Rename the Decimal type to Float?

[radeusgd]

The discussion on #6817 showed that our number type names and Value_Type are misaligned:

Unfortunately with Value_Type with have a bit of inconsistency between it and the Enso types.

Value Type	Enso Type
Value_Type.Integer	Integer
Value_Type.Float	Decimal
Value_Type.Decimal	(currently not supported) BigDecimal

The names for value type were chosen to be close to the SQL type names. Maybe we should rename Enso Decimal into Float?? It is indeed a bit confusing to have this inconsistency. Also indeed, Decimal often suggests precise decimals, where we have IEEE-754 floating points.

Indeed Decimal suggests precise decimals which is currently not the case, as Enso uses IEEE-754 floats. We shouldn't rename Value_Type as it does follow the commonly known conventions of decimal for precise numbers and float for imprecise ones and it is consistent with SQL naming (which was its purpose).

But now, Value_Type and the Enso numeric types are misaligned, and as one can see in the table in a pretty confusing way - both have a Decimal type but it means different things.

We should do something to avoid the confusion.

Shall we rename the Enso Decimal type to Float? That would make it consistent with Value_Type and more precisely reflect that it is IEEE-754 floating points and not precise decimals.

cc: @jdunkerley @wdanilo @JaroslavTulach

[GregoryTravis]

Or Double, since that's what it is?

[JaroslavTulach]

Remove Integer and Decimal, keep just Number? That would at least simplify our inheritance hierarchy!

There is no inheritance in Enso - every object is of its type and of type Any. Except numbers are special. There is no reason for engine to expose its internal implementations of numbers to user. We could call all our internal variants Number.

[radeusgd]

Also, if we do this, there is no way to tell reliably if I'm currently working with precise integers or approximating floats - and sometimes I want to guarantee the precision. Currently, I can guarantee this by working with only Integer type. If we merge the types, I don't know if the precise or approximating mechanism is used under the hood.

[radeusgd]

If anything, I'd rather just remove the Number type and keep separate types Integer and 'whatever name comes from this discussion'.

After all, Number = Integer | Decimal - it's currently a special type, but since a long time there were plans to allow 'naming' sum types and then Number would just be an alias for this sum type.

[GregoryTravis]

Agreed, we need this type hierarchy; even languages with no type hierarchy (C) have numeric type hierarchies of some kind.

[JaroslavTulach]

C needs this some kind of numeric type hierarchy to be fast. Dynamic languages (like JavaScript) don't need that and are OK with single Number type. Internally we of course have to speculate on particular type of value, but there is no need to get such hint from programmers.

[radeusgd]

Internally we of course have to speculate on particular type of value, but there is no need to get such hint from programmers.

But that is just the 'efficiency' part. What about correctness? Many functions make sense only on integers (like indexing an array). Don't we want to be able to express this in the type?

[wdanilo]

There are multiple things to consider here:

1. The simplicity for the user. From this perspective, the Jaroslav's idea to keep just Number is most beautiful, but it will not work, as it is unfortunately important for the user to know what number precision they use when working with databases or data frames. Please confirm this with James and Ned. If we could have just a single Number type, I would be the happiest man alive, but probably its not the case.
2. The current name Decimal was chosen because of point 1, to keep it simple for the user, because the name "float" does not tell anyone who is not computer scientist anything sensible. However, as we need to be synchronized with databases types, this probably was wrong choice, so we should use the same names as databases use.
3. Regarding "double", I hope it was a joke :D

[farmaazon]

3. Regarding "double", I hope it was a joke :D

Of course, it was. f64 is the only answer.

[jdunkerley]

We need to keep the difference between Integer and Decimal (or Float if renamed).

Float is a common name for DB users (as that is the SQL standard).
Decimal implies an accurate base 10 representation which we don't have (and again is a SQL term).

[wdanilo]

In such a case, I believe that Float is the best name here (unfortunately).

[jdunkerley]

@radeusgd - lets add a ticket to do this please.