None. It wasn't implemented when I wrote the original patch.

BTW: changing to the other syntax, we should just have a *args overload. Forcing every call site to wrap in a %W, Array or Tuple is just noise.

class Process
  def self.run(*args : String, **kwargs)
    run(args, **kwargs)
  end
end

Let's add that to RFC 25!

Why would the aarch64 CI fail?

I reverted the change to Process.run(args) because it fails with a freshly built compiler:

$ make
$ bin/crystal spec spec/compiler/loader/unix_spec.cr
In src/crystal/system/unix/process.cr:314:21

 314 | {pathname, argv.to_unsafe}
                       ^--------
Error: undefined method 'to_unsafe' for Tuple(Pointer(UInt8), Pointer(UInt8), Pointer(UInt8), Pointer(UInt8), Pointer(UInt8), Pointer(UInt8))

I guess the new API doesn't work with anything but Array?

Yeah. Looks like we're not actually testing that the Enumerable(String) restriction is fully supported. The implementation seems to expect Array(String). 🙈

I removed the revert commit and replaced the Tuple an Array literal so we have Process.run [...] that's a bit more clear than Process.run({...}). I'd have merely dropped the tuple, but Process.run(*args) isn't available (yet).