From e4bcf7d6fc602c822083c7c0b7bf462feb20a80b Mon Sep 17 00:00:00 2001
From: simba <simba@Mac-mini.local>
Date: Thu, 24 Sep 2026 22:39:45 -0700
Subject: [PATCH] fix: remove Swift 6 captured-var warnings in Server startup
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Three `var`s were captured by concurrently-executing closures, which is a
warning today and an error in the Swift 6 language mode:
- Server.swift:1337 `isVision`: made mutable by the VLM→LLM fallback (#173)
  and read in the /health handler.
- Server.swift:1056 / 1368 `mtpAssistantModelRef`: read inside
  `container.perform` and in the chat-completion route.

Each is now captured through a `let` snapshot taken after its last mutation
(`loadedAsVision`, `loadedMTPAssistant`, and a local `assistant` for the
mainModelRef wiring; DualModelMTP is class-bound via Module, so setting
through the `let` still reaches the same object).

Clean rebuild on Xcode 27 / Swift 6.4 shows no captured-var warnings. Smoke-tested on a
Mac mini M6: Qwen3.6-35B (VLM fallback) /health "vision":false, Qwen3.8-27B
"vision":true, Gemma 4 + bf16 MTP assistant "MTP assistant ready (3 tokens/round)",
and all three answer.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 Sources/SwiftLM/Server.swift | 15 ++++++++++-----
 1 file changed, 10 insertions(+), 5 deletions(-)

diff --git a/Sources/SwiftLM/Server.swift b/Sources/SwiftLM/Server.swift
index 4e0e185..3cc7cb0 100644
--- a/Sources/SwiftLM/Server.swift
+++ b/Sources/SwiftLM/Server.swift
@@ -964,6 +964,9 @@ struct MLXServer: AsyncParsableCommand {
             }
         }
 
+        // Final after the VLM→LLM fallback above; later closures capture this `let`.
+        let loadedAsVision = isVision
+
         print("[SwiftLM] Loaded model configuration. Inferred tool call format: \(String(describing: await container.configuration.toolCallFormat))")
 
         // ── Check if target model supports DFlash ──
@@ -1052,12 +1055,14 @@ struct MLXServer: AsyncParsableCommand {
                 print("[SwiftLM]    Ignoring --mtp-assistant-model; generation will not use MTP.")
             } else {
                 // The assistant drafts *for* this trunk, so it needs a reference to it.
+                let assistant = mtpAssistantModelRef
                 await container.perform { mainContext in
-                    mtpAssistantModelRef?.mainModelRef = mainContext.model
+                    assistant?.mainModelRef = mainContext.model
                 }
                 print("[SwiftLM] MTP assistant ready (\(self.numMtpTokens) tokens/round)")
             }
         }
+        let loadedMTPAssistant = mtpAssistantModelRef
 
         // ── Load DFlash draft model for block-diffusion speculative decoding ──
         let dflashModel: DFlashDraftModel?
@@ -1249,7 +1254,7 @@ struct MLXServer: AsyncParsableCommand {
             minP: self.minP,
             repeatPenalty: self.repeatPenalty,
             thinking: self.thinking,
-            isVision: isVision,
+            isVision: loadedAsVision,
             prefillSize: self.prefillSize,
             turboKV: self.turboKV,
             mtp: self.mtp,
@@ -1334,7 +1339,7 @@ struct MLXServer: AsyncParsableCommand {
                 }
             }
             let payload = """
-{"status":"ok","model":"\(modelId)","vision":\(isVision),"memory":{"active_mb":\(activeMemMB),"peak_mb":\(peakMemMB),"cache_mb":\(cacheMemMB),"total_system_mb":\(totalMemMB),"gpu_architecture":"\(deviceInfo.architecture)"},"stats":{"requests_total":\(snapshot.requestsTotal),"requests_active":\(snapshot.requestsActive),"tokens_generated":\(snapshot.tokensGenerated),"avg_tokens_per_sec":\(String(format: "%.2f", snapshot.avgTokensPerSec))}\(partitionJson)}
+{"status":"ok","model":"\(modelId)","vision":\(loadedAsVision),"memory":{"active_mb":\(activeMemMB),"peak_mb":\(peakMemMB),"cache_mb":\(cacheMemMB),"total_system_mb":\(totalMemMB),"gpu_architecture":"\(deviceInfo.architecture)"},"stats":{"requests_total":\(snapshot.requestsTotal),"requests_active":\(snapshot.requestsActive),"tokens_generated":\(snapshot.tokensGenerated),"avg_tokens_per_sec":\(String(format: "%.2f", snapshot.avgTokensPerSec))}\(partitionJson)}
 """
             return Response(
                 status: .ok,
@@ -1365,7 +1370,7 @@ struct MLXServer: AsyncParsableCommand {
                     draftModelRef: draftModelRef, numDraftTokens: numDraftTokensConfig,
                     dflashModel: dflashModel, dflashBlockSize: dflashBlockSizeConfig,
                     dflashTargetModel: dflashTargetModel,
-                    mtpAssistant: mtpAssistantModelRef
+                    mtpAssistant: loadedMTPAssistant
                 )
             } catch {
                 return Response(
@@ -1465,7 +1470,7 @@ struct MLXServer: AsyncParsableCommand {
             "port": port,
             "model": modelId,
             "engine": "mlx",
-            "vision": isVision
+            "vision": loadedAsVision
         ]
         if let plan = partitionPlan {
             var info = plan.healthInfo