macOS x86_64 port

[DarthMDev]

Discuss bugs and problems found here. Direct issues you find to me please.

[nattthebear]

This is interesting stuff. I'm sorry about the Waterbox TLS problems; that part of the code is a bit of a hack on all OSes. If I recall correctly (and maybe I don't; it has been a long time), we can't use a host OS TLS slot because the guest wouldn't know which slot it is.

It's possible all the TLS stuff could be abandoned; we could instead store struc Context at some well known memory location which we map ourselves. Probably far away from the "normal" waterbox memory. But I'm not sure what other work that would entail at this point.

Stepping back a bit, this port is in an unfortunate place: Rosetta 2 is EOL in 2027, and all the nix winforms stuff has been dead for decades. But if you're fine with that, don't let me discourage you.

Waterbox on ARM is a potentially doable thing, but we'd need to rework the guest toolchain and it would have separate core builds than the x86-64 one.

[DarthMDev]

Yeah, it’s a shame Rosetta 2 is EOL in 2027. That might make things difficult down the road, but baby steps for now!

Your point about bypassing the host OS TLS entirely and mapping ⁠struc Context⁠ to a well-known, isolated memory location makes a lot of sense. It seems like a much cleaner, cross-platform approach than fighting with platform-specific TLS hacks, even if it requires some extra scaffolding. If I hit a wall with the current TLS implementation on macOS, I might explore what shifting away from native TLS slots would look like.

Yeah, the WinForms situation on Unix isn't ideal, but my main goal right now is just seeing how far I can push the x86_64 translation layer and getting the underlying Waterbox logic happy on macOS. Native ARM64 support down the line would be amazing, but dealing with separate core toolchains is definitely a bigger beast for later.