[Rcpp-devel] check validity of object created from Rcpp module

[John Chambers]

On 6/27/12 7:07 AM, Douglas Bates wrote:
> On Tue, Jun 26, 2012 at 7:04 PM, Dirk Eddelbuettel <edd at debian.org> wrote:
>> On 26 June 2012 at 17:21, Jiqiang Guo wrote:
>> | Dear List,
>> |
>> | I am wondering if there is a way to check whether an object created from Rcpp's
>> | module is valid or not especially for objects loaded in another session.
>> |
>> | In detail, we know that we could not save and load objects created from Rcpp's
>> | module [section 5 of http://cran.r-project.org/web/packages/Rcpp/vignettes/
>> | Rcpp-modules.pdf].  But that does not prevent users doing that. So if an
>> | object created from Rcpp's module is loaded in another session, we would end up
>> | with errors or even segfault.  It would be nice we have a method to check if
>> | it is valid or not so that when we are writing a package, we could do a check
>> | before calling those methods of that object.
>>
>> It's a very good idea. I don't have a good idea as to how we'd do this though.
> There has to be one or more external pointers in the R object
> somewhere (although I spent some time looking at the Rcpp-modules code
> and came away very confused so I can't tell you exactly where.)  That
> pointer will become a NULL pointer after a serialize/unserialize
> sequence so you need to check for its being NULL.  This is why the
> reference classes in the development version of lme4 have a field
> named ptr and a method named Ptr.  The Ptr method first checks if ptr
> is a null pointer and, if so, regenerates the object then returns ptr.
>   I can say from experience that this all gets rather complicated.
> However I haven't looked at John Chambers' new code for Rcpp classes
> in R and he might have addressed the issue.
Well, no, and it doesn't seem particularly a modules issue, but rather a 
general point in using C++ data in R.  Since C++ deals with run-time 
object pointers, there is always the likelihood that an arbitrary 
reference to such objects will fail to serialize.

What one could easily write is a recursive check on an object that looks 
for any elements, attributes or objects in environments that are 
external pointer types with NULL values.  That would be a good thing to 
have--maybe someone did this as part of general QC software?

Given that, we could perhaps embed this in code that unserializes class 
"C++Object".  Of course, one can check but clearly there is no universal 
repair method.

John
>
>> Dirk
>>
>> | An illustrative example:
>> |
>> | require("Rcpp")
>> | require("inline")
>> |
>> | inc <- '
>> | class Foo {
>> | private:
>> |   int x_;
>> | public:
>> |   Foo(int x) : x_(x) { }
>> |   int getx() const { return x_;}
>> | };
>> |
>> | RCPP_MODULE(foo) {
>> |   class_<Foo>("Foo")
>> |   .constructor<int>()
>> |   .method("getx", &Foo::getx)
>> |   ;
>> | }
>> |
>> | fx <- cxxfunction(signature(), "", include = inc, plugin = "Rcpp", verbose =
>> | FALSE)
>> | m <- Module("foo", getDynLib(fx))
>> | cppo  = new(m$Foo, 5)
>> |
>> | print(cppo$getx())
>> | save("cppo", file = 'cppo.RData')
>> |
>> | rm(list = ls())
>> | load("cppo.RData")
>> | print(cppo$getx())
>> |
>> | For the above code, we know the last line would not work.  If we have a
>> | function to test the validity of cppo,then if not valid, we would not call
>> | cppo's function and are able to tell users that cppo is not valid any more.
>> |
>> | Best,
>> | Jiqiang
>> |
>> | ----------------------------------------------------------------------
>> | _______________________________________________
>> | Rcpp-devel mailing list
>> | Rcpp-devel at lists.r-forge.r-project.org
>> | https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/rcpp-devel
>> --
>> Dirk Eddelbuettel | edd at debian.org | http://dirk.eddelbuettel.com
>> _______________________________________________
>> Rcpp-devel mailing list
>> Rcpp-devel at lists.r-forge.r-project.org
>> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/rcpp-devel
> _______________________________________________
> Rcpp-devel mailing list
> Rcpp-devel at lists.r-forge.r-project.org
> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/rcpp-devel
>