[Rd] question

[Gabor Grothendieck]

On Sat, Mar 7, 2009 at 3:19 PM, Wacek Kusnierczyk
<Waclaw.Marcin.Kusnierczyk at idi.ntnu.no> wrote:
> Gabor Grothendieck wrote:
>>
>> I've provided an argument against it and no one has provided one
>> for it. The so-called identical code Ivo showed was not identical
>> and, in fact, was flawed.
>
> no, you're wrong.  you think of the part where ivo shows what he'd like
> to have;  the example i was referring to was almost identical with
> yours, except for the explicit return and '=' used for assignment.
>
>
>> Your first/last example could be
>> written:
>>
>> f <- function() letters
>> L <- structure(f()[1:2], names = c("first", "last"))
>>
>
> indeed, but:
>
> - this still does not allow one to use the names directly, only as
> L$first etc., with the syntactic and semantic (longer lookup times) penalty;

That's how it should be done. Using the auto split you get many
variables which is not desirable.  it encourages bad programming.

>
> - using structure you add yet another source of performance penalty; a
> quick naive benchmark hints that it doubles the time elapsed if the
> returned list is inaccessible otherwise, and adds one order of magnitude
> if the list has to be copied:

There is no difference between structure and assignment. They are both
operations.  If there is a timing difference that is a different question.

>
>    f1= function() as.list(letters)
>    f2 =local({ letters = as.list(letters);  function() letters })
>
>    source('http://rbenchmark.googlecode.com/svn/trunk/benchmark.r')
>    benchmark(replications=10000, columns=c('test', 'elapsed'),
>       'f1 direct'=f1(),
>       'f1 structure'=structure(f1(), names=letters),
>       'f2 direct'=f2(),
>       'f2 structure'=structure(f2(), names=letters))
>
>    #           test elapsed
>    # 1    f1 direct   0.171
>    # 2 f1 structure   0.693
>    # 3    f2 direct   0.048
>    # 4 f2 structure   0.594
>
> instead of a syntactically (and semantically, if done appropriately)
> clean solution:
>
>    c(a, b) = f()[1,3]
>    # work with a and b
>
> you offer a glut:
>
>    l = structure(f()[1,3], names=c('a', 'b'))
>    # work with l$a and l$b
>
>
>> or one could define a function to do that without having
>> to modify the language.   Given the relative infrequency
>> of this it hardly seems to merit a language feature.
>>
>
> infrequency of what?  of people's inventing ugly hacks to get arround
> the inability to capture multiple return values directly?  sure, this is
> a good argument against having someone do the job, but is it a good
> argument against having the feature in the language?

I have never had to use it even though I had it available for years.
I do lots of R code so I think it speaks for itself.