compiler: Use -T instead of --script for linker scripts

[workingjubilee]

Note that our run-make test suite uses -T and not --script, interestingly enough:

run-make/rust-lld-link-script-provide/rmake.rs
2:// when the symbols in the `PROVIDE` of the link script can be eliminated.
16:        .link_arg("-Tscript.t")

Fixes #127183

[rustbot]

r? @WaffleLapkin

rustbot has assigned @WaffleLapkin.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[rustbot]

Some changes occurred in compiler/rustc_codegen_ssa

cc @WaffleLapkin

[petrochenkov]

link_arg("-T") will emit -Wl,-T on command line, so it will go directly to linker and clang won't see it, so the explanation doesn't seem right.

[workingjubilee]

wait, then why is it broken for them?

[workingjubilee]

🤔

[workingjubilee]

It must be something about building specifically for a custom target. Thanks for catching this, I'll take a closer look. This patch's code change might still be correct but it might be for a weird reason.

[Noratrieb]

gcc does not support --script either. I suspect the people in #127183 configured their target incorrectly such that cc was invoked with ld-style arguments

[WaffleLapkin]

r? @petrochenkov
since you seem to have context here

[petrochenkov]

I'm ok with using -T instead of --script as a more popular and widely supported option (gcc and clang support -T as an alias for -Wl,-T).

As I understand, this change only "fixes" something when the linker flavor is somehow messed up and ld-style options are passed to cc (#142445 (comment)).

r=me with the comment removed.
If everything is configured correctly, .link_arg("-T") should never pass -T to gcc or clang.
@rustbot author