Comments on SRFI 122 draft 14 Sudarshan S Chawathe (04 Dec 2016 19:21 UTC)
Re: Comments on SRFI 122 draft 14 John Cowan (04 Dec 2016 23:13 UTC)
Re: Comments on SRFI 122 draft 14 Bradley Lucier (15 Dec 2016 18:46 UTC)

Comments on SRFI 122 draft 14 Sudarshan S Chawathe 04 Dec 2016 19:21 UTC

These questions and comments refer to draft 14 of SRFI 122 and are in
roughly document order (not significance order).

  * (trivial) Extra ';;;' at the end of the description of
    interval-upper-bounds->list.

  * Should interval= be named interval=? instead, by analogy with
    string=?, char=?, etc.?

  * Since interval-intersect? returns a usable 'true' value, should it
    be named interval-intersect, by analogy with member, etc.

  * In the description of storage-class-default et al., the invocation
    of make-storage-class has the positions of getter and setter
    switched.

  * Also at the above location, storage-class-length is missing from
    the sentence describing the return values.

  * Arrays and intervals are specified as being Scheme types different
    from all others.  However, storage classes do not seem to have
    this requirement.  I am guessing this design is intentional,
    perhaps to make some optimized implementations easier, but am not
    sure.

  * (nit) In the description of procedure array?, I assume the
    intention is (the obvious interpretation) that the returned value
    must be #f in other cases but, strictly speaking, that is not
    required by the current phrasing.  Ditto for mutable-array?.

  * For completeness, perhaps the description of array-domain and
    array-getter should allow an optional setter in the invocation
    used in the "if" part of the first sentence there.

  * In the description of specialized-array-default-safe?, is the #t
    intentionally specific, or is any true value (i.e., anything other
    than #f and #false) permissible there?  It would be useful to
    clarify in any case.  There are a few other similar occurrences in
    the document (true v. #t).  Similarly, there are a few places
    where an argument is restricted to being a boolean.  I assume this
    would be interpreted as limiting valid values to those for which
    boolean? yields true.  Unless this restriction is intentional, it
    may be better to go with the usual Scheme convention of
    interpreting any values as boolean.

  * The position of the "new value" argument for array setters (first)
    is different from that used by the storage-class setters.  I
    assume this decision is motivated by reasons similar to those
    noted in SRFI 63, but highlighting of the difference may be
    useful.

  * (noted in an earlier by John Cowan) specialized-array may be
    better named make-specialized-array.  (And perhaps ditto for the
    u16-array example procedure later.)

  * In the description of specialized-array, there's a missing closing
    parenthesis near the end of the sentence just before "If safe is
    #t...".

  * Above location: Though implied, the sentence "If safe is #t..."
    should perhaps say something like "in addition to the following".

  * In the description of specialized-array, the code defining
    (array-setter array) has storage-class-getter where it should use
    storage-class-setter.

  * I am confused by the description of array-indexer and array-body.
    Regarding the statement "(array-indexer array) is assumed to...":
    Isn't (array-indexer array) provided by the implementation instead
    of the user (directly)?  If so, then the assumption seems like it
    should be a guarantee by the SRFI/implementation to the user.
    Also, I could not find a description of array-body (although I can
    guess, especially given the sample implementation, that it returns
    the implied backing-store object for the storage class used by the
    specialized array).

  * There are three instances of "specialized-array-indexer" in the
    document that I assume refer to the "array-indexer" procedure
    above, so that one name should be changed to the other.  (Perhaps
    it would be useful for the procedures that can only be applied to
    specialized array to have a 'specialized-' prefix or some other
    distinguishing feature in their name.)

  * (minor) The example on "shearing" seems to have extra indentation
    for the (define b ...) portion.

  * (minor) In the description of array-permute, 'Array' should be 'array'.

	* (minor) The explanation of array-reverse may be easier to
    understand (and a bit shorter) if the common parts of the code
    were factored out.  Perhaps something along the lines:

      (define (flip-multi-index array flip? . multi-index)
        (map (lambda (i_k flip?_k l_k u_k)
               (if flip?
                   (- (+ l_k u_k -1) i_k)
                   i_k))
             multi-index
             (vector->list flip?)
             (interval-lower-bounds->list (array-domain array))
             (interval-upper-bounds->list (array-domain array))))

	* Would it be desirable for array->list to guarantee lexicographic
    traversal of the array (similar to the guarantee provided by
    array->specialized-array)?  If such a guarantee is added, then it
    would also propagate to array-fold and array-fold-right given
    their definitions in terms of array->list.

  * The description of array-map (and array-for-each) refers to the f
    argument as 'function' instead of the more conventional
    'procedure'.  Since 'function' is used in its mathematical sense
    elsewhere in the document, it may be better to change these
    instances to 'procedure'.

  * Given array-every?, should array-any? also be provided?

	* By analogy with SRFI-1, should array-every? (and array-any? if
    added) return the true value (if any) from the final proc
    invocation?  (In that case, the names array-every and array-any
    would be more conventional.)

  * Should array-every? (and array-any? if added) provide any
    tail-call guarantees similar to the analogous procedures in SRFI 1
    (or would that create implementation problems)?

  * SRFI 25's array-ref and array-set! permit array indices to be
    provided either as separate arguments or as a vector or as a
    1-dimensional array.  Would a similar strategy be useful for the
    relevant procedures of this SRFI?

  * (very minor) Perhaps a small change to the PGM example would
    showcase the SRFI features a bit more: It could use a more
    restrictive storage class for the array it builds, based on the
    PGM header (i.e., u8-storage-class for greys < 256,
    u16-storage-class otherwise, etc.).  But I may be missing some PGM
    format subtleties here.

Regards,

-chaw