Nightly docs missing relevant float parsing impls

[hanna-kruppe]

#31903 hid all trait impls from doc(hidden) modules. This was a step in the right direction, since it removed several dead links (mentions of traits that really should be hidden). Unfortunately currently FromStr is implemented in the hidden module core::num::dec2flt and thus the current nightly doesn't document this impl at all (neither on the trait's page nor on the primitive pages). For the same reason, ParseFloatError doesn't seem to implement any traits (except Error o.O perhaps because of the bound on the associated type FromStr::Error).

[hanna-kruppe]

cc @mitaa @alexcrichton

Since AFAIK one can't override this behavior with an explicit #[doc(..)] annotation (and several of the traits that ParseFloatError is missing are derived) I fear we have to either ditch the strategy introduced by #31903 or move impl FromStr for $float_type and ParseFloatError back up to core::num. The latter doesn't help other code which might have the same problem. Is there an easy way to implement a rule such as this?

if the trait or implementing type is hidden, don't document the impl

[alexcrichton]

To me when thinking about #31903 it seemed like a mild antipattern to have public interfaces you want documented in a #[doc(hidden)] module. Would it be "easy enough" to move these impls up to core::num? If privacy or something gets in the way we can reconsider reverting perhaps.

[hanna-kruppe]

Well it used to be that way, then (IIRC) you moved the impls into core::num::dec2flt, so it should just work.

[mitaa]

FWIW, I thought of another related testcase which is broken in the same way

// foo.rs
extern crate bar;

pub use bar::Foo;

// bar.rs
pub struct Foo;

mod private {
    trait Bar {}
    impl Bar for ::Foo {}
}

(the impl-inlining code doesn't look at privacy at all)

AIUI this would need to be solved in the same or similar way as necessary for

if the trait or implementing type is hidden, don't document the impl

[alexcrichton]

Oh maybe it was privacy related in that case...