Unexpected behavior looping through sequence of dates

Thanks for the clarification, Luke.

That is really counter-intuitive behavior. I 100% agree with you that the "for" documentation should state that assumption explicitly.

I would also like to suggest changing the "for" implementation to issue a warning if the "seq" argument is a vector of a non-primitive type. That would have saved me a few hours of debugging in the last few days, so I imagine it must be relevant for more people as well.

--
Alexandre Sieira
CISA, CISSP, ISO 27001 Lead Auditor

"The truth is rarely pure and never simple."
Oscar Wilde, The Importance of Being Earnest, 1895, Act I


Sent with Sparrow (http://www.sparrowmailapp.com/?sig)


On Saturday, 9 de March de 2013 at 13:10, luke-tierney at uiowa.edu wrote:

R's for loop is only designed to iterato over primitive types. The
help file says of the seq argument:

seq: An expression evaluating to a vector (including a list and an
expression) or to a pairlist or ?NULL?. A factor value will
be coerced to a character vector.

[This could be more emphatic by stating that any class attributes are
igonred or something of that nature.]

Having for() do anything else would require designing an iteration
protocol -- probably would be nice in principle but not easy to do.

Best,

luke

On Sat, 9 Mar 2013, Peter Ehlers wrote:

On 2013-03-09 11:14, R. Michael Weylandt wrote:

On Sat, Mar 9, 2013 at 6:50 PM, David Winsemius <dwinsemius at comcast.net (mailto:dwinsemius at comcast.net)>
wrote:

I was unable to find the reason for the original coercion in the
help("for") page or the R
Language Definition entry regarding for-loops. On the hunch that coercion
via as.vector
might be occurring,

Behaviorally, it seems to, but the code for do_for in eval.c has
factors special-cased to call
asCharacterFactor so that might not be a robust detail to rely on. The
relevant behavior seems instead to be that there's a
switch on val_type which creates the loop index but doesn't copy all
attributes (like class)

Note that this means a user's as.vector wouldn't be called here:

as.vector.flub <- function(x, ...) letters[x]

foo <- 1:5
class(foo) <- "flub"

as.vector(foo)

for(j in foo) {print(j); print(class(j))}

as.vector.grub <- function(x, ...) match(x, letters)

bar <- letters[1:5]
class(bar) <- "grub"

as.vector(bar)

for(j in bar) {print(j); print(class(j))}

Cheers,
Michael

I think that Michael is right - the problem is with val_type
in the do_for code.

Here's a simplified version of Alexandre's example:

d <- as.Date("2013-03-10")
for(i in seq_along(d)) print(i)
#[1] 1

for(i in d) print(i)
#[1] 15774
where we might have expected to see "2013-03-10".

The essential line in the do_for code seems to me to be:

val_type = TYPEOF(val);

?typeof tells us that R does not have a 'date' type, so:

typeof(d)
#[1] "double"

And the for-loop results follow.

Peter Ehlers

--
Luke Tierney
Chair, Statistics and Actuarial Science
Ralph E. Wareham Professor of Mathematical Sciences
University of Iowa Phone: 319-335-3386
Department of Statistics and Fax: 319-335-3017
Actuarial Science
241 Schaeffer Hall email: luke-tierney at uiowa.edu (mailto:luke-tierney at uiowa.edu)
Iowa City, IA 52242 WWW: http://www.stat.uiowa.edu