Repository navigation
test: use Task.sleep(nanoseconds:) so SwiftLMTests runs in Release on Swift 6.2 - #256
Merged
Merged
Conversation
…ift 6.2 On Swift 6.2.4 (Xcode 26.3) in a Release build, any Task.sleep(for:) call in the SwiftLMTests bundle aborts the XCTest process with "freed pointer was not the last allocation" (swift_task_dealloc, called from the specialized Task<>.sleep<A>(for:tolerance:clock:)). It needs neither cancellation nor any SwiftLM code: a test containing only `try? await Task.sleep(for: .milliseconds(10))` crashes, while the nanoseconds overload does not. Swift 6.4 is not affected. Switch the test sleeps to Task.sleep(nanoseconds:) with the same durations. ClientDisconnectTests (7) and the whole SwiftLMTests bundle (464) pass in Release on Swift 6.2.4. Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Problem
mlx-swift-lmruns the SwiftLM tests in Release on Xcode 26.3 (Swift 6.2.4). There,ClientDisconnectTestsaborts the whole XCTest process:This hid every test after it. The suite passes in Debug and on Swift 6.4.
What I found (reproduced locally with
/Volumes/Work/Xcode-26.3.app, Swift 6.2.4)swift_task_dealloccalled from the specializedTask<>.sleep<A>(for:tolerance:clock:)compiled into the test bundle. It is not inServer.swift's cancellation code.try? await Task.sleep(for: .milliseconds(10))crashes. It does not need cancellation, aTask,DeferredCancellation, NIO, or animport SwiftLM.ContinuousClock().sleep(for:)crashes too.Task.sleep(nanoseconds:)does not crash. An empty async test and a trivialTask { 1 }do not crash.swiftc -Oprogram, nor in a minimal package with one XCTest (tools 5.9 or 6.0, executable target with@main,-enable-testing, Release) on 6.2.4. So I could not reduce it to an upstream bug report. The exact compiler trigger is unknown.Change
Replace the five
Task.sleep(for: ...)calls inClientDisconnectTestsandDFlashStreamBridgeTestswithTask.sleep(nanoseconds:)and the same durations. No production code changes.Verification (Swift 6.2.4,
swift test -c release --no-parallel --skip-build --disable-swift-testing,mlx.metallibcopied next to the test bundle)ClientDisconnectTests: 7 tests, 0 failures, EXIT=0 (before: signal 6).DFlashStreamBridgeTests: 4 tests, 0 failures, EXIT=0.SwiftLMTestsbundle: 464 tests, 10 skipped, 0 failures, EXIT=0.Not verified
Server.swifthas sevenTask.sleep(for:)calls in production code. The same toolchain's Release build runs them inmlx-swift-lm's speculative-decoding script, which passed, but I have not shown they are unaffected. I did not change them.🤖 Generated with Claude Code