Fix clippy lint clippy::float_equality_without_abs

[gsserge]

Which issue does this PR close?

This is part of #1255

Rationale for this change

It's beneficial to run clippy as strict as possible.

What changes are included in this PR?

This PR enables clippy::float_equality_without_abs lint in the arrow crate and fixes all the relates testing code for this check to pass.

Are there any user-facing changes?

No.

[jhorstmann]

This float comparison with epsilon hint is a pet peeve of mine. The epsilon should not be needed if we expect a result that can be exactly represented in a float. I think the corresponding float_cmp lint was actually downgraded from "correctness" to "pedantic" recently, so I would prefer to rewrite those asserts without epsilon.

[codecov-commenter]

Codecov Report

Merging #1305 (4673194) into master (8f7c56e) will increase coverage by 0.03%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #1305      +/-   ##
==========================================
+ Coverage   83.01%   83.05%   +0.03%     
==========================================
  Files         180      180              
  Lines       52731    52824      +93     
==========================================
+ Hits        43775    43873      +98     
+ Misses       8956     8951       -5     

Impacted Files	Coverage Δ
arrow/src/array/array_union.rs	90.71% <100.00%> (-0.05%)	⬇️
arrow/src/compute/kernels/aggregate.rs	73.33% <100.00%> (ø)
arrow/src/compute/kernels/arithmetic.rs	90.86% <100.00%> (ø)
arrow/src/compute/kernels/cast.rs	95.21% <100.00%> (ø)
arrow/src/csv/reader.rs	88.12% <100.00%> (ø)
arrow/src/json/reader.rs	83.39% <100.00%> (ø)
arrow/src/datatypes/datatype.rs	66.40% <0.00%> (-0.40%)	⬇️
parquet/src/arrow/converter.rs	63.96% <0.00%> (-0.39%)	⬇️
arrow/src/array/transform/mod.rs	84.51% <0.00%> (-0.13%)	⬇️
... and 8 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 8f7c56e...4673194. Read the comment docs.

[alamb]

@jhorstmann to be clear, would you prefer that we keep the module level #![allow(clippy::float_equality_without_abs))]?

Or perhaps annotate the individual tests with

#![allow(clippy::float_equality_without_abs))]

Or something else?

[gsserge]

It's also possible to allow only in the tests

#![cfg_attr(test, allow(clippy::float_equality_without_abs))]

but I'm actually curious how would you do the check without epsilon.

[jhorstmann]

My suggestion was to remove the #![allow(clippy::float_equality_without_abs))], but also replace assertions like

assert!(5.0 - c.value(0) < f64::EPSILON);

with

assert_eq!(5.0, c.value(0));

where the asserted number can be exactly represented in floating point and where the operation does not involve any rounding. In some earlier rust version the latter assertion would have triggered the float_cmp lint, but I don't think it does that anymore for exact numbers.

[gsserge]

@jhorstmann thanks for the explanation, I'm going to update the PR

[alamb]

Looks great. Thank you @gsserge and @jhorstmann