Implement concrete Error types for each distribution's new function

[FreezyLemon]

See #221 (comment).

Ignore the massive amount of commits, that's for my organisation and can/will be squashed later.

Outstanding issues:

• Clippy (Empirical): Returning a Result<_, ()> is not best practice. clippy suggests an Option which makes less sense here, so either ignore or return a Empirical directly (the function is infallible) ignored

• MultinomialError has not been implemented (yet). I noticed that the parameter validation is moved to the internal module which implies the check_multinomial function will be reused somewhere else, so I didn't want to start changing stuff before that project is done Changed my mind on this. Should be fine to change, it's not much code to replicate inside new (e.g. Categorical does basically the same check inside new)

• Unit Tests

• Proofread

• After everything is done: Check StatsError to see where it is still used and if it can be replaced by something else

[codecov]

Codecov Report

Attention: Patch coverage is 98.91540% with 5 lines in your changes missing coverage. Please review.

Project coverage is 93.58%. Comparing base (d0a5b04) to head (6cd01db).
Report is 4 commits behind head on master.

Files with missing lines	Patch %	Lines
src/stats_tests/fisher.rs	62.50%	3 Missing ⚠️
src/distribution/multivariate_normal.rs	95.23%	1 Missing ⚠️
src/distribution/multivariate_students_t.rs	96.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #265      +/-   ##
==========================================
+ Coverage   93.54%   93.58%   +0.04%     
==========================================
  Files          53       53              
  Lines       11657    11822     +165     
==========================================
+ Hits        10904    11064     +160     
- Misses        753      758       +5     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[FreezyLemon]

#269 would reduce the effort needed to add tests in this PR significantly, so I'll wait for that.

[FreezyLemon]

• rebased
• added a new test function test_create_err which expects a specific error variant
• added a lot of tests
• added FishersExactTestError (only one variant atm, which wraps around Hypergeometric -> bit boilerplate-y, but most flexible)

edit: Coverage is still only so-so because there is no code path hitting the Display implementations. I think we'll have to take that hit, writing unit tests for that is too much IMO

[YeungOnion]

Regarding coverage, our CI is running nightly cargo, so we could use this attribute on functions (from cargo-llvm-cov)

in lib.rs

/// ... docs

#![cfg_attr(coverage_nightly, feature(coverage_attribute))]

in modules

impl std::fmt::Display for D {
    #[cfg_attr(coverage_nightly, coverage(off))]
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        //...
    }
}

in manifest

[lints.rust]
unexpected_cfgs = { level = "warn", check-cfg = ['cfg(coverage,coverage_nightly)'] }

Let me know if you want me to do the proofreading as part of my review and I'll check that off when I've done it.

[FreezyLemon]

Let me know if you want me to do the proofreading as part of my review and I'll check that off when I've done it.

Thanks for offering. I plan on going through the docs and Display impls once myself, there's some inconsistencies that I know of which I want to fix.

Feel free to proofread when reviewing, but I'd wait until I've done my first pass.

[FreezyLemon]

Alright, I changed the things I wanted to and finally added the #[non_exhaustive] attribute to the error types. Should be good to go now

[YeungOnion]

Gave it all a read, everything matches up and the language is consistent, thanks for this.

[YeungOnion]

added a Send + Sync check like in #226

let me know if that works for you.

[FreezyLemon]

Yeah. I refactored it a bit.. I realized after 226 that it's trivial to do T: Sync + Send, only requiring one helper function which makes it a bit shorter.