[R] R_unload_[PACKAGENAME] not called

Henrik Alsing Friberg henrik.alsing.friberg at mosek.com
Wed Mar 28 16:25:48 CEST 2012


I experimented with this earlier, but it is indeed tricky business.
Last I tried my registered finalizer addressed unmapped memory,
because garbage collection happened after my package (and thus the
clean-up stuff) was unloaded. This happened using the
R_RegisterFinalizerEx function from C++ on some dummy SEXP object.

Since cholmod_finish is an exported function from the Matrix package,
it has to be called before the Matrix package is unloaded. I will try
something like
  >> reg.finalizer(environment(Matrix), fun, onexit=TRUE)

Thanks.


On Mon, Mar 26, 2012 at 4:21 PM, Duncan Murdoch
<murdoch.duncan at gmail.com> wrote:
> On 26/03/2012 6:24 AM, Henrik Alsing Friberg wrote:
>>
>> Hi R-help users..
>>
>> I mistakenly posted a now rejected bug report
>> (https://bugs.r-project.org/bugzilla3/show_bug.cgi?id=14855), but the
>> subject still bugs me..!
>>
>> If I have a package called MyPackage with C code inside, then library
>> or require(MyPackage) will call loadNamespace() which will look for
>> and call:
>> void R_init_MyPackage(DllInfo *info)
>>
>> At the end of the R session, the corresponding unload function is
>> however *not* called.
>> void R_unload_MyPackage(DllInfo *info)
>>
>> I use the two functions to call R_cholmod_start(CHM_CM) and
>> cholmod_finish(CHM_CM) respectively, and I think this might leak
>> memory if the latter is not called. Firstly, why is one called without
>> the package maker having to think about it, while the other is not?
>> Secondly, how do I make sure that the unload function is called --
>> using either .Last.lib or .onUnload to call unloadNamespace() seems
>> like a bad idea, as unloadNamespace() is documented to call back to
>> these hooks (the call cycle would have to by broken by other means)?
>
>
> You would normally use reg.finalizer() for this, or the C level equivalents
> (see the section on external pointers in Writing R Extensions).
>
> Duncan Murdoch



More information about the R-help mailing list