Apple Silicon Support with Fibers Working

[pouyakary]

Description

Port the coroutine subsystem to a custom ARM64 context implementation so Io builds and runs natively on Apple Silicon macOS, removing the old Rosetta-only guidance.

Type of Change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update
• Performance improvement
• Code refactoring

Related Issues

Fixes #465

Changes Made

• Added USE_ARM64_CONTEXT handling and an IoCoroARM64Context field so coroutines can store native register state on macOS arm64 (libs/coroutine/source/Coro.h:12).
• Reworked the ARM64 coroutine implementation to manage stacks via Io allocators, register them with Valgrind, and launch callbacks through the assembly bridge (libs/coroutine/source/Coro_arm64.c:1).
• Simplified the Apple-specific context header to the minimal register snapshot required by the assembly helpers and documented native Apple Silicon support in the README (libs/coroutine/source/arm64-ucontext.h:1, README.md:200).

Testing

• All existing tests pass
• Added new tests for the changes
• Tested on the following platforms:
  • Linux
  • macOS (Intel)
  • macOS (Apple Silicon)
  • Windows
  • Other:

Test Commands

# Commands used to test the changes
make test

[pedro-w]

This change is incorrect. The original assert is correct; test is failing on MacOS (actually on AARCH64 Linux too) in release mode though it passes (for me anyway) in debug mode.

I have a patch to fix this but it is not quite ready yet (more testing needed)

[pouyakary]

Ah sorry this is my bad 😞 Forgot to undo this 😔

[pedro-w]

Just my opinion but I think this needs more scrutiny. For example the commit 92dff38 "fix object tests" makes the test pass by matching the "expected" value to the incorrect "observed" value
Commit ae761b7 "fix the fibers" introduces a shell script, not related to fibers AFAICS (Windows only feature)

And, Io was working on Apple Silicon before - at least I have been running it on my M4 Mac. (since #476 I guess?)

[PedroBatista]

@pedro-w I agree, improving current code would be nice an encouraged if any problem is identified, but that lack of testing in multiple platforms scares me. Especially because these changes are VERY platform and ISA dependent.

[pedro-w]

For background I did contribute a tiny bit to Io many years ago (2016?) and 'rediscovered' it recently; it's a neat project.

I have a few patches on the go but they need sorting out before I can do PRs, briefly

1. getting the build to work on MSVC windows (and in general the cmake scripts to take advantage of newer features in cmake)
2. ARM32 support (tested on an old raspberry pi, which worked but then the same code failed on Debian running in qemu-system-arm)
3. getting the memory tracking option to work (IO_CHECK_ALLOC)
4. ARM64 when compiled with optimisation - this is, I think, because the context object doesn't store the floating point registers and they are used, amongst other things, for loop indexes (that's why the coro test in ObjectTest.io fails)
5. One of the PRs I did submit 3229936 has an error 😨

On this PR specifically I am not criticising but I would like to understand better what it is doing as some of the commits don't seem to match their description or the aim of the PR.

Thanks.

[pedro-w]

Especially because these changes are VERY platform and ISA dependent.

I can test Windows 32 and 64 bit, WSL (Debian 64 bit) and MacOS M4, also ARM32 Debian on Raspberry Pi. I don't have anything on FreeBSD / OpenBSD.

[pouyakary]

I actually thought that Io is dead for good and then tried to revive it to work on my system and felt of it as a shot in the dark using many patchy work and GPT help. I'm happy to see this is not! @pedro-w would you contribute your changes to the Io then? The official build needs Rosetta 2 and feels wrong.

[pouyakary]

P.S. Given your comments, I'll close this one since there seems to be better fixes.

[stevedekorte]

Pouyakary, Just wanted to let you know that I value your PRs and I'm sorry that I haven't been more active here.

[pedro-w]

Likewise I apologise if my reply was offputting, I wrote it in a rush and the tone perhaps wasn't what I wanted.

I actually thought that Io is dead for good

Maybe not dead but "moribund" 😉 . I think Io is a really nice code base and I just would like to see it get a polish up for modern platforms.