Place TLS initializers with relocations in .tdata

[ecstatic-morse]

Should fix #70673, although I'm not sure how to test this. Perhaps @joshlf could find a MCVE?

Also adds more context to the FIXME.

r? @oli-obk

[joshlf]

I'll try to pull this and build and test locally. If you're developing on a Mac, you could also just try building alloc-tls as described in the original issue, and see if it both fails before this change and succeeds with this change.

[joshlf]

OK, I built your patch locally, and with it, cargo test succeeds in building and running alloc-tls. That's not an MCVE, but hopefully that gives a bit more confidence that this is correct.

[oli-obk]

I'd assume

#![feature(thread_local)]

#[thread_local]
static A: &u8 = &42;

fn main() {
    dbg!(*A);
}

would be a minimal repro, unsure how to test this without a mac though.

The PR looks fine to me, so

@bors r+

[bors]

📌 Commit fffbcc8 has been approved by oli-obk

[joshlf]

@oli-obk Right you are! Tested your proposed repro locally, and it both crashes the current compiler and also doesn't crash with this PR applied.

[oli-obk]

Superb! Thanks for checking

@ecstatic-morse if you get around to it before this PR goes to bors, please add the repro to the test suite, if not let's keep the issue open as needs-test

[ecstatic-morse]

Added the regression test.

@bors r=oli-obk

[bors]

📌 Commit f030635 has been approved by oli-obk