returning vectors of unknown size from C (with .C)

Dear All,

In a package, I am using ".C" to call some C functions. In one case,
the number of elements of the return vectors are not known in R before
the C call. (Two of the vectors are integers, the third is vector of
character strings).

Passing from R a vector of the maximum possible size would be a huge
waste. I understand one alternative is to use ".Call", but I'd rather
avoid it if I can (all of the code seems working except for the return
of values into R). Another would be to write to a file from C and then
read that into R, but this looks very ugly. Are there any other
reasonable alternatives, or should I just use .Call?
  

Thanks,

R.

Ramon Diaz-Uriarte wrote:

Dear All,

In a package, I am using ".C" to call some C functions. In one case,
the number of elements of the return vectors are not known in R before
the C call. (Two of the vectors are integers, the third is vector of
character strings).

Passing from R a vector of the maximum possible size would be a huge
waste. I understand one alternative is to use ".Call", but I'd rather
avoid it if I can (all of the code seems working except for the return
of values into R). Another would be to write to a file from C and then
read that into R, but this looks very ugly. Are there any other
reasonable alternatives, or should I just use .Call?

 .Call is usually easiest, but another possibility is to have two entry
points:  one to calculate how much space you need, a second to pass in a
vector that's the right size to hold the result.

 Duncan Murdoch

Thanks,

R.

On Sat, Apr 26, 2008 at 3:19 AM, Duncan Murdoch <murdoch at stats.uwo.ca> wrote:

Ramon Diaz-Uriarte wrote:

Dear All,

In a package, I am using ".C" to call some C functions. In one case,
the number of elements of the return vectors are not known in R before
the C call. (Two of the vectors are integers, the third is vector of
character strings).

Passing from R a vector of the maximum possible size would be a huge
waste. I understand one alternative is to use ".Call", but I'd rather
avoid it if I can (all of the code seems working except for the return
of values into R). Another would be to write to a file from C and then
read that into R, but this looks very ugly. Are there any other
reasonable alternatives, or should I just use .Call?

 .Call is usually easiest, but another possibility is to have two entry
points:  one to calculate how much space you need, a second to pass in a
vector that's the right size to hold the result.

You mean making two successive calls to the C code? The problem is
that the size of the result is not known until the result is obtained
(in my C code, the underlying structure is a linked list that gets
stretched as needed as the computation proceeds). So I would not know
"where to leave the result from C" in between the two calls to C.

On Sat, 26 Apr 2008, Ramon Diaz-Uriarte wrote:

On Sat, Apr 26, 2008 at 3:19 AM, Duncan Murdoch <murdoch at stats.uwo.ca>

wrote:

Ramon Diaz-Uriarte wrote:

Dear All,

In a package, I am using ".C" to call some C functions. In one case,
the number of elements of the return vectors are not known in R before
the C call. (Two of the vectors are integers, the third is vector of
character strings).

Passing from R a vector of the maximum possible size would be a huge
waste. I understand one alternative is to use ".Call", but I'd rather
avoid it if I can (all of the code seems working except for the return
of values into R). Another would be to write to a file from C and then
read that into R, but this looks very ugly. Are there any other
reasonable alternatives, or should I just use .Call?

 .Call is usually easiest, but another possibility is to have two entry
points:  one to calculate how much space you need, a second to pass in a
vector that's the right size to hold the result.

You mean making two successive calls to the C code? The problem is
that the size of the result is not known until the result is obtained
(in my C code, the underlying structure is a linked list that gets
stretched as needed as the computation proceeds). So I would not know
"where to leave the result from C" in between the two calls to C.

 But that is possible (you malloc the memory for a local copy in the rist
call), and rpart does something like it.

 --
 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

 But that is possible (you malloc the memory for a local copy in the rist
call), and rpart does something like it.

Aha, thanks, I didn't know it was doable (or easy). I'll look at the
rpart code. One further question, though, what is "the rist call"?