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

Re: s/-LEFT//g?



> On Thu, Oct 16, 2003 at 01:06:29PM -0700, Bradd W. Szonye wrote:
>> As an extension of my idea to change collection-fold-left to
>> collection-fold, how about removing all of the -lefts? Rationale:
>> 
>> 1. It's the naming convention in SRFI-1.
>> 2. The -left suffix doesn't make sense for "disordered" collections.
>> 3. It's less to type.

scgmille@xxxxxxxxxxxxxxxxxx wrote:
> Its less to type, but in this case its important, as it really is
> indicating directionality.  If it was just get-value or such, it
> wouldn't convey the message that its retrieving a very specific value
> from the collection, the left most or least precedent.

But that's only true for directional collections. Nondirectional
collections (like grab-bags and true mathematical sets) have no concept
of "first" or "last." In those cases, there is no message to convey,
because they necessarily return an arbitrary value, not a specific one.

> *-left provides a symmetry that the *-right functions can compare
> themselves against.  

While the names aren't symmetrical, there's just as much *conceptual*
symmetry in all of these:

    Convention                              Precedent
    *-method-left and *-method-right        PLT list.ss library
    *-method and *-method-right             SRFI-1
    *-method and *-reverse-method           C++ Standary Library

I would peronally prefer one of the latter two conventions, because
(1) *-method-left doesn't make sense for non-directional collections,
and (2) it's more to type.

I would also accept an alternate approach where all collections have
*-method and only directional collections have *-method-left and -right.

    Convention          Meaning
    *-method-left       operate from the left (least) end
    *-method-right      operate from the right (greatest) end
    *-method            operate in arbitrary order

Directional (aka "reversible") collections would probably define
*-method to be equivalent to *-method-left, but that shouldn't be a
requirement. This way, you could implement an ordered-hash-table which
would provide these enumerators:

    collection-fold         enumerate in efficient (storage) order
    collection-fold-left    enumerate in sorted order
    collection-fold-right   enumerate in reverse-sorted order

In general, I would expect a directional collection to use *-method for
the most efficient order, *-method-left for guaranteed in-order, and
*-method-right for guaranteed reverse-order. For many directional
collections, but not all of them, *-method and *-method-left will be the
same.

I like this approach because it allows totally generic algorithms (i.e.,
algorithms that don't care whether it's an ordered collection) to use
the most efficient methods -- you only get the (possibly-inefficient)
in-order methods when you specifically ask for them.

I think this would resolve both my objections and yours:

1. No more applying "directional" method names like *-left to
   non-directional collections.
2. Directional/reversible collections have symmetric method names.
3. Less verbose naming convention for common cases.

Plus an added benefit:

4. Directional collections have (possibly more efficient) nondirectional
   methods for users who don't care about the order of elements.

SUMMARY OF PROPOSAL

All containers have *-method, which uses an arbitrary order.

Reversible containers have *-method-left and *-method-right, which use
the container's ordering-function or inherent order (for sequences).

Reversible containers may optionally define *-method to use a specific
ordering (usually "same as *-method-left"  or "most efficient order").
-- 
Bradd W. Szonye
http://www.szonye.com/bradd