Turn warnings into errors in std and extra

[brson]

stage1+ only. The deluge of warnings is never ending. Fortunately, some excellent and fearless people do clean them up, but it feels like this would be more efficient to not commit the warnings in the first place.

Having too many warnings looks bad.

[catamorphism]

Agreed. I'd even go so far as to say that all code in the rust repository should be warning-clean. We should set a good example :-) (or if there are warnings we want to ignore in rustc, it would probably be better to opt out of them explicitly).

[graydon]

Agreed

[pnkfelix]

@brson what do you think about turning off warnings entirely for the stage0 build, so that we don't get annoyed by bugs in lint that are not yet fixed in the snapshot version of the compiler?

(or at least add [or just document?] an easy way, perhaps a configure flag, to let one easily opt into ignoring all warnings in the stage0 build.)

[brson]

@pnkfelix That sounds good to me.

@alexcrichton has a pull request open for this but I think we should do a little more. In Alex's formulation all the #[deny(*)] attributes are passed as -D flags on the command line in stage1+. Passing flags on the command line though isn't very rusty - I would rather see these as attributes at the top of the crate.

Two ways we could achieve this would be to: a) add command line flags for converting warnings to errors and ignoring warnings, so each stage just has one extra command line flag instead of all the lint flags, or b) making attributes themselves conditional. I prefer the former.

[alexcrichton]

@brson, what do you think about -D warnings along with a #[deny(warnings)] attribute?

That role of it would be to deny all warnings, turning them into errors by default. There could then be some validation as well to say -A warnings allows all warnings to go unnoticed, and -W warnings itself does nothing (treats warnings as warnings, non fatal)

That way stage0 would have -A warnings and stage1+ would have -D warnings and that's it. That way we could still control the lint levels manually inside the crates without having to specify them all on the command line. Additionally whenever a crate were ready to deny common lints by default, they just need to add #[deny(warnings)]

[alexcrichton]

As an example, see this