Add #[inline] to Decimal::from_parts for performance

[finnbear]

This makes decoding Decimal in bitcode 10X faster.

It works since bitcode checks the range of limited-range integers (like scale) up front, all at once, using vector instructions, and then uses unreachable_unchecked right before calling Decimal::from_parts to communicate that fact to rustc. This only works if Decimal::from_parts is inlinable.

[Tony-Samuels]

I'm surprised the compiler doesn't just do this TBH, the inline attribute is only a suggestion.

If you'd like to see this in the next v1 release, please raise the same MR against the v1 branch. master is currently working towards an eventual v2.

[finnbear]

I'm surprised the compiler doesn't just do this TBH, the inline attribute is only a suggestion.

AFAIK, the Rust compiler doesn't consider inlining non-generic functions across crate boundaries unless using LTO (link-time optimization), which increases compile times and isn't on by default.

Arguably, all simple functions in rust_decimal should be #[inline].

If you'd like to see this in the next v1 release, please raise the same MR against the v1 branch.

Good to know, I'll keep that in mind!