Convert run-pass tests with trivial main to build-pass.

[cjgillot]

If main is empty, having a run-pass test does not bring anything more than a build-pass one.

[rustbot]

r? @petrochenkov

(rustbot has picked a reviewer for you, use r? to override)

[tmiasko]

Multiple entry points (under #[cfg(...)], --test, etc):

• tests/ui/abi/abi-sysv64-register-usage.rs
• tests/ui/env-null-vars.rs
• tests/ui/issues/issue-21634.rs
• tests/ui/rfcs/rfc-1014-stdout-existential-crisis/rfc-1014-2.rs
• tests/ui/simple_global_asm.rs
• tests/ui/test-attrs/test-type.rs
• tests/ui/test-attrs/test-vs-cfg-test.rs
• tests/ui/threads-sendsync/thread-local-extern-static.rs
• tests/ui/weird-exit-code.rs

Potentially testing if built executable works:

• tests/ui/abi/relocation_model_pic.rs
• tests/ui/cfg/crt-static-off-works.rs
• tests/ui/cfg/crt-static-on-works.rs

[petrochenkov]

I assume you used some script or regex to remove the attributes.
Could you tweak it to remove lines entirely, instead of leaving empty lines?

[petrochenkov]

Maybe it makes sense to add #[warn(...)] attributes to keep the warnings when they are significant.

[petrochenkov]

False positive, the empty main is under cfg.

[petrochenkov]

Many of these can probably be check-pass, but build-pass is already an improvement.

Also need to double check any tests involving potential life before and after main - global variables with constructors/destructors, #[link_section], #[used], maybe something else + also conditional main like in #119484 (comment).
+ also custom codegen options that may result in the produced binary crashing even with empty main.

[Noratrieb]

this main is not very trivial

[the8472]

This tests dynamic linking. The original issue #12133 also looks like it encountered issue at runtime.

I'm concerned that these changes are applied too broadly and some other subtle things that happen before/after main are being missed.

[cjgillot]

I should really avoid filing PR 30mins before new year. Thankfully that happens only once a year.