[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Miscellaneous loose ends

This page is part of the web mail archives of SRFI 76 from before July 7th, 2015. The new archives for SRFI 76 contain all messages, not just those from before July 7th, 2015.

I list a few things that are unclear to me from the spec.
Some may be omissions or rationale omissions, and some may be just my own
blindness :-)

- Why do the field /name/s in the procedural layer /not/ need to be distinct?
  I could see this feature causing lots of pain.  A rationale may help.

- Are the syntactic layer <field name>s identifier literals (like ELSE) 
  or implicitly quoted symbols?  There is a big difference within the 
  context of macro behaviour, or even simply in the context of deciding the 
  meaning of things like
    (let ((x 1))
      (define-type foo (fields (x immutable))))   
  especially in the context of extensions that might use field names 
  for purposes such as pattern matching.         

- Are the syntactic layer <field name>s the same as the /name/s 
  in the procedural layer?  A priori they do not have to be.  If they
  are, it would imply that they are implicitly quoted symbols.  
  If they are not, they could still be symbols, or they could be identifiers.  
  In either case, do they have to be distinct, and if identifiers, under 
  what predicate (e.g. free-identifier=?, literal-identifier=?)?

- Can <field-name> be reused for <accessor-name> or <mutator-name> as in

   (fields (x-coord (x-coord x-coord-set!))

  or, for that matter, can <field-name> be bound somewhere else in the 
  program without breaking the record type?
- How does define-type bind <record-name> exactly?  For example, in

   (define-type foo ....)
   (let () 
     (define-type foo ....))
   (type-descriptor foo)  
  which type descriptor will be returned?  For uniformity with internal
  defines, I would assume that internal define-types remain local and the
  toplevel FOO would be returned.  Is this correct? 
- Is the procedural layer type descriptor /name/ the same as the 
  syntactic layer <record name>.  This would imply that <record name> is
  a symbol, which makes it unclear how DEFINE-TYPE binds it exactly
  (since it is not symbols, but indentifiers, that are bound by binding forms).
  The big problem I could see is that this would be incompatible with 
  locality of internal define-type's. 

- Why do (sealed) and (opaque) have unnecessary parentheses?

- Why is INIT! a binding form?  It forces one to eta-convert 
  reusable procedures, e.g. (init! (x) (display x)) instead of 
  just (init! display).  It also builds a performance inefficiency 
  into the syntax.  
- In the (nongenerative <uid>) clause, why not have a <uid-expression>
  that can be evaluated.  I can think of cases where you might, e.g., read
  a previously stored uid from a database and wish to insert it in the 
  type declaration.  The current syntax does not allow that. 
  (By the way, I think an evaluable <uid-expression> would actually
  alow one to go the other way and make generative from nongenerative.)   
- I do not understand the comment about the UUID namespace.  The record 
  type is nongenerative after all, whereas UUID is concerned with generativity
  (generating a never before seen ID).
  Specifically, what problems may I run into if I do not use a UUID, e.g.
  what is wrong with:
    (define-type point ...
       (nongenerative point))
- Why is generativity the default?  I should think that normally, every
  time I run a program, I would like my types to be the same, especially
  if I read or write record values.   
- Can I forward an unknown number of arguments to the parent?

- Will there be an external representation for records?