Renaming and refactoring DataKey, DataEntry, and other traits; adding ErasedDataProvider

[sffc]

Fixes #243
Fixes #244
Fixes #422
Fixes #433

This PR makes a number of changes to the icu_provider crate.

• The DataProvider crate now has a type parameter, allowing it to be used with generic types and zero-copy deserialization. Types that accept a DataProvider should accept the generic version, which is auto-implemented for the type-erased version.
• The type-erased version is now named ErasedDataProvider, and it lives in the erased module.
• New names for DataKey and friends: a DataRequest encapsulates a ResourcePath. A ResourcePath encapsulates a ResourceKey (compile time) and a ResourceOptions (runtime). The ResourceKey is const and copyable; the ResourceOptions is neither.
• Organized the symbols better in the icu_provider crate, with spot improvements to API.
• Added a new built-in provider named HelloWorldProvider.
• Made the LanguageIdentifier field in ResourceOptions (formerly DataEntry) optional.
• Added benchmarks to provider_fs.

[codecov-io]

Codecov Report

Merging #431 (0a1ee06) into master (90f7381) will decrease coverage by 0.72%.
The diff coverage is 63.51%.

@@            Coverage Diff             @@
##           master     #431      +/-   ##
==========================================
- Coverage   75.56%   74.83%   -0.73%     
==========================================
  Files          92       90       -2     
  Lines        4534     4641     +107     
==========================================
+ Hits         3426     3473      +47     
- Misses       1108     1168      +60     

Impacted Files	Coverage Δ
components/datetime/src/format.rs	100.00% <ø> (ø)
components/datetime/src/lib.rs	100.00% <ø> (ø)
components/plurals/src/data.rs	79.54% <ø> (ø)
components/plurals/src/lib.rs	66.66% <ø> (ø)
components/provider/src/error.rs	0.00% <0.00%> (ø)
components/provider/src/lib.rs	100.00% <ø> (ø)
components/provider/src/structs/dates.rs	0.00% <ø> (-44.69%)	⬇️
.../provider_cldr/src/download/cldr_paths_download.rs	34.37% <ø> (ø)
components/provider_cldr/src/lib.rs	100.00% <ø> (ø)
components/provider_cldr/src/support.rs	11.76% <0.00%> (ø)
... and 24 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 90f7381...0a1ee06. Read the comment docs.

[coveralls]

Pull Request Test Coverage Report for Build d8325f0e02aa3eb26a2accdd518dc1c82d781e69-PR-431

• 329 of 520 (63.27%) changed or added relevant lines in 18 files are covered.
• 26 unchanged lines in 8 files lost coverage.
• Overall coverage decreased (-0.7%) to 74.02%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
components/provider/src/inv.rs	24	25	96.0%
components/provider_fs/src/fs_data_provider.rs	24	25	96.0%
components/provider/src/structs/icu4x.rs	5	7	71.43%
components/provider_fs/src/bin/icu4x-cldr-export.rs	0	2	0.0%
resources/testdata/src/bin/icu4x-gen-testdata.rs	0	2	0.0%
components/provider_fs/src/export/fs_exporter.rs	5	8	62.5%
components/provider/src/hello_world.rs	47	51	92.16%
components/provider/src/structs/plurals.rs	1	6	16.67%
components/provider_cldr/src/support.rs	0	5	0.0%
components/provider/src/data_provider.rs	13	19	68.42%

Files with Coverage Reduction	New Missed Lines	%
components/provider_cldr/src/transform/dates.rs	1	69.03%
components/provider_cldr/src/transform/mod.rs	1	12.5%
components/provider_cldr/src/transform/plurals.rs	1	56.64%
components/provider_fs/src/deserializer.rs	1	41.67%
components/provider/src/data_provider.rs	1	61.9%
components/provider_cldr/src/support.rs	2	11.76%
components/provider/src/error.rs	3	0%
components/provider/src/structs/dates.rs	16	0%

Totals
Change from base Build 90f7381e0cbea53b38d7b6d53279903e4f19efe5:	-0.7%
Covered Lines:	3587
Relevant Lines:	4846

---

💛 - Coveralls

[sffc]

I updated the PR to also add ErasedDataProvider. There is a failing test, which is pending an answer to this question:

I have a trait with several methods. Some of the methods have default implementations, and I want to override those implementations if a generic type parameter satisfies a trait bound. For example, I have a method set_to_default(&mut self), which returns an error by default, but if the struct implements Default, then I want to have a custom implementation of that method.

Here's what I tried, but I am getting an error: https://play.rust-lang.org/?version=stable&mode=debug&edition=2018&gist=d3ace0821bbc3527c8586e45f9ef1200

trait Foo<T> {
    fn method1(&self) -> T;
    fn method2(&self) -> Option<T> {
        None
    }
}

struct Bar<T> where T: Clone {
    data: T,
}

impl<T> Foo<T> for Bar<T> where T: Clone {
    fn method1(&self) {
        self.data.clone()
    }
}

impl<T> Foo<T> for Bar<T> where T: Default {
    fn method2(&self) {
        Some(T::default())
    }
}

[sffc]

I also included Criterion benchmarks comparing the generic version of DataProvider to the dynamic ErasedDataProvider that goes through Any. The results:

     Running /home/sffc/projects/omnicu/target/release/deps/provider_fs-410d506882edbd70
json/overview           time:   [19.138 us 19.361 us 19.586 us]                           
                        change: [-2.5016% -0.9742% +0.6057%] (p = 0.22 > 0.05)
                        No change in performance detected.
Found 1 outliers among 100 measurements (1.00%)
  1 (1.00%) low mild

json/generic            time:   [12.444 us 12.620 us 12.797 us]                          
                        change: [-4.5120% -2.4169% -0.5153%] (p = 0.02 < 0.05)
                        Change within noise threshold.
Found 2 outliers among 100 measurements (2.00%)
  1 (1.00%) low mild
  1 (1.00%) high mild

json/erased             time:   [13.270 us 13.425 us 13.580 us]                         
                        change: [-1.1545% +0.2018% +1.6537%] (p = 0.78 > 0.05)
                        No change in performance detected.

bincode/generic         time:   [10.014 us 10.173 us 10.330 us]                             
                        change: [-6.3577% -4.8819% -3.4220%] (p = 0.00 < 0.05)
                        Performance has improved.

bincode/erased          time:   [10.523 us 10.647 us 10.774 us]                            
                        change: [-0.9207% +0.3940% +1.7870%] (p = 0.57 > 0.05)
                        No change in performance detected.
Found 5 outliers among 100 measurements (5.00%)
  3 (3.00%) low mild
  2 (2.00%) high mild

In table form:

Data Format	Runtime, Generic	Runtime, Erased
JSON	12.620 us	13.425 us
Bincode	10.173 us	10.647 us

So, Generic is faster than Erased, and Bincode is faster than JSON, but the differences are not earth-shattering.

[sffc]

I plan to resolve this problem by requiring Default to be implemented when ErasedDataProvider is used, but not when plain DataProvider is used.

[jira-pull-request-webhook]

Notice: the branch changed across the force-push!

• components/provider/src/data_receiver.rs is no longer changed in the branch

View Diff Across Force-Push

~ Your Friendly Jira-GitHub PR Checker Bot

[Manishearth]

Got the test compiling at https://github.com/Manishearth/omnicu/tree/works

[sffc]

Okay, this is ready for review. When reviewing, you can focus on (1) the API docs in the icu_provider crate and (2) code changes outside of that crate. There are a lot of code changes inside the crate, but I feel pretty confident about them.

[zbraniecki]

lgtm! Minor nits and suggestions inline