R devel - May 2008 - rawToChar(raw(0))

Hi,

right now we have (on R v2.7.0 patched (2008-04-23 r45466)) that:
> rawToChar(raw(0))
[1] ""
> rawToChar(raw(0), multiple=TRUE)
character(0)

Is this intended or should both return character(0)?  Personally, I
would prefer that an empty input vector returns an empty output
vector.  Same should then apply to charToRaw(), but right now we get:
> x <- character(0)
> charToRaw(x)
Error in charToRaw(x) : argument must be a character vector of length 1

I am aware of the new rules for R v2.8.0dev on truncating nuls in
character strings.  That would make charToRaw(x) return the same
regardless of x==character(0) or x=="" ("\0").  We have with
R
v2.8.0dev:
> x <- ""
> charToRaw(x)
raw(0)

Is guess this is on purpose because "" == "<nul>" and
"a" == "a<nul>",
but what would happen if one use the special rule that "<nul>"
returns
as.raw(0) instead?  Would that break much downstream?

If one have it return as.raw(0), and add the rule that an empty input
returns an empty output, then charToRaw(rawToChar(x)) [and vice versa
rawToChar(charToRaw(x))] would always return 'x'.  With R v2.7.0 we
now have:
> x <- raw(0)
> identical(x, charToRaw(rawToChar(x)))
[1] TRUE
> x <- as.raw(0)
> identical(x, charToRaw(rawToChar(x)))
[1] TRUE

but the latter will be FALSE in R v2.8.0 (because of the new rules on
'\0'):
> x <- as.raw(0)
> identical(x, charToRaw(rawToChar(x)))
[1] FALSE
Warning message:
In rawToChar(x) : truncating string with embedded nul: '\0'


Just some thoughts.

Cheers

Henrik

On 5/21/08, Henrik Bengtsson <hb at stat.berkeley.edu>
wrote:
> Hi,
>
>  right now we have (on R v2.7.0 patched (2008-04-23 r45466)) that:
>
>  > rawToChar(raw(0))
>  [1] ""
>  > rawToChar(raw(0), multiple=TRUE)
>  character(0)
>
>  Is this intended or should both return character(0)?  Personally, I
>  would prefer that an empty input vector returns an empty output
>  vector.
I don't see why; rawToChar(, multiple=FALSE) is meant to be a map from
an n-vector to a scalar. Why should there be an exception when n==0?
Would you expect

mean(numeric(0))

to return numeric(0)?
> Same should then apply to charToRaw(), but right now we get:
>
>  > x <- character(0)
>  > charToRaw(x)
>  Error in charToRaw(x) : argument must be a character vector of length 1
The error message says it all: charToRaw() maps a scalar to an
n-vector. For vectors of length > 1, it uses the first element with a
warning.

No comments on the rest.

-Deepayan
>  [...]