[R] deSolve, unresolved namespace error

[Prof Brian Ripley]

On 06/11/2013 17:20, Adam Clark wrote:
> I'm having trouble running the "ode" function from the "deSolve" package.
>
> I am running RStudio under Ubuntu 13.1
>
> I am running ode() on compiled code that returns delta values using the .C
> convention. While I can include an example of the code, I suspect that it
> will not be helpful, since the problem is not replicable among systems
> (e.g. Solaris or Mac).
>
> When I call ode() on my compiled code, it occasionally will return:
>
> Error in .C("unlock_solver") :
>    "unlock_solver" not resolved from current namespace (deSolve)
>
>
> All subsequent calls to ode() return the same error message, regardless of
> what I run. The problem is resolved only if I restart RStudio. However,
> since I am running many iterations of this command, I would like to find a
> way to resolve this without restarting RStudio.
>
> If I call: is.loaded("unlock_solver", PACKAGE="deSolve"), R returns TRUE.
>
> If I call: .C("unlock_solver"), R returns:
> list()
>
> If I first unload "deSolve" using
> detach("package:deSolve", unload=TRUE), deSolve disappears from my search()
> space, but  is.loaded("unlock_solver", PACKAGE="deSolve") still returns
> TRUE. If I reload deSolve, the problem persists.
>
> Based on what I have read in the help files on namespace conventions in
> packages,  I suspect that the problem is that the .C function
> "unlock_solver" is not corrected loaded, or was unloaded but not marked as
> unloaded.
>
> I have two guesses for what is going on:
> 1) "unlock_solver" was loaded using the library.dynam() function, but
> unloaded using the dyn.unload() function. As I understand it, this would

Who said it was unloaded?  That it is not by default is explicit in ?detach.

> leave a blank entry in the namespace, leading R to think that
> "unlock_solver" is loaded, even though the function no longer does
> anything. However, even when deSolve is working correctly,
> .C("unlock_solver") returns a blank list, so this may not be the case.
>
> 2) Some call deep in deSolve is not pointed towards the right package, and
> therefore cannot find "unlock_solver".
>
>>From the source code for deSolve, posted at
> https://r-forge.r-project.org/scm/viewvc.php/pkg/deSolve/src/deSolve_utils.c?view=log&root=desolve&pathrev=344,
> "unlock_solver"
> seems to be a pretty simple function, inside the deSolve_utils.c file:
>
> void unlock_solver(void) {solver_locked = 0;}
>
> This command is a pretty recent addition to deSolve (it appeared
> somewhere between revision 319 and 324), and is meant to "prevent
> nesting of solvers that have global variables", according to the
> change annotation.
>
>
> In any case, I'd like to find a way to specifically unload "unlock_solver"
> and reload it, or barring that, unload as many of the DLL's associated with
> deSolve as possible and reload them. I suspect that this will solve my
> problems.

I am not sure why you think it is reasonable to do that.  Clearly the 
designers of deSolve did not think so, as it does not have an .onUnload 
action.

 > library(deSolve)
 > names(getLoadedDLLs())
[1] "base"      "utils"     "methods"   "grDevices" "graphics"  "stats"
[7] "deSolve"   "tools"
 > detach("package:deSolve", unload=TRUE)
 > names(getLoadedDLLs())
[1] "base"      "utils"     "methods"   "grDevices" "graphics"  "stats"
[7] "deSolve"   "tools"

> Many thanks, and sorry if this is a silly question,

It is really an R-devel question: see the posting guide.  In particular 
OSes differ in how (or even if) they can reload DLLs, and the details 
are way too technical for R-help.

In any case, there are no reproduction instructions here: see the 
posting guide.

> Adam
>


-- 
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