Re: Re: Function memoization

Dimitre Novatchev wrote:

  Hi

> If in the context of memoisation we accept that "identity"
> is equivalent to all-memberwise identity of the items of
> the two sequences, then the effect of memoising long
> sequences arguments will be negative! It will actually be
> harmful!

  Good point.

> I am proposing to leave developers the potential to define
> and implement sequence "identity" in the context of
> memoisation using other definitions (such as the
> reasonable comparison of a single "sequence-reference").

  Good idea.  Do you think to something like the following:

    <xsl:function name="...">
      <xsl:param name="..."/>
      <func:memoisation comparator="..."/>
      ...
    </xsl:function>

where @comparator is a QName identifying a function (with
something like func:sequence-identity() to check identity of
two sequence, which is not implementable in plain XSLT or
XPath).  If yes, two questions:

    - what could be the default comparator?

    - does the arity of the comparator have to be always 2,
      as it will be called once for each pair of argument,
      or 2*(arity of the memoised function), as it will be
      called only once for each memoised value?

> Also, developers could have a flexible definition, so that
> memberwise comparison could be used for sequences up to,
> let's say, ten items and reference (ID) comparison will be
> used for comparing longer sequences.  This treshold can
> even be determined dynamically depending on the speed of
> execution.

  IMHO, it is a bad idea, as this could result in different
results when arguments are sequences of different size.  But
if it is needed, having the ability to set the comparator
let the developer the ability to get it.

> This is the way to achieve maximum effect of memoisation
> without the harm (that would result for long sequence
> arguments) of imposing strict memberwise comparison.

> Think of developers as often clever people that are
> capable of making the right decision in a particular
> situation.

  I completely agree.  Tools have to be designed to minimize
the possibility of writing bad things, default values have
to be well-choosed, but the developer always has to get the
ability to express what he wants.  IMHO.

  Regards,

--drkm























	

	
		
___________________________________________________________________________ 
Yahoo! Mail réinvente le mail ! Découvrez le nouveau Yahoo! Mail et son interface révolutionnaire.
http://fr.mail.yahoo.com