Stabilize rounding to increment methods for FixedDecimal

[jedel1043]

Follow up of #4571.

[jedel1043]

The failing action seems to be sporadic; I ran it locally without problems.

[sffc]

Note: No binsize diff.

Main:

-rwxr-xr-x 1 root root    24248 Feb  5 23:27 optim4.elf
-rwxr-xr-x 1 root root    13520 Feb  5 23:27 optim5.elf

PR:

-rwxr-xr-x 1 root root    24256 Feb  5 14:00 optim4.elf
-rwxr-xr-x 1 root root    13520 Feb  5 14:00 optim5.elf

[sffc]

https://github.com/unicode-org/icu4x/blob/main/docs/tutorials/c-tiny/fixeddecimal/test.c doesn't include calls to rounding functions. I think we should probably add that.

[sffc]

On main:

-rwxr-xr-x 1 root root    25568 Feb  6 16:43 optim4.elf
-rwxr-xr-x 1 root root    14464 Feb  6 16:43 optim5.elf

On this branch:

-rwxr-xr-x 1 root root    29112 Feb  6 01:16 optim4.elf
-rwxr-xr-x 1 root root    17520 Feb  6 01:16 optim5.elf

🐖

[sffc]

@jedel1043 I think we can use generics to solve this and stay DRY. Sandbox example:

https://play.rust-lang.org/?version=stable&mode=release&edition=2021&gist=c9c7c4788888d36fc3e1b37e7b9dd36a

Build it as WASM and check the output. Note that trunc and trunc_to_increment compile to different code, with trunc being substantially smaller than trunc_to_increment.

This will cause duplicated code between the increment and non-increment functions, but I think this is okay because if you care about code size you should just use either one or the other, and we know that many users don't need increments so we don't want to penalize those callers.

[jedel1043]

Did a size comparison of all possible solutions:

main

-rwxr-xr-x 1 jedel users    25376 Feb  6 23:19 optim4.elf
-rwxr-xr-x 1 jedel users    14728 Feb  6 23:19 optim5.elf

RoundingIncrement

-rwxr-xr-x 1 jedel users    28768 Feb  6 23:27 optim4.elf
-rwxr-xr-x 1 jedel users    17656 Feb  6 23:27 optim5.elf

IncrementLike

-rwxr-xr-x 1 jedel users    26128 Feb  7 00:25 optim4.elf
-rwxr-xr-x 1 jedel users    15000 Feb  7 00:25 optim5.elf

IncrementLike + #[inline(never)] (current)

-rwxr-xr-x 1 jedel users    25648 Feb  7 00:39 optim4.elf
-rwxr-xr-x 1 jedel users    14920 Feb  7 00:39 optim5.elf

Adding #[inline] to all the internal methods barely changes the size of optim4.elf (16 bytes) and doesn't change the size of optim5.elf.

[sffc]

Thanks for making this change! This looks good to me now to land. After it lands we can play around a little more with the #[inline] attributes to measure the impact.

I'm going to copy @markusicu on this PR in order to grant the stabilization review.

[robertbastian]

Please wait for @markusicu review as well

[sffc]

Notes from the API Review:

https://docs.google.com/document/d/1gKZCaiTG2eeyxrIKUF2tf6PIsaPCIBaUVRXD-JDh6dI/edit#heading=h.7pm5mstzwo40

@markusicu left numerous suggestions about documentation and generally the FixedDecimal API shape which we can fix in the 2.0 timeframe. I will file issues about those. However, @markusicu was happy with the shape and naming of RoundingIncrement and *_to_increment, so we have the i18n Correctness approval to land the new API.

Thanks @jedel1043!