[Rcpp-devel] Interfacing Rcpp with a C library: memory allocation

Alessandro Mammana mammana at molgen.mpg.de
Thu Nov 21 14:37:20 CET 2013


I found out what the problem was, luckily it was just a stupid bug in
my code (an array index out of bounds).
I still didn't quite get how memory allocation works. As far as I
understood there are these two ways of allocating memory:

1. Using malloc/free, new/delete, normal C++ constructors and
destructors: this memory area is completely separated from R memory
area and it is erased at the end of the Rcpp call. Pointers to these
addresses will not be valid in a second Rcpp call.
2. Using Rcpp constructors/destructors, Calloc() and Free(): this is
the same memory used by R and it is suitable for passing objects in
and out.

Is that right?
Thanks again,
Ale


On Wed, Nov 20, 2013 at 7:25 PM, Dirk Eddelbuettel <edd at debian.org> wrote:
>
> On 20 November 2013 at 18:54, Alessandro Mammana wrote:
> | Thanks a lot for your help, I'll look at the references. I saw Section
> | 6.1.2 already. Does that mean that I should replace  all the calls to
> | malloc() and to free() with those to Calloc() and Free() in the
> | external C code (maybe with a macro?)? Or does it just mean that I
> | would get the same memory errors, but this way they are handled by R?
>
> The idea is not to get errors.
>
> Let's assume that the library you use is in itself correct.  You could then
> use it and write wrapper functions. In the wrapper functions, you create Rcpp
> type and assign to those. That would avoid having to alter the library. A
> good defensive approach.
>
> Depending on how well you know what you are doing, you can do something like
> too where we assume we have  someFunction(double* p):
>
>     int n;
>
>     Rcpp::NumericVector x(n);      // vector of n element
>
>     someFunction(x.begin());       // someFunction sees a double*
>
> which is how I would interface a C library.  The devil is in the detail, and
> you still have not provide any so we can't help you much more than via
> generalities.
>
> Pick one function from the library. Interface it. Get it work without a
> segfault.  Apply what you learned on another function etc pp.
>
> There are other approaches (Rcpp modules is one) but you may to make sure
> you clear a few conceptual hurdles first.
>
> Dirk
>
>
>
> |
> | Thanks a lot!
> | Ale
> |
> | On Wed, Nov 20, 2013 at 6:35 PM, Dirk Eddelbuettel <edd at debian.org> wrote:
> | >
> | > On 20 November 2013 at 18:27, Alessandro Mammana wrote:
> | > | Dear all,
> | > | I'm trying to write some efficient code for analyzing sequencing data
> | > | in R. To do this I would like to use the C library samtools. I've
> | > | created a package where the src directory looks like this:
> | > |
> | > | src
> | > | |-- Makevars
> | > | |-- RcppExports.cpp
> | > | |-- mysourcecode.cpp
> | > | `-- samtools
> | > |     |-- all *.c and *.h files as well as an independent Makefile
> | > |
> | > | My Makevars file looks like this:
> | > |
> | > | PKG_CPPFLAGS = -Isamtools
> | > | PKG_LIBS = `$(R_HOME)/bin/Rscript -e "Rcpp:::LdFlags()"`
> | > | PKG_LIBS += -Lsamtools  -lbam -lz -lpthread
> | > |
> | > | $(SHLIB): samtools/libbam.a
> | > |
> | > | samtools/libbam.a:
> | > |       @(cd samtools-0.1.19 && $(MAKE) libbam.a \
> | > |           CC="$(CC)" CFLAGS="$(CFLAGS) $(CPICFLAGS)" AR="$(AR)"
> | > | RANLIB="$(RANLIB)")
> | > |
> | > | Everything compiles and I get my cpp functions in R, however I am
> | > | getting some weird segfaults, I think they are due to memory
> | > | allocation, but it's hard for me to track them. Especially now, these
> | > | errors are showing up not immediately, but at the second time that I
> | > | call a Rcpp function.
> | > |
> | > | I wanted to ask the following:
> | > | 1. Is it the right way of using external C libraries? I couldn't find
> | > | much documentation around
> | >
> | > Sure.
> | >
> | > | 2. The C library uses malloc and free, and so do I (as little as
> | > | possible, just to interface with the library), is this mechanism
> | > | clashing against Rcpp/R memory management? Could it happen, for
> | > | instance, that R tries to free allocated memory that I already
> | > | manually freed myself?
> | >
> | > Yes.
> | >
> | > Please read the "Writing R Extensions" manual, section 6.1.2, on
> | > User-controlled memory.
> | >
> | > And/or see the Rcpp documentation: if you actually use our data containers,
> | > things "just work", ie
> | >
> | >   R> cppFunction("NumericMatrix foo(int n) { return NumericMatrix(n); }")
> | >   R> foo(2)
> | >        [,1] [,2]
> | >   [1,]    0    0
> | >   [2,]    0    0
> | >   R> foo(4)
> | >        [,1] [,2] [,3] [,4]
> | >   [1,]    0    0    0    0
> | >   [2,]    0    0    0    0
> | >   [3,]    0    0    0    0
> | >   [4,]    0    0    0    0
> | >   R>
> | >
> | > will never create a segfault. Ditto for using proper C++ containers (eg
> | > std::vector<double>).
> | >
> | > But you cannot randomly match the C approach (see Section 6.1.2, as mentioned
> | > above) and the R/Rcpp approach.
> | >
> | > | In general I didn't understand much about memory allocation in Rcpp
> | > | and I couldn't find many resources talking about it. Is there anything
> | >
> | > Well I could recommend a book to you...  There are also eight vignettes in
> | > the package, and the more on other sites (the list archive, our blogs,
> | > StackOverlow, Hadley's draft book).
> | >
> | > | R- or Rcpp-specific that I have to keep in mind or should I program as
> | > | if I were programming in C/C++?
> | >
> | > You can provided you do it right.
> | >
> | > Hope this helps, if it was unclear please come back with more questions.
> | > Also see the unit tests and examples.
> | >
> | > Cheers, Dirk
> | >
> | > --
> | > Dirk Eddelbuettel | edd at debian.org | http://dirk.eddelbuettel.com
> |
> |
> |
> | --
> | Alessandro Mammana, PhD Student
> | Max Planck Institute for Molecular Genetics
> | Ihnestraße 63-73
> | D-14195 Berlin, Germany
>
> --
> Dirk Eddelbuettel | edd at debian.org | http://dirk.eddelbuettel.com



-- 
Alessandro Mammana, PhD Student
Max Planck Institute for Molecular Genetics
Ihnestraße 63-73
D-14195 Berlin, Germany


More information about the Rcpp-devel mailing list