R devel - Nov 2009 - two small wishes (with code sugegstions) for R-core

Dear R developers,

It would be great if you could implement the two minor code changes suggested
below, which would help processing large objects in R.


Jens Oehlschl?gel


# Wish no. 1: let [.AsIs return the class AFTER subsetting, not the class of the
original object
# Wish no. 2: adjust write.csv and write.csv2 for multiple calls in chunked
writing

# Rationale no. 1: a couple of packages will return a different class than
SomeClass when subsetting with [.SomeClass
# and still need to keep the AsIs property
# Examples for classes returning different classes on subscipting are in
packages 'bit', 'ff', 'bigmemory'
# For classes where [.SomeClass will return class SomeClass, such a change will
not hurt

# Code suggestion no. 1: please use 
"[.AsIs" <- function (x, i, ...){
      ret <- NextMethod("[")
      oldClass(ret) <- c("AsIs", oldClass(ret))
      ret
}
# instead of
"[.AsIs" <- function (x, i, ...)
structure(NextMethod("["), class = class(x))


# Rationale no. 2: write.csv and write.csv2 currently enforce that a header must
be written, even with append=TRUE
# This prevents a csv file being written in chunks.
# If argument append=TRUE is used, a header should not be enforced (may be even
be forbidden)

# Code suggestion no. 2: please use
write.csv <-
function (...)
{
    Call <- match.call(write.table, expand.dots = TRUE)
    for (argname in c("col.names", "sep", "dec",
"qmethod")) if (!is.null(Call[[argname]]))
        warning(gettextf("attempt to set '%s' ignored",
argname),
            domain = NA)
    rn <- eval.parent(Call$row.names)
    ap <- eval.parent(Call$append)
    Call$col.names <- if (is.logical(ap) && ap) FALSE else {if
(is.logical(rn) && !rn) TRUE else NA}
    Call$sep <- ","
    Call$dec <- "."
    Call$qmethod <- "double"
    Call[[1L]] <- as.name("write.table")
    eval.parent(Call)
}
write.csv2 <-
function (...)
{
    Call <- match.call(write.table, expand.dots = TRUE)
    for (argname in c("col.names", "sep", "dec",
"qmethod")) if (!is.null(Call[[argname]]))
        warning(gettextf("attempt to set '%s' ignored",
argname),
            domain = NA)
    rn <- eval.parent(Call$row.names)
    ap <- eval.parent(Call$append)
    Call$col.names <- if (is.logical(ap) && ap) FALSE else {if
(is.logical(rn) && !rn) TRUE else NA}
    Call$sep <- ";"
    Call$dec <- ","
    Call$qmethod <- "double"
    Call[[1L]] <- as.name("write.table")
    eval.parent(Call)
}
# instead of
write.csv <- function (...) 
{
    Call <- match.call(expand.dots = TRUE)
    for (argname in c("col.names", "sep", "dec",
"qmethod")) if (!is.null(Call[[argname]]))
        warning(gettextf("attempt to set '%s' ignored",
argname),
            domain = NA)
    rn <- eval.parent(Call$row.names)
    Call$col.names <- if (is.logical(rn) && !rn) 
        TRUE
    else NA
    Call$sep <- ","
    Call$dec <- "."
    Call$qmethod <- "double"
    Call[[1L]] <- as.name("write.table")
    eval.parent(Call)
}
write.csv2 <- 
function (...) 
{
    Call <- match.call(expand.dots = TRUE)
    for (argname in c("col.names", "sep", "dec",
"qmethod")) if (!is.null(Call[[argname]]))
        warning(gettextf("attempt to set '%s' ignored",
argname),
            domain = NA)
    rn <- eval.parent(Call$row.names)
    Call$col.names <- if (is.logical(rn) && !rn) 
        TRUE
    else NA
    Call$sep <- ";"
    Call$dec <- ","
    Call$qmethod <- "double"
    Call[[1L]] <- as.name("write.table")
    eval.parent(Call)
}

I've incorporated a more correct version of 1: think about what 
happens if the next method also copies over the class, for example.
(It was more complicated than that: [.AsIs was used for other classes 
in R -- did you actually test your suggestion, as 'make check' failed 
for me?)

But write.csv[2] are just wrappers, and if you want to do something 
complicated simply don't use them.  They are intended to help naive 
users of .csv files get them right, and banning append=TRUE seems the 
best way to help do that.  (A recent R-help thread shows that 
non-users of Excel are often unaware of its exact requirements -- and 
that includes me.)

The wishes may be small: the work to implement and test them is often 
not.

On Mon, 2 Nov 2009, Jens Oehlschl?gel wrote:
> Dear R developers,
>
> It would be great if you could implement the two minor code changes
suggested below, which would help processing large objects in R.
>
>
> Jens Oehlschl?gel
>
>
> # Wish no. 1: let [.AsIs return the class AFTER subsetting, not the class
of the original object
> # Wish no. 2: adjust write.csv and write.csv2 for multiple calls in chunked
writing
>
> # Rationale no. 1: a couple of packages will return a different class than
SomeClass when subsetting with [.SomeClass
> # and still need to keep the AsIs property
> # Examples for classes returning different classes on subscipting are in
packages 'bit', 'ff', 'bigmemory'
> # For classes where [.SomeClass will return class SomeClass, such a change
will not hurt
>
> # Code suggestion no. 1: please use
> "[.AsIs" <- function (x, i, ...){
>      ret <- NextMethod("[")
>      oldClass(ret) <- c("AsIs", oldClass(ret))
>      ret
> }
> # instead of
> "[.AsIs" <- function (x, i, ...)
> structure(NextMethod("["), class = class(x))
>
>
> # Rationale no. 2: write.csv and write.csv2 currently enforce that a header
must be written, even with append=TRUE
> # This prevents a csv file being written in chunks.
> # If argument append=TRUE is used, a header should not be enforced (may be
even be forbidden)
>
> # Code suggestion no. 2: please use
> write.csv <-
> function (...)
> {
>    Call <- match.call(write.table, expand.dots = TRUE)
>    for (argname in c("col.names", "sep",
"dec", "qmethod")) if (!is.null(Call[[argname]]))
>        warning(gettextf("attempt to set '%s' ignored",
argname),
>            domain = NA)
>    rn <- eval.parent(Call$row.names)
>    ap <- eval.parent(Call$append)
>    Call$col.names <- if (is.logical(ap) && ap) FALSE else {if
(is.logical(rn) && !rn) TRUE else NA}
>    Call$sep <- ","
>    Call$dec <- "."
>    Call$qmethod <- "double"
>    Call[[1L]] <- as.name("write.table")
>    eval.parent(Call)
> }
> write.csv2 <-
> function (...)
> {
>    Call <- match.call(write.table, expand.dots = TRUE)
>    for (argname in c("col.names", "sep",
"dec", "qmethod")) if (!is.null(Call[[argname]]))
>        warning(gettextf("attempt to set '%s' ignored",
argname),
>            domain = NA)
>    rn <- eval.parent(Call$row.names)
>    ap <- eval.parent(Call$append)
>    Call$col.names <- if (is.logical(ap) && ap) FALSE else {if
(is.logical(rn) && !rn) TRUE else NA}
>    Call$sep <- ";"
>    Call$dec <- ","
>    Call$qmethod <- "double"
>    Call[[1L]] <- as.name("write.table")
>    eval.parent(Call)
> }
> # instead of
> write.csv <- function (...)
> {
>    Call <- match.call(expand.dots = TRUE)
>    for (argname in c("col.names", "sep",
"dec", "qmethod")) if (!is.null(Call[[argname]]))
>        warning(gettextf("attempt to set '%s' ignored",
argname),
>            domain = NA)
>    rn <- eval.parent(Call$row.names)
>    Call$col.names <- if (is.logical(rn) && !rn)
>        TRUE
>    else NA
>    Call$sep <- ","
>    Call$dec <- "."
>    Call$qmethod <- "double"
>    Call[[1L]] <- as.name("write.table")
>    eval.parent(Call)
> }
> write.csv2 <-
> function (...)
> {
>    Call <- match.call(expand.dots = TRUE)
>    for (argname in c("col.names", "sep",
"dec", "qmethod")) if (!is.null(Call[[argname]]))
>        warning(gettextf("attempt to set '%s' ignored",
argname),
>            domain = NA)
>    rn <- eval.parent(Call$row.names)
>    Call$col.names <- if (is.logical(rn) && !rn)
>        TRUE
>    else NA
>    Call$sep <- ";"
>    Call$dec <- ","
>    Call$qmethod <- "double"
>    Call[[1L]] <- as.name("write.table")
>    eval.parent(Call)
> }
>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>
-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595