[R-sig-dyn-mod] Using dynamically compiled code for use with deSolve in a new package

Dear Thomas,

Thanks again for this reply. While this may work locally and for some
users, it does not work in my case where my namespace file is generated
by roxygen2. Also, CRAN does not like dll-file to be in the package
(inst folder), so it needs to be built at installation. How do I do
this? Simply including the Fortran code in the src directory does not
work (gives error 127, whatever that is). Any suggestions?

Cheers, Rampal


On 11-6-2017 21:15, Thomas Petzoldt wrote:

Hi,

that's straightforward.

1) Put deSolve in Depends of your package DESCRIPTION:

Depends: deSolve


2) Import the solver function (e.g. ode or lsoda) in the NAMESPACE of
your package and declare your dynamic library (.dll or .so)

importFrom(deSolve,ode)

useDynLib(mypackage)


3) and then use this .dll/.so in the call to ode():

ode(...., dllname = "mypackage", .....)


An example can be found in package growthrates:

https://github.com/tpetzoldt/growthrates/blob/master/pkg/R/grow_twostep.R

Hope it helps,

Thomas



On 11.06.2017 13:20, Rampal Etienne wrote:

Dear all,

I have written C and FORTRAN functions to use with deSolve and found
that it works significantly faster than the analogous R functions. So
far so good! However, now I want to build a package that uses this code.
How do I do this? I can include the DLL in the package (so no more
compilation by the user of the package, just loading it), but then the
user needs to be able to find it. Where will this be installed? That may
be different on different systems. Any suggestions?

Cheers, Rampal

Hi,

including Fortran and/or C sources in /src is exactly what we (and the
other CRAN authors) do. It is self-evident, that binary code (like
dlls) cannot be included in portable packages, that are expected to
run on different operating systems. Details are found in "Writing R
Extensions".

I don't see a problem with roxygen2, this is exactly how it was
implemented in package "growthrates", mentioned earlier. Let's assume
we have a package called "foo", add the following roxygen-directives
to a file "R/foo-package.R":

#' @useDynLib foo
#'
#' @importFrom deSolve ode

Hope it helps,

Thomas


On 29.11.2017 14:44, Rampal S. Etienne wrote:

Dear Thomas,

Thanks again for this reply. While this may work locally and for some
users, it does not work in my case where my namespace file is generated
by roxygen2. Also, CRAN does not like dll-file to be in the package
(inst folder), so it needs to be built at installation. How do I do
this? Simply including the Fortran code in the src directory does not
work (gives error 127, whatever that is). Any suggestions?

Cheers, Rampal


On 11-6-2017 21:15, Thomas Petzoldt wrote:

Hi,

that's straightforward.

1) Put deSolve in Depends of your package DESCRIPTION:

Depends: deSolve


2) Import the solver function (e.g. ode or lsoda) in the NAMESPACE of
your package and declare your dynamic library (.dll or .so)

importFrom(deSolve,ode)

useDynLib(mypackage)


3) and then use this .dll/.so in the call to ode():

ode(...., dllname = "mypackage", .....)


An example can be found in package growthrates:

https://github.com/tpetzoldt/growthrates/blob/master/pkg/R/grow_twostep.R


Hope it helps,

Thomas



On 11.06.2017 13:20, Rampal Etienne wrote:

Dear all,

I have written C and FORTRAN functions to use with deSolve and found
that it works significantly faster than the analogous R functions. So
far so good! However, now I want to build a package that uses this
code.
How do I do this? I can include the DLL in the package (so no more
compilation by the user of the package, just loading it), but then the
user needs to be able to find it. Where will this be installed?
That may
be different on different systems. Any suggestions?

Cheers, Rampal