NaN cannot be used as a key in Map

[JaroslavTulach]

The primary motivation forMap existance is implementation of Vector.distinct. To make such implementation easy/fast/possible we don't want errors from Map.insert. On contrary, we want every object to provide hash code - so it can be used as a key for the Map. Constraints:

1. it is desirable for Map to always satisfy: m.insert k v . contains_key k == True
2. For Number we are going to use JavaScript's Equality comparisons and sameness to ensure NaN works as a key
3. Custom types - shall have a forced reflexivity - e.g. Meta.is_same_object implies that they are Any.== which in turn satisfies the first point

Following code shows the expected behavior for custom types that violate reflexivity:

type My_Nan
    Value comment:Text

...comparator such that it always returns Nothing

k = My_Nan.Value "foo"
k2 = My_Nan.Value "foo"
(k==k).should_be_true
(k==k2).should_be_false
m = Map.empty.insert k 10
m.contains_key k . should_be_true
m.get k . should_equal 10
m.contains_key k2 . should_be_false

m2 = m.insert k2 20
m2.get k . should_equal 10
m2.get k2 . should_equal 20
m2.size . should_equal 2

m3 = m2.insert k 30
m3.size . should_equal 2
m3.get k . should_equal 30

e.g. , we will do a best effort solution. Thanks to Any.== relying on Meta.is_same_object, even if compare implementation is always incomparable, x==x will be True. Multiple separate instances of incomparable atoms will however not be.

Follow up of

• Add support for partial ordering to Comparators #5771
• Merge ordered and unordered comparators #5818

Tasks

Preview Give feedback

1. Make sure NaN can be a key in Map using the JavaScript's Equality comparisons and sameness
2. Make sure x == x via Meta.is_same_object (tracked in Any.== should check for Meta.is_same_object as the first step #6065)
3. Update tests
4. Update documentation

[radeusgd]

Probably also worth adding a test for Vector.distinct:

[Number.nan, 1, Number.nan, 2, 1].distinct.to_text . should_equal "[NaN, 1, 2]"

(using to_text as regular equality may actually yield False due to the NaNs...)

[radeusgd]

Let's be sure to test the custom types properly, i.e. have tests for a custom type like My_Nan above.

As @Akirathan mentioned, this may be blocked by #5709

[enso-bot]

Pavel Marek reports a new STANDUP for today (2023-04-17):

Progress: Updating bench results - I had broken cache locally, so I have spent some time investigating why some older bench results are not displayed. Benchmarks https://enso-org.github.io/engine-benchmark-results/ now contain results from Dec 2022, until today. NaN can now be used as a key in the map, and complies to the Same value zero equality JS specs It should be finished by 2023-04-20.

[enso-bot]

Pavel Marek reports a new STANDUP for today (2023-04-18):

Progress: Polish, add some tests. Ready for review. Bookclub, review other PRs. It should be finished by 2023-04-20.

[enso-bot]

Pavel Marek reports a new STANDUP for yesterday (2023-04-19):

Progress: Last nits before merging, adding some last tests. It should be finished by 2023-04-20.