%dorng% does not work unless the 'foreach' package is attached

[HenrikBengtsson]

%dorng% does not work unless the 'foreach' package is attached. This becomes a problem for packages who want to use 'foreach' and 'doRNG' internally without attaching those packages. The workaround for such packages is to also import %dopar% from 'foreach' even if they never use it.

A minimal example:

`%dorng%` <- doRNG::`%dorng%`
y <- foreach::foreach(x=1:2) %dorng% x
## Error in list(args = (1:2)(.doRNG.stream = list(c(10407L, -255388091L,  : 
##  could not find function "%dopar%"

A similar do.call() problem was reported in Issue #13, which was fixed.

[HenrikBengtsson]

I've created PR #19 that fixes this problem. Proof of concept afterward:

> `%dorng%` <- doRNG::`%dorng%`
> y <- foreach::foreach(x=1:2) %dorng% x
Loading required package: foreach
Loading required package: rngtools
Warning message:
executing %dopar% sequentially: no parallel backend registered 
> str(y)
List of 2
 $ : int 1
 $ : int 2
 - attr(*, "rng")=List of 2
  ..$ : int [1:7] 10407 812739260 -134609779 1241370090 -241383837 922805320 -74025559
  ..$ : int [1:7] 10407 -237543692 -1522949811 894917114 315123266 1852579936 210565358
 - attr(*, "doRNG_version")= chr "1.7.4"

[HenrikBengtsson]

Move #19 (comment) here:

@renozao wrote:

Wondering why this was not working:

• foreach is in Depends

Packages under 'Depends' are only loaded when loading your doRNG package. However, they are attached if one attaches doRNG, e.g.

> ## Loading only
> loadNamespace("doRNG")
<environment: namespace:doRNG>
> search()
[1] ".GlobalEnv"        "package:stats"     "package:graphics" 
[4] "package:grDevices" "package:utils"     "package:datasets" 
[7] "package:methods"   "Autoloads"         "package:base"     

> ## Attaching
> library(doRNG)
Loading required package: foreach
Loading required package: rngtools
> search()
 [1] ".GlobalEnv"        "package:doRNG"     "package:rngtools" 
 [4] "package:foreach"   "package:stats"     "package:graphics" 
 [7] "package:grDevices" "package:utils"     "package:datasets" 
[10] "package:methods"   "Autoloads"         "package:base"     
> 

• I import the whole package in the NAMESPACE with import(foreach)

Is it that calling do.call() with a string lookup for the function in the incorrect place?

When using:

    res <- do.call("%dopar%", list(obj, ex), envir = parent.frame())

do.call() will have to first locate "%dopar%", which, although the help leaves room for interpretation, I guess does by using something like get("%dopar%", mode = "function", envir = envir) internally. Since envir = parent.frame(), it's searched from the environment where %dorng% was called and not from inside you packages - they are two different search paths. I think that if you would have done:

    fcn <- get("%dopar%", mode = "function")  # search using the search path that 'doRNG' sees
    res <- do.call(fcn, list(obj, ex), envir = parent.frame())

it would have worked. But, yeah, it's basically always better to use the function object rather than its name when using do.call(), so:

    res <- do.call(foreach:`%dopar%`, list(obj, ex), envir = parent.frame())

or

#' @importFrom foreach %dopar%
...
    res <- do.call(`%dopar%`, list(obj, ex), envir = parent.frame())

[markopalangetic]

The same issue here. Is there any progress on this topic?