Unstable functionality in Serde

[dtolnay]

Currently unstable

• Never type Tracking issue for promoting ! to a type (RFC 1216) rust-lang/rust#35121
  • Serialize and Deserialize impls for !

Stabilized

• Serialize and Deserialize for NonZero* Tracking issue for std::num::NonZero* types rust-lang/rust#49137
  • Stabilized in 1.0.60 Implement for std::num::NonZero* on Rust 1.28+ #1278
• Deserialize for Box<CStr> Tracking issue for CString, OsString, PathBuf extra methods rust-lang/rust#40380
  • Requires CString::into_boxed_c_str
  • Stabilized in 1.0.61 Stabilize Deserialize for Box<CStr> #1284
• Deserialize for dynamically sized Rc / Arc Tracking issue for From<&[T]> for Rc & co rust-lang/rust#40475
  • Requires From<Box<T>> for Rc<T> / Arc<T>
  • Stabilized in 1.0.61 Stabilize impls for dynamically sized Rc / Arc #1285
• core::error::Error Tracking Issue for Error in core rust-lang/rust#103765
  • Stabilized in 1.0.210 Stabilize no-std Error trait #2818

Removed

• Specialization Tracking issue for specialization (RFC 1210) rust-lang/rust#31844
  • The default collect_seq and collect_map used to specialize for ExactSizeIterator.
  • Removed in Remove len hint specialization #1841.

[nox]

Feature zero_one just died. Can we have new versions on both 0.9 and 1.0?

[SimonSapin]

Submitted #898 for zero_one.

[Techcable]

593bcb0 breaks the unstable feature because you removed the nonzero feature flag which is still required by NonZeroU16 and friends.

[dtolnay]

The nonzero feature flag was removed from nightly in rust-lang/rust#50808.

[Techcable]

Thanks so much for the clarification, I didn't realize I was using an outdated nightly!

[dtolnay]

I updated the list to reflect functionality that was stabilized in Serde 1.0.60 and 1.0.61. These are no longer behind the "unstable" cfg.

[dtolnay]

Added never_type in #1302.

[dtolnay]

Added error_in_core in #2344.

[gootorov]

Would it be appropriate to add support for Vec<T, A: Allocator>/Box<T, A: Allocator> under the unstable feature flag?
If so, I can submit a PR in some near future.

[jdygert-spok]

Error in core has been stabilized. Would it be appropriate to make StdError = core::error::Error when minor >= 81?