgs.swift             | +28 (errorResultFromParse helper)
 Server.swift                  | +4 -10 (refactor 2 catch chains)
 HandlerGlueTests.swift        | +144 (new file, 6 tests)
 CHANGELOG.md                  | (in 0.5.4)

Test counts

• Before: 172 tests
• After: 178 (+6)
• All pass

Subsumed

• Catch chain integration tests for handler error wiring (#7/#13 follow-up) #18 (Catch chain integration tests for handler error wiring) — fully covered by Test: handler glue integration (errorResult conversion + tdlib wiring) #14 implementation; will close as subsumed by Test: handler glue integration (errorResult conversion + tdlib wiring) #14 with cross-link.

Out of scope (new follow-up)

• TDLib parameter spy verification: 需 TDLibProtocol abstraction 才能驗 「parsed.chatId 真的傳到 tdlib.getChatHistory(chatId:)」。涉及 21+ caller refactor + dependency injection — 應為獨立 issue（會在 verify 階段開）。

Scope Compliance

✅ 完整覆蓋 #14 catch chain 部分 + subsume #18。Part B (TDLib spy) 如 plan 所述留 follow-up，避免 architectural change 混入本 batch。

Next: verification pending

[kiki830621]

Verify: Batch 4 (#9 + #10 + #14 + subsumed #18)

Engine

Agent Team (5 Claude reviewers: requirements / logic / security / regression / devils-advocate) + Codex (gpt-5.5 xhigh) = 6-AI review

要求覆蓋率

• 改善 stderr log 格式: ASCII 箭號 + DateParseError internal visibility #9 2/2 reqs addressed (B1 FULLY; B2 won't-fix with cross-link to MCP handler 型別強制 + CLI parity: max_messages string bypass + CLI date behavior 不一致 #8 A2)
• Input hardening: DateParseError input truncate + since/until range sanity check #10 2/2 reqs FULLY (C1 truncate + C2 since>until check)
• Test: handler glue integration (errorResult conversion + tdlib wiring) #14 PARTIALLY — Req 1 catch chain wiring (helper extracted + 6 tests but tests don't exercise actual Server.handleToolCall), Req 2-3 isError + message format FULLY for helper, Req 4 TDLib spy NOT (intentional Part B deferral)
• Catch chain integration tests for handler error wiring (#7/#13 follow-up) #18 PARTIALLY by association (catch chain integration tests via helper, not via real handler)

Findings（合併去重，按 severity 排序）

#	Severity	Finding	Source	Action
1	—	Behavior 100% equivalent for legitimate callers; truth tables verified across 14 cases for parseMaxMessages, 8 cases for validateDateRange	team:logic + codex	PASS
2	—	DateParseError truncation 4 boundary tests pass; #11 contract assertion "2026/04/17" preserved	team:logic + codex	PASS
3	—	Scope check passed: 8 files in scope, no Package.swift changes	team:regression + codex	PASS
4	—	DateParseError public visibility (#9 B2 won't-fix) supported by Package.swift cross-target requirement (CLI uses)	team:codex	PASS
5	HIGH	#14/#18 PARTIALLY: HandlerGlueTests doesn't exercise actual Server.handleToolCall — it tests errorResultFromParse helper + manually mimics catch chain. If Server.swift dropped helper for hardcode errorResult("changed!"), all tests still PASS but production breaks. The test models the glue but doesn't invoke the glue.	team:devils-advocate + codex	Follow-up #27
6	MEDIUM	HandlerArgError had no input truncation (asymmetric to #10 C1's DateParseError protection). Future HandlerArgError(message: "invalid output_path: \(longPath)") could reflect 1MB user input. In-scope fix 67451ef — added 256-char cap in errorResultFromParse for ALL message sources.	team:devils-advocate	In-scope fix ✓
7	MEDIUM	Error precedence change undocumented: validateDateRange runs before parseMaxMessages, so requests with both invalid will now return date-range error first (was max_messages error in 0.5.3). In-scope fix 67451ef — added "Note (verify-DA F3 — error precedence)" to CHANGELOG 0.5.4.	team:devils-advocate	In-scope fix ✓
8	LOW	testSinceEqualsUntilSameDayAccepted name misleading: tests "same date with start vs end of day asymmetry" not "since == until equal". Lucky alignment due to parseISODate + parseUntilDate asymmetry.	team:devils-advocate	Follow-up
9	LOW	errorResultFromParse fallback branch is dead code today (only HandlerArgError + DateParseError reach it). testUnknownErrorFallsBackToLocalizedDescription pins behavior that won't execute.	team:devils-advocate + security forward-looking	Won't fix (defensive value preserved)
10	LOW	Commit e7a9515 coalesces #9 B1 (ASCII arrow) + #10 C1 (truncate) — git bisect would point to commit with two unrelated fixes.	team:devils-advocate	Won't fix (already past bisect window; future best practice noted)
11	LOW	CHANGELOG "Won't fix" section is non-standard per Keep a Changelog 1.1.0 (only Added/Changed/Deprecated/Removed/Fixed/Security). Tools may misclassify.	team:devils-advocate	Won't fix (human-readable, edge tool concern)
12	INFO	validateDateRange placed in parser layer not handler layer (issue #10 plan said handler) — better architecture but spec deviation.	team:devils-advocate + requirements (acknowledged in implementation history)	Won't fix (deviation rationalized, accepted)
13	INFO	DateParseError truncation tests sparse (10/128/129/200 only — no 50/100/500/1000 char cases). Edge: threshold change from 128→127 wouldn't be caught.	team:devils-advocate	Won't fix (boundary tests cover principal mutations)
14	INFO	DST fall-back boundary not tested — but inherited from #15/#19 gap, not new in batch 4.	team:devils-advocate	Follow-up #28 (continuation of #16/#19 family)
15	LOW (forward-looking)	errorResultFromParse fallback uses error.localizedDescription — could leak NSError metadata (file path, errno) if 3rd-party Error type with sensitive localizedDescription is introduced.	team:security	Acknowledge; covered by F2 cap

Devil's Advocate 反駁有效率

DA + Codex 一致指出 HIGH F1: HandlerGlueTests 是 structural mimicry，不打 actual handler。其他 4 reviewer 都判 #14 為 PASS，DA + Codex 反駁 → 降為 PARTIALLY。

• 2 in-scope fix (67451ef)：F2 (HandlerArgError truncation gap) + F3 (CHANGELOG precedence note)
• 2 follow-up issues (will be opened): Real Server.handleToolCall integration test (HandlerGlueTests is structural mimicry only) #27 real handler integration test, DST fall-back boundary tests for parseUntilDate / validateDateRange (fixed TimeZone) #28 DST tests
• 5 won't-fix with rationale (forward-looking, edge tool concerns, defensive value)

Codex 獨立貢獻

• 跑 swift test --filter HandlerGlueTests (6 pass) + --filter for range/truncation (5+4 pass)
• 直接讀 Package.swift 確認 CLI cross-target dependency
• 與 DA 一致認定 HandlerGlueTests 是 structural mimicry — 跨模型 corroboration 非常有力

Scope Check

PASS — 改動限於 8 檔案（HandlerArgs / Server / DateParsing / TDLibClient / DateParsingTests / HandlerGlueTests / ServerHandlerLogicTests / CHANGELOG），全部追溯 #9/#10/#14/#18 + verify-DA in-scope fixes。

Tests

• 163 → 180 tests pass, 0 failures (+17 net)
  • +4 DateParseError truncation (Input hardening: DateParseError input truncate + since/until range sanity check #10 C1)
  • +5 since/until range (Input hardening: DateParseError input truncate + since/until range sanity check #10 C2)
  • +6 HandlerGlueTests for Test: handler glue integration (errorResult conversion + tdlib wiring) #14
  • +2 HandlerGlueTests for verify-DA F2 (commit 67451ef)

Commits

• e7a9515 ASCII arrow + DateParseError input truncate (改善 stderr log 格式: ASCII 箭號 + DateParseError internal visibility #9 B1, Input hardening: DateParseError input truncate + since/until range sanity check #10 C1)
• d58f6b2 validate since_date <= until_date in both parsers (Input hardening: DateParseError input truncate + since/until range sanity check #10 C2)
• d0750a2 extract errorResultFromParse + handler glue tests (Test: handler glue integration (errorResult conversion + tdlib wiring) #14, subsumes Catch chain integration tests for handler error wiring (#7/#13 follow-up) #18)
• 67451ef cap errorResultFromParse messages + document precedence change (Test: handler glue integration (errorResult conversion + tdlib wiring) #14 verify DA F2+F3)

判決

PASS — 無 blocking findings 留存；2 個 MEDIUM in-scope fix 已修復（67451ef）；2 個 valid follow-ups (#27 + #28) 待 user triage；其餘 LOW/INFO won't-fix 理由透明。#14/#18 標 PARTIALLY closed — open follow-up #27 處理 real handler integration。

[kiki830621]

Closing Summary (PARTIALLY closed — see Follow-up section)

⚠️ This issue closes with PARTIALLY closed status. The core architectural concern (real handler-level integration test that exercises Server.handleToolCall) is intentionally tracked in follow-up #27. This is honest closure — not pretending Part B is done.

Problem

#7 verify by Codex (MEDIUM) flagged:

「ServerHandlerLogicTests 直接呼叫 parseGetChatHistoryArgs() pure function，但 handler glue code (Server.swift:351-364) 負責 (1) HandlerArgError/DateParseError → errorResult 轉換 (2) CallTool.Result.isError = true 旗標 (3) 從 parsed args 6 個欄位 wire 到 tdlib.getChatHistory(...)。這批新增測試沒有真正測到 handler 行為，只測到抽出的 parser。若之後是 handler glue 壞掉而不是 parser 壞掉，這組測試仍會全綠。」

Root Cause

• Pure function 路徑全 cover，wiring 層 0 cover：MCP 層補 handler 行為 unit test（get_chat_history） #7 抽 parser pure function 是 testability win，但只解了「parser 邏輯」一半。Handler glue 仍是 untested path
• TDLibClient 是 public final class，無 protocol abstraction：mock TDLib 需要 architectural change（21+ caller refactor + dependency injection），會把單一 issue 變大型重構

Solution (Part A — in-scope)

Pragmatic split: Part A (catch chain wiring helper, in-scope) + Part B (TDLib parameter spy, follow-up due to architectural change required)。

Part A 完整實作：

• 抽 errorResultFromParse(_ error: Error) -> CallTool.Result module-level helper (HandlerArgs.swift)
• Server.swift case "get_chat_history": 改用 helper（5 行 → 3 行）
• Server.swift case "dump_chat_to_markdown": 改用 helper
• 新增 Tests/CheTelegramAllMCPTests/HandlerGlueTests.swift（拆檔避免 ServerHandlerLogicTests 越界 Split ServerHandlerLogicTests.swift (426 lines) into focused per-concern files #26 split trigger）
• 6 tests cover：HandlerArgError → isError=true / DateParseError → 同樣 shape / unknown error fallback / get_chat_history glue / dump_chat_to_markdown glue / DateParseError 不誤入 unknown branch
• Subsume Catch chain integration tests for handler error wiring (#7/#13 follow-up) #18 (DA finding from Refactor dump_chat_to_markdown handler to use parseGetChatHistoryArgs pattern #13 verify, catch chain integration tests — overlapping scope)

verify-DA F2 in-scope fix (commit 67451ef): 抽 messageLengthCap = 256 給 errorResultFromParse — 補上 HandlerArgError 對稱保護缺口（之前 DateParseError 自身有 #10 C1 truncation，HandlerArgError 沒有；reflected user input 上限 256 chars + marker）。+2 new tests cover 256-char boundary + 500-char truncate。

verify-DA F3 in-scope fix: CHANGELOG 加 error precedence note — validateDateRange 在 parseMaxMessages 之前，同時觸發兩 errors 時 caller 收 range error。

What's NOT solved (Part B → #27)

HandlerGlueTests 是 structural mimicry test — 它測 errorResultFromParse(_:) helper 與「parser → helper」程式形狀，但沒打 actual Server.handleToolCall。如果有人把 Server.swift catch arm 從 errorResultFromParse(error) 改成 hardcode errorResult("changed!")，全部 6 個 HandlerGlueTests 仍會 PASS（因為它們測 helper），但 production MCP client 會收到錯誤行為。

Codex 獨立確認此 gap（cross-model corroboration）。需要 architectural change：

• Server.handleToolCall 從 private 升 internal（or 抽 testable wrapper）
• Stub TDLibClient init flow 給 test instance
• 或 protocol abstraction TDLibProtocol 給 dependency injection

→ Follow-up #27 tracks this。

Verification

Engine：5 Claude reviewers + Codex = 6-AI。

維度	Result
Req 1 catch chain helper extracted	PARTIALLY (helper 存在 + 6 tests 鎖 helper contract，但 tests 不打 actual handler — DA F1 + Codex agree)
Req 2 isError true assertion	FULLY for helper
Req 3 message format assertion	FULLY for helper
Req 4 TDLib parameter spy	NOT (intentional Part B deferral → #27)
#18 catch chain integration	PARTIALLY (subsumed by #14, 同 PARTIALLY status)

DA F1 (HIGH) + Codex 一致：HandlerGlueTests 是 structural mimicry。4/5 reviewer 都判 PASS，DA + Codex 反駁 → 降為 PARTIALLY。Honest closure 而非 close-and-pretend。

Changes

• d0750a2 refactor: extract errorResultFromParse + handler glue tests (Test: handler glue integration (errorResult conversion + tdlib wiring) #14, subsumes Catch chain integration tests for handler error wiring (#7/#13 follow-up) #18)
• 67451ef fix: cap errorResultFromParse messages + document precedence change (Test: handler glue integration (errorResult conversion + tdlib wiring) #14 verify DA F2+F3)

Subsumed

• Catch chain integration tests for handler error wiring (#7/#13 follow-up) #18 (Catch chain integration tests for handler error wiring) — 同 PARTIALLY status

Follow-up

• Real Server.handleToolCall integration test (HandlerGlueTests is structural mimicry only) #27 — Real Server.handleToolCall integration test (HandlerGlueTests is structural mimicry only) — HIGH priority, Codex + DA cross-confirmed
• DST fall-back boundary tests for parseUntilDate / validateDateRange (fixed TimeZone) #28 — DST fall-back boundary tests (continuation of Test design: fixed-TZ assertion to catch symmetric TZ mutations #16/TZ symmetric mutation guard for testUntilDateUsesEndOfDay (parity with #16) #19, DA INFO)

Why close as PARTIALLY?

• Part A is genuinely complete：errorResultFromParse helper + 8 tests + Server.swift catch refactor + F2 truncation gap fix
• Part B (real handler integration) requires architectural change — 不能在單一 issue 內 hack 進去而不破壞 scope discipline
• Open Real Server.handleToolCall integration test (HandlerGlueTests is structural mimicry only) #27 with explicit rationale 比 close 假裝完整解決 issue 誠實，且 audit trail 清楚（commit d0750a2 + 67451ef 對應的是 Part A，未來有人考古不會誤以為 Part B 也修了）