Just noticed I had a partially drafted progress update I didn't end up
sending. It was just about a successful build regarding something I was
trying that I'd talked about, but reading through the thread, it looks
like I just kept working and sent a new progress update email instead
with all the updates folded into it.

I should make clear that none of the -opt*--{target,sysroot} flags in
stage1.*.{ghc,c,link}.opts are necessary; just as I removed the --host
stuff, it seems pretty much all of that is useless. That means for stage
2, only the stage1.*.ghc.c.opts += -optc-fno-ret-protector is necessary.

And I've already mentioned these in previous emails (with patches), but
just to clarify, in case it helps (since I don't know what opaque FILE
means, or how it relates to needing cross-compile scaffolding; sounds
like an abstract FILE pointer in C):

  - We can produce arm64 bindists instead of copying arm64 binaries to an
    amd64 bindist.
  - We can remove the need to provide --sysroot and -pgm* flags to the
    cross-compiled GHC compiler that runs on arm64.
  - I think we can remove the need to provide -L flags to the GHC compiler
    that runs on arm64, but I don't have a patch (though I can provide one).
  - We'll still have to provide -Wl,--no-execute-only but that's already
    the case.

Cheers,
Habib