[ObjC][Gen] Emit strs into __cstring if willBeRelativelyAddressed

[ellishg]

Emit ObjC strings to the __cstring section if willBeRelativelyAddressed is true.

#84300 recently emitted these special strings into their respective sections. However, I recently discovered an error when linking an x86_64 binary.

ld64.lld: error: a.o:(__objc_classname): offset is outside the section

I believe the issue is there is a relocation that has an offset (-4 in this case) outside of the referent section (__objc_classname). After looking at 638e4b0 and 50b5044, I suspect they are a similar issues.

When we relatively address them (which occurs in protocol conformance records), the linker may compute the relative offset before coalescing, leading to an incorrect result.

These cases already set willBeRelativelyAddressed, so if that's the case, also make sure they are emitted into the __cstring section.

[rjmccall]

We've definitely run into that linker bug before, which is why the flag exists, so I'm fine with this.

[ellishg]

I've requested membership, but I don't have it yet. @rjmccall could you test and merge for me?

[rjmccall]

@swift-ci Please test

[ellishg]

@swift-ci Please test macOS platform

[ellishg]

@swift-ci Please test

[ellishg]

@swift-ci Please test

[ellishg]

import-as-instance-method.swift keeps failing for macOS, but it passes locally for me. It doesn't seem related to me. @rjmccall is there a way to merge anyway? Or should I keep trying to rebase?

[rjmccall]

@ellishg There's a PR up to fix it, #84708. We just need to wait until it's merged and then re-test.

[hamishknight]

@swift-ci please test macOS

[hamishknight]

@swift-ci please test Windows

[ellishg]

@swift-ci please test macOS

Oops, I thought I had to rebase and then retest. I'll trigger the tests again

[ellishg]

@swift-ci Please test

[ellishg]

@swift-ci please test Windows