On Wed, Jan 2, 2019 at 1:47 AM Alex Shinn <xxxxxx@gmail.com> wrote:

Given that in the current R7RS there is no notion of type
beyond type predicates, this implementation-strategy would
be allowed, but it may leak with future extensions.

I assume you mean record introspection? Let it leak. Anyone is

free to implement any type as a record or not. In SICL, a new

in-progress implementation of Common Lisp, everything is a

CLOS instance (basically a tuple with a pointer to the class object

and a pointer to the instance data, which always includes the version

number of the class object, to detect obsolete instances) except

immediates and pairs (because they are very common and are

already just two pointers), which are recognized by type tags.

Would it be reasonable to say that the types are disjoint from
all other types except possibly each other?

Disjointness just means that the objects on which

the type predicates return #t are belong to disjoint

(mathematical) sets. Since we don't yet know all the

type predicates, this is a promise that can't quite be

fulfilled yet, but the intention is that it will end up being

true at the end of the R7RS-large process. So the

important point here is that u8vector? and u16vector?

never both return #t on the same object.

There are a few cases where non-disjointness is explicit:

in particular, the fact that SRFI 127 lazy sequences are

(typically improper) lists is deliberately exposed, as is

the fact that the insides of SRFI 117 list queues are

in fact lists rather than random sequence objects.

Note that sets and bags in the sample

implementation of SRFI 113 are a single record type: a

comparator, a hash table mapping elements to the number of

occurrences, and a flag. The set-* and bag-* procedures just

check the flag for correctness and then call the corresponding

unexported sob-* procedure. If saving code space is an issue,

you can always make all three of set-*, bag-* and sob-* the

same procedures.

There turn out to be only two places in the code where it's

necessary to check the flag and do something different:

in the primitive routine to add an element used by various

sob-* procedures, and in another primitive routine that

is used by the set-theoretic operators. Both basically

bound the count to 1.

John Cowan http://vrici.lojban.org/~cowan xxxxxx@ccil.org

Her he asked if O'Hare Doctor tidings sent from far coast and she with

grameful sigh him answered that O'Hare Doctor in heaven was. Sad was the

man that word to hear that him so heavied in bowels ruthful. All she

there told him, ruing death for friend so young, algate sore unwilling

God's rightwiseness to withsay. Joyce, Ulysses, "Oxen of the Sun"