Can we stop rustc --test warning of an unused main()?

[chris-morgan]

$ echo 'fn main(){}' | rustc --test -
<anon>:1:1: 1:12 warning: code is never used: `main`, #[warn(dead_code)] on by default
<anon>:1 fn main(){}
         ^~~~~~~~~~~

I have put #[cfg(not(test))] on main a couple of times (though perhaps #[allow(dead_code)] would be better) and I have just observed another person doing it. Would it be reasonable to make the dead code lint permit main to be unused?

[ghost]

Perhaps it would be better to disable all warnings by default in test mode. Some people use tests as a "poor man's REPL", and warnings are just noise there.

[huonw]

The dead_code lint is really helpful for indicating tests that are missing their #[test] (and also the lints are useful for generally making sure that the tests do what they're supposed to).

[nixpulvis]

👍 for ignoring unused main. I agree keeping other dead code warning is defiantly a good idea, but warning about main is just noise.

[doy]

+1 to specifically ignoring main, but keeping other warnings.

[nixpulvis]

On a related note, ignoring main seems to ignore things when compiled with --test when trying to circumvent this issue.

[brycefisher]

:+ for ignoring main

[steveklabnik]

I'm interested in sheparding a patch like this, if anyone wants to give it a shot.

[Siosm]

@steveklabnik I've made a first try to fix this. This tentative fix is not valid but I already got failed tests that I'm investigating.
I think that I need to find a way to get the attributes associated with a function to check if it as the 'start' or 'main' attribute. The current documentation is unclear on this specific point (see http://doc.rust-lang.org/reference.html#function-only-attributes and http://doc.rust-lang.org/reference.html#marker-types).
Also, are functions starting with a '_' internal only? (src/librustc/middle/dead.rs:521)

[steveklabnik]

Starting with a _ is intended to supress warnings, yeah

[huonw]

@Siosm, NB. it is purely a warning/conventions thing, it just controls the lints, it doesn't have any effect on what is accessible externally or anything like that (that's the pub keyword).

[Siosm]

@huonw Ok. Is this documented somewhere? So far I can not find any mention of it in the reference or in the guide.

[felixc]

This issue appears to also affect the Emacs Flycheck mode for Rust. Emacs users will get Flycheck warnings saying main() is dead code. Arguably it's a bug in their end (is it?), but it's another data point to suggest it's unexpected behaviour.

[tbelaire]

It'd be also a good idea to not just use #[cfg(not(test))] on main, since that'll cause spurious import not used errors on functions that are just imported for use in main.

[daboross]

@tbelaire Using #[cfg_attr(test, allow(dead_code))] instead of #[cfg(not(test))] would fix this error without causing import not used errors. It doesn't look the best, but it does do that.

[steveklabnik]

🎊