My bad, I forgot to call the compileAttributes function... T.T Everything works fine!!! Yes, I don't need object code. "tweak the Makevars", I meant in package P, because now the files are not in src/ anymore, but I moved them to /inst/include. In package Q no need to tweak the Makevars :D
On Tue, May 20, 2014 at 5:41 PM, Dirk Eddelbuettel <e...@debian.org> wrote: > > On 20 May 2014 at 17:26, Alessandro Mammana wrote: > | Dear All, > | first of all thanks again for Rcpp, I am becoming more and more > | dependent on it and I am filling my directories with .cpp scripts... > | > | I have a Rcpp package P using some header file F.hpp in its src/ > | directory. This file F.hpp contains templated functions and structs. > | Is there a way for a new Rcpp package Q to include file F.hpp? > > Yes. > > But you need to move it from src/F.hpp to inst/include/F.hpp which, once > installed, will become include/F.hpp. > > The big trick then is that by just saying "LinkingTo: P" you get R to do the > right thing via the appropriate -I.... flags. > > This is essence all that the package BH does to give everybody Boost headers. > > | I read already Writing R Extensions "Linking to native routines in > | other packages" as well as Rcpp-attriibutes, "Providing a C++ > | Interface". It looks like there are two mechanisms for sharing c++ > | code. > | 1. The "interface" mechanism: the real code is in the src/ directory > | but it is referred to from some headers in the inst/include directory. > | Additionally, these headers need to be wrapped with some (boilerplate > | and obscure to me) C code. This works only with functions. The > | boilerplate code can be nicely generated automatically with Rcpp > | attributes. > > Yes. That is probably the best route __when you need object code__. I am > using that in package RcppXts from (non-Rcpp) package xts, and in the new > RcppRedis from RApiSerialize. > > The exporting package also needs to do some work. See for example > RApiSerialize -- all it does is register two functions. The mechanism is > pure R mechanics and not Rcpp specific but can be used by Rcpp if you want > to. > > | 2. The "direct" mechanism: just put the code in inst/include and tweak > | the Makevars so that the compiler knows where to look for those > | headers. This should work also with structs and templated cpp code. > > See above. No tweaks to Makevars needed (!!) but works only for template > headers. > > | I think I need the "direct" mechanism. > | In package P, I moved my hpp files in inst/include and tweaked the > | Makevars, the package compiled fine. > | Now in package Q I am trying to use some of the stuff, and I am > | including the files I need. I also did the following: > | 1. put import(P) in Q's namespace file > | 2. put Import P and LinkingTo P in Q's description file > | > | however during compilation I got an error like: > | undefined symbol: _Z16rcpp_hello_worldv > > It is a linking error. > > | do you know what could be wrong? > > Are you expecting object code? Then you need to do more. > > You are doing the right thing by working it out with small examples. > > Dirk > > | Thanks a lot and sorry for bothering. > | > | Ale > | > | > | -- > | Alessandro Mammana, PhD Student > | Max Planck Institute for Molecular Genetics > | Ihnestraße 63-73 > | D-14195 Berlin, Germany > | _______________________________________________ > | Rcpp-devel mailing list > | Rcpp-devel@lists.r-forge.r-project.org > | https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/rcpp-devel > -- > Dirk Eddelbuettel | e...@debian.org | http://dirk.eddelbuettel.com -- Alessandro Mammana, PhD Student Max Planck Institute for Molecular Genetics Ihnestraße 63-73 D-14195 Berlin, Germany _______________________________________________ Rcpp-devel mailing list Rcpp-devel@lists.r-forge.r-project.org https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/rcpp-devel
