C# Weekly #7: .NET 11 and C# 15 Ship, a Runtime Async Spin Regression Sparks Debate, and LDM Sets the C# 16 Shortlist

C# · Weekly #7

C# Weekly #7: .NET 11 and C# 15 Ship, a Runtime Async Spin Regression Sparks Debate, and LDM Sets the C# 16 Shortlist

csharpC#Language Weekly.NET 11C# 15

Sources:GitHub Releases + 官方博客 + HN

📦 Releases

Stable Branches and Mainline Status

On November 10, 2026, Microsoft shipped .NET 11.0.0 (GA, SDK 11.0.100) at .NET Conf 2026, with C# 15 as the default language version for net11.0. .NET 11 is a Standard Term Support (STS) release, supported for two years (until November 14, 2028). Visual Studio 2026 v18.12.0 ships alongside it. The release delivers native union types (union), closed hierarchies (the closed modifier), and labeled jump control flow. For the syntax in detail, see our earlier piece C# 15 Explained: Union Types and Closed Hierarchies Land.

The servicing branches received their November security updates:

  • .NET 10.0.15 (SDK 10.0.404, active support);
  • .NET 9.0.23 (SDK 9.0.321, end of life);
  • .NET 8.0.34 (SDK 8.0.428, end of life).

The update contains more than twenty security fixes, most notably an ASP.NET Core elevation-of-privilege flaw (CVE-2026-114146, CVSS 8.8), a .NET denial-of-service flaw (CVE-2026-109475), and Visual Studio remote code execution flaws (CVE-2026-114151, CVE-2026-114152, CVE-2026-118745).

Key Engineering Changes and Their Impact

  1. dotnetup, the first official cross-platform toolchain manager (public preview): dotnetup isolates installs in a per-user directory. Run dotnetup init for an interactive setup, then dotnetup install latest to deploy the toolchain. It resolves the SDK version from the nearest global.json and needs no admin elevation.
  2. C# 15 features are fully in production: union types combined with pattern matching reshape polymorphic data modeling, closed hierarchies eliminate redundant defensive default branches, and extension indexers and collection expression arguments fill in the remaining container-syntax gaps.

Key Roslyn and Tooling Commits This Week

Over the past seven days, Roslyn and the language tooling merged or committed several patches that improve semantic analysis and refactoring:

Repo & PRStatus & DateCore ChangePractical Impact
Roslyn #86446Merged 2026-11-15Fixes address-of data-flow analysis through syntax wrappers such as parentheses, checked, unsafe, and null-suppressionFixes the missing CS0170 diagnostic when taking the address of an unassigned local, closing an uninitialized stack-pointer leak.
Roslyn #86404Merged 2026-11-13Makes CS8940 and CS8204 build-only diagnosticsLightens the IDE language server’s live analysis load; no more spurious editor squiggles while typing custom async builder code.
Roslyn #86378Merged 2026-11-13Fixes Source Generator diagnostic leaks from stale project snapshots in Edit and ContinueClears incremental compilation history so stale error markers no longer linger after a generator error is fixed.
Roslyn #86481Committed 2026-11-15Extends the IDE0410 analyzer to infer both polarities of boolean loop-exit flagsDetects reversed exit logic initialized to either true or false and refactors nested loops into C# 15 labeled jumps in one click.

📝 Deep Dives

1. LDM Publishes Its November 2 Notes: Backlog Purged, C# 16 Candidate Set Drawn

  • What happened: On November 10, the C# Language Design Meeting (LDM) merged its November 2 notes (PR #10421). The team triaged long-pending proposals: breaking-change warnings (#8865), fieldof() for exposing backing fields directly (#9031), and standard union types (#9565) moved to proposals/rejected; code-block switch arms (#8610) and user-defined positional patterns (#8613) moved to the backlog. Dictionary expressions (#8659), closed enums (#9011), and target-typed static member access (#9138) stay in the active set for implementation.
  • Why it matters: The cleanup marks the boundaries of C# 16. Rejecting fieldof() protects property encapsulation from metaprogramming sugar that would erode field accessibility; standard union types are left to the library layer while the language core focuses on type inference. Dictionary expressions ([key: value]) move toward implementation and complete the collection expression story.
  • Who is affected: Architects following C# syntax evolution and open source library maintainers who rely on pattern matching and metaprogramming.

2. .NET 11 Runtime Async Shows a Thread Pool Spin Regression (Issue #135738)

  • What happened: In mid-November, developer hez2010 filed a benchmark report in dotnet/runtime (Issue #135738). It shows .NET 11 GA regressing against RC1 in async suspension scenarios: Task.Yield latency rose from 34.30 ns in RC1 to 249.03 ns, and TaskCompletionSource continuation scheduling from 115.66 ns to 530.47 ns. The cause traces to the spin threshold mechanism (SpuriousDispatchNoSpinThreshold) introduced in PR #132765: idle workers spin excessively while waiting on the semaphore, adding 50% to 110% CPU overhead in low-concurrency, I/O-bound workloads.
  • Why it matters: The bug illustrates the tension between burst throughput and single-request latency under light load. The default spin threshold is tied to two thirds of the processor count, so on servers with 24 or more cores idle threads keep contending. The current production workaround is the environment variable DOTNET_ThreadPool_SpuriousDispatchNoSpinThreshold=0, which forces workers with no work to sleep immediately and restores throughput.
  • Who is affected: Teams running high-frequency-dispatch microservices and engineers building low-latency network gateways (such as gRPC and WebSocket).

3. Visual Studio 18.12 Opens BYOM: Offline Local Models and Enterprise Gateways

  • What happened: On November 11, Microsoft announced general availability of Bring Your Own Model (BYOM) in Visual Studio 18.12. The feature drops the mandatory GitHub sign-in and adds a dedicated Custom Endpoint setting. Developers can point the IDE at private OpenAI-compatible endpoints, connect local inference engines (such as Ollama or LM Studio) or an internal enterprise AI gateway, and keep separate “work” and “personal” credentials for a single provider.
  • Why it matters: Cloud AI assistants have been blocked in finance, defense, and other compliance-heavy settings by the risk of code leaving the network. BYOM brings AI-assisted coding to fully air-gapped environments, so enterprises can satisfy security audits while their developers get architecture explanations, unit test generation, and semantic analysis for C# 15 codebases inside the IDE.
  • Who is affected: Teams on classified intranets, enterprise security and compliance specialists, and engineers who self-host LLMs.

4. C# 15 Union Types Meet the Zero-Value Problem: default Breaks Compile-Time Exhaustiveness

  • What happened: Frontend and functional programming engineer cekrem published an in-depth analysis of C# 15 union types (union Pet(Cat, Dog)). When a union is not explicitly assigned (for example Pet pet = default; or an element of new Pet[3]), its internal reference is actually null. In a switch, the compiler considers the match exhaustive after checking only the concrete cases (Cat and Dog), but at runtime a default instance throws a SwitchExpressionException.
  • Why it matters: Purely functional languages rule out uninitialized zero values at the language level, but C# keeps default because of the CLR memory model. When consuming union instances produced by external serialization or array allocation, developers still need a defensive null arm in switch expressions. Static checking guarantees the cases are complete, but it does not shield you from CLR zero-value semantics leaking through at runtime.
  • Who is affected: Engineers modeling business domains with C# 15 unions and functional programming practitioners.

🔥 Community Buzz

1. Hacker News, 240 Points, “Announcing .NET 11”: Native Runtime Async and the Two-Year Upgrade Treadmill

  • Heat: Hacker News thread (https://news.ycombinator.com/item?id=50442225, 240 points, 313 comments).
  • Focus and core disagreements:
    • Disagreement 1: native async stack readability vs. extensibility. Commenter bob1029 praised .NET 11 for handling async scheduling in the runtime, so stack traces are no longer cluttered with state-machine glue, which cuts debugging time on multithreaded failures. Low-level author ablob worries that once the runtime owns async control flow, advanced extension points such as a custom AsyncMethodBuilder or F#-style Computation Expressions lose room for customization.
    • Disagreement 2: STS support window vs. the cost of breaking changes. Developer feoren criticized .NET 11 for dozens of breaking changes and an STS window of just two years, forcing companies into constant regression testing. Opponents such as AbbeFaria, who operate large services, countered that modern .NET migration has shed the friction of the .NET Framework era, and that a fast major-version cadence is the price of capturing JIT performance gains and keeping cloud resource utilization high.

2. Ivan Gavryliuk Archives Parquet.Net: .NET’s Role in Modern Big Data Infrastructure Diverges

  • Heat: Hacker News thread (https://news.ycombinator.com/item?id=50487044, source blog https://www.aloneguid.uk/projects/parquet-dotnet/#why-am-i-killing-it).
  • Focus and core disagreements: After years of maintaining Parquet.Net, author Ivan Gavryliuk archived the project on the day .NET 11 shipped, saying bluntly that .NET is the wrong ecosystem for handling Apache Parquet files, and that such workloads should move to DuckDB or Rust/Arrow-native engines.
    • Disagreement 1: has the ecosystem been squeezed out by Python and Rust? Supporters argue that in columnar analytics, OLAP, and data lakes, upstream specs and C/C++ bindings create very high ecosystem barriers, .NET has no native commercial user base in data science, and a solo-maintained managed parser has a very poor return on effort.
    • Disagreement 2: CLR compute performance vs. library investment. The other side says that after generations of performance work, C# has unmanaged pointers, SIMD vectorization, and zero-copy slicing, so runtime horsepower is not the problem. They attribute Parquet.Net’s stall to Microsoft steering data engineering resources toward Fabric and Spark wrappers, never cultivating a DuckDB-class managed data compute core.

🔭 Next Week

  • Watch 1: Track the dotnet/runtime team’s fix for Issue #135738 (the spin regression). In particular, when the adaptive idle-dispatch heuristic lands on the main branch, and whether it makes the first cumulative servicing update, 11.0.1.
  • Watch 2: Follow the LDM’s review of PR #10416 (attributes on union cases) in upcoming meetings, and see how attribute metadata is pushed down onto synthesized constructors.