[R] Maximum # of DLLs reached, or, how to clean up after yourself?

[Henrik Bengtsson]

As Jeff says, I think the common use case is to run/rerun in fresh R sessions.

But, yes, if you'd like to have each script clean up after itself,
then you need to check with pkgs0 <- loadedNamespaces() to see what
packages are loaded when the script starts (not just attached) and
then unload the ones added at the end by pkgsDiff <-
setdiff(loadedNamespaces(), pkgs0).   However, it's not as simple as
calling unloadNamespace(pkgsDiff), because they need to be unloaded in
an order that is compatible with the package dependencies.   One way
is to too use while(length(pkgDiffs) > 0) loop over with a
try(unloadNamespace(pkg)) until all are unloaded.   At the end, run
R.utils::gcDLLs() too (now on CRAN).

unloadNamespace("foo") should result in the same as
detach("package::foo", unload=TRUE) [anyone correct me if I'm wrong].

Hope this helps

Henrik

On Wed, Sep 14, 2016 at 6:41 AM, Jeff Newmiller
<jdnewmil at dcn.davis.ca.us> wrote:
> I never detach packages. I rarely load more than 6 or 7 packages directly before restarting R. I frequently re-run my scripts in new R sessions to confirm reproducibility.
> --
> Sent from my phone. Please excuse my brevity.
>
> On September 14, 2016 1:49:55 AM PDT, Alexander Shenkin <ashenkin at ufl.edu> wrote:
>>Hi Henrik,
>>
>>Thanks for your reply.  I didn't realize that floating DLLs were an
>>issue (good to know).  My query is actually a bit more basic.  I'm
>>actually wondering how folks manage their loading and unloading of
>>packages when calling scripts within scripts.
>>
>>Quick example:
>>Script1:
>>       library(package1)
>>       source("script2.r")
>>       # do stuff reliant on package1
>>       detach("package:package1", unload=TRUE)
>>
>>Script2:
>>       library(package1)
>>       library(package2)
>>       # do stuff reliant on package1 and package2
>>       detach("package:package1", unload=TRUE)
>>       detach("package:package2", unload=TRUE)
>>
>>Script2 breaks Script1 by unloading package1 (though unloading package2
>>
>>is ok).  I will have to test whether each package is loaded in Script2
>>before loading it, and use that list when unloading at the end of the
>>Script2.  *Unless there's a better way to do it* (which is my question
>>-
>>is there?).  I'm possibly just pushing the whole procedural scripting
>>thing too far, but I also think that this likely isn't uncommon in R.
>>
>>Any thoughts greatly appreciated!
>>
>>Thanks,
>>Allie
>>
>>On 9/13/2016 7:23 PM, Henrik Bengtsson wrote:
>>> In R.utils (>= 2.4.0), which I hope to submitted to CRAN today or
>>> tomorrow, you can simply call:
>>>
>>>    R.utils::gcDLLs()
>>>
>>> It will look at base::getLoadedDLLs() and its content and compare to
>>> loadedNamespaces() and unregister any "stray" DLLs that remain after
>>> corresponding packages have been unloaded.
>>>
>>> Until the new version is on CRAN, you can install it via
>>>
>>>
>>source("http://callr.org/install#HenrikBengtsson/R.utils@develop")
>>>
>>> or alternatively just source() the source file:
>>>
>>>
>>source("https://raw.githubusercontent.com/HenrikBengtsson/R.utils/develop/R/gcDLLs.R")
>>>
>>>
>>> DISCUSSION:
>>> (this might be better suited for R-devel; feel free to move it there)
>>>
>>> As far as I understand the problem, running into this error / limit
>>is
>>> _not_ the fault of the user.  Instead, I'd argue that it is the
>>> responsibility of package developers to make sure to unregister any
>>> registered DLLs of theirs when the package is unloaded.  A developer
>>> can do this by adding the following to their package:
>>>
>>> .onUnload <- function(libpath) {
>>>     library.dynam.unload(utils::packageName(), libpath)
>>>  }
>>>
>>> That should be all - then the DLL will be unloaded as soon as the
>>> package is unloaded.
>>>
>>> I would like to suggest that 'R CMD check' would include a check that
>>> asserts when a package is unloaded it does not leave any registered
>>> DLLs behind, e.g.
>>>
>>> * checking whether the namespace can be unloaded cleanly ... WARNING
>>>   Unloading the namespace does not unload DLL
>>> * checking loading without being on the library search path ... OK
>>>
>>> For further details on my thoughts on this, see
>>> https://github.com/HenrikBengtsson/Wishlist-for-R/issues/29.
>>>
>>> Hope this helps
>>>
>>> Henrik
>>>
>>> On Tue, Sep 13, 2016 at 6:05 AM, Alexander Shenkin <ashenkin at ufl.edu>
>>wrote:
>>>> Hello all,
>>>>
>>>> I have a number of analyses that call bunches of sub-scripts, and in
>>the
>>>> end, I get the "maximal number of DLLs reached" error.  This has
>>been asked
>>>> before (e.g.
>>>>
>>http://stackoverflow.com/questions/36974206/r-maximal-number-of-dlls-reached),
>>>> and the general answer is, "just clean up after yourself".
>>>>
>>>> Assuming there are no plans to raise this 100-DLL limit in the near
>>future,
>>>> my question becomes, what is best practice for cleaning up
>>(detaching)
>>>> loaded packages in scripts, when those scripts are sometimes called
>>from
>>>> other scripts?  One can detach all packages at the end of a script
>>that were
>>>> loaded at the beginning of the script.  However, if a package is
>>required in
>>>> a calling script, one should really make sure it hadn't been loaded
>>prior to
>>>> sub-script invocation before detaching it.
>>>>
>>>> I could write a custom function that pushes and pops package names
>>from a
>>>> global list, in order to keep track, but maybe there's a better way
>>out
>>>> there...
>>>>
>>>> Thanks for any thoughts.
>>>>
>>>> Allie
>>>>
>>>> ______________________________________________
>>>> R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
>>>> https://stat.ethz.ch/mailman/listinfo/r-help
>>>> PLEASE do read the posting guide
>>http://www.R-project.org/posting-guide.html
>>>> and provide commented, minimal, self-contained, reproducible code.
>>
>>______________________________________________
>>R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
>>https://stat.ethz.ch/mailman/listinfo/r-help
>>PLEASE do read the posting guide
>>http://www.R-project.org/posting-guide.html
>>and provide commented, minimal, self-contained, reproducible code.
>