SET/CASE and GET/CASE for case-sensitive selection in MAP! and OBJECT!

Submitted by: fork

The nature of OBJECT! vs. MAP! has often been questioned, as many dynamic languages (such as JavaScript) have a single key/value store that is used for both purposes. This brings to the table the question of how the two are different--or should be different--at both an interface and implementation level.

The desire of case-sensitivity in MAP! has brought to the forefront an issue to look at. I will summarize an explanation I gave in chat of the design space here:

A general rule of Rebol is that case is preserved, but not significant unless a lookup requests it.

>> blk: ["A" 1 "a" 2]
>> select blk "a"
== 1
>> select/case blk "a"
== 2

This applies whether you are using strings or words:

>> blk: [A 1 a 2]
>> select blk 'a
== 1
>> select/case blk 'a
== 2

With objects, however, assignments run in sequence:

>> select object [A: 1 a: 2] 'a
== 2
>> select/case object [A: 1 a: 2] 'a
== 2

This is in accordance with assignments through set-word!, which in Rebol do not have a way to set with case. The default evaluator ignores case, and you cannot override it with a /CASE refinement on GET or SET:

>> A: 1
== 1
>> a: 2
== 2
>> print a
== 2
>> print A
== 2

But imagine that there existed a /CASE refinement on OBJECT, MAP, SET and GET. It would be theoretically possible for OBJECT and MAP (a.k.a. MAKE OBJECT! and MAKE MAP!) to store information in a case-sensitive way...accounting for duplicates, and then retrieve it by default in a case-insensitive way, with an override if you wanted case-sensitivity.

Unlike blocks, there is not a linear order to determine who would be the "winner" in a case-insensitive read or write to an object. If it is storing values for both the key a and A then which should obj/a or obj/A overwrite?

What you'd need would be to basically have a list of the case-equivalent written variants. Each SET/CASE call would move the case instance you used last to the top of the list as the next one to be read by an un-cased GET. If you were to ever use a SET without a /CASE it would clear out the list and replace it with a single entry of that casing.

>> obj: object/case [XyZ: 100 xYz: 200]
>> print obj/xyz
200
>> print get/case 'obj/XyZ
100
>> set/case 'obj/AbC 10
>> set/case 'obj/aBc 20
>> print obj/abc
20
>> print get/case 'obj/AbC
10
>> obj/abc: 3
>> print get/case 'obj/AbC
** Script error: cannot access AbC in path obj/AbC
>> print obj/AbC
3
>> print get/case 'obj/abc
3

While this may seem weird for OBJECT!, it is pretty much what you need to get the behavior working for MAP! that people are asking for. MAP/CASE would need to be pretty much the same to make sense.

^{CC - Data [ Version: r3 master Type: Wish Platform: All Category: Datatype Reproduce: Always Fixed-in:none ]}

metaeducation / rebol-issues

SET/CASE and GET/CASE for case-sensitive selection in MAP! and OBJECT! #2172