Make static ns_string!s work inside closures

[madsmtm]

They don't currently for some reason!

Discovered in #254.

[madsmtm]

Similar error when using:

RUSTFLAGS="-Clink-dead-code" cargo build --example basic_usage

[madsmtm]

I've cfg-gated the functionality behind the feature flag unstable-static-nsstring for now: #350

[madsmtm]

objrs adds special export_names, so perhaps that's required?

[madsmtm]

A change in rustc dead code elimination causes the UTF16 to appear in the assembly output, even though it's known statically that the symbol is unused.

Note that this happens even if the #[link_section = "..."] attribute is removed.

Will probably need to bisect rustc to find the regression, it's somewhere between nightly-2023-05-25 and nightly-2023-08-27.

[madsmtm]

I bisected it to rust-lang/rust#115126, probably rust-lang/rust#115114, but also found that it only appears if you place the string in another static, not if you just use it as an expression.

So probably not worth doing more about.

[complexspaces]

Sorry to necropost, but given the Rust bisection work you did last year what is the status of the unstable-static-nsstring flag? Is this something that can be tested as correct now?

[madsmtm]

I doubt it, I suspect the two statics that's needed for this, the one in __DATA,__cfstring and the one in __TEXT,__cstring,cstring_literals, still need to end up in the same codegen unit for the linker to be able to pick this up for sure, and I'm not sure how to make rustc guarantee that happens?