Surprising behavior of literal bindings when `case` expansion applies about rhombus-prototype HOT 6 CLOSED

Two possibilities for fixing this: a keyword argument for the equal? procedure, or a syntax parameter that holds the identifier.

from rhombus-prototype.

I'd lean towards an optional keyword.
Would a syntax parameter ever accidentally apply to situations where someone wants the normal equal-case behavior inside a clause of another case expression overridden for a different equality?

from rhombus-prototype.

eq?, eqv?, equal?, and equal-always? are all that apply to case, right? (It can’t be an arbitrary procedure due to the dispatch algorithm used.) I’m not sure having an optional directive that recognizes exactly those identifiers is a better idea than having different versions of case altogether.

from rhombus-prototype.

Maybe add a new case variant to Racket, but in a racket/case (or racket/case-always) module instead of racket/base or racket?

from rhombus-prototype.

A new case variant seems like it would either require duplication of the case implementation or internally providing a dispatched variant from the implementation in racket/private/case.rkt.

from rhombus-prototype.

eq?, eqv?, equal?, and equal-always? are all that apply to case, right? (It can’t be an arbitrary procedure due to the dispatch algorithm used.)

I don't think it would need to be limited to those in principle, since it is possible to build custom hash tables, AVL trees, tries, range maps, etc. using arbitrary Racket code and syntax-local-lift-expression their construction. Moreover, an internal data structure for this purpose could get even cleverer and represent itself strictly in terms of vectors and such just to get access to quotability if that's important.

Better than lifting or quoting might be to do a define/match-like design—as in, to introduce a form for defining a function that immediately invokes case on one of its arguments—so that the programmer knows exactly what "definition time" the data structure construction is performed at and can reposition it to wherever in their code they like it. Having a well defined "definition time" would even open up the potential for non-constant expressions in branch conditions, since those expressions could be documented to run specifically at definition time and in the definition site's lexical scope.

from rhombus-prototype.