[Discussion] Brainstorming the Core Runtime & Library Primitives

[axx83]

Hello everyone,

During the initial discussion on csharplang, a piece of insightful feedback was raised: before we can finalize the language syntax for CScaf/Spaces, we need to "hammer out the core runtime, libraries, and other data/messaging primitives."

I couldn't agree more. Elegant syntax is only the surface; the true power of the "Spaces" model lies in a robust runtime foundation.

The goal of this thread is to collectively brainstorm what those foundational primitives might look like. These are big, open questions, and all ideas, from nascent thoughts to radical proposals, are welcome.

---

Key Areas for Discussion

Here are a few core areas that I believe we need to solve.

1. Isolation Primitives

The "Spaces" model proposes two primary levels of isolation: ThreadGroup (in-process) and Isolate (a separate process).

• How can we implement this efficiently on the .NET CLR?
• Creating a new process (Isolate) can be expensive. Are there more lightweight process models we could leverage or build?
• For ThreadGroup, how do we guarantee absolute memory (heap) isolation between Spaces?

2. Memory Management for Unmanaged Spaces

One of the most ambitious features is the concept of a GC-less Unmanaged Space for maximum performance.

• How can a non-GC environment coexist with a GC-managed one within the same application?
• What memory management model should we adopt? Could we draw inspiration from Rust's ownership and borrowing model?
• What should the API for allocating and freeing memory in an Unmanaged Space look like to be both safe and efficient?

3. Inter-Space Communication (ISC) Layer

swait and srun require a highly efficient underlying transport layer.

• For in-process communication (SharedMemory, NamedPipes), how can we optimize data transfer to achieve zero-copy?
• How should the API for this transport layer be designed to be easily extensible with new transport mechanisms in the future?

4. Data Marshalling

When an object is sent from one Space to another (e.g., from Managed to Unmanaged), its state needs to be reconstituted. The initial proposal suggested a special constructor convention.

• Is this the best approach?
• Should we consider other mechanisms, such as implicit/explicit conversion operators, or perhaps using source generators to create mappers automatically?

---

These are just starting points. I am eager to hear your thoughts, analyses, and concerns.

What do you see as the biggest challenge or the most exciting opportunity in building the runtime foundation for CScaf?