[Rcpp-devel] Rcpp and VC++ (was: How get colnames and rownames in Rcpp method?)

[Romain Francois]

Le 31/07/10 22:04, Dominick Samperi a écrit :
> On Wed, Jul 28, 2010 at 10:11 PM, Dirk Eddelbuettel <edd at debian.org
> <mailto:edd at debian.org>> wrote:
>
>     ... you missed entry 2.7 in the FAQ which I quote here with the
>     mark-up; see
>     the pdf version on my site or the nearest CRAN mirror:
>
>        \subsection{Can I use \pkg{Rcpp} with Visual Studio}
>
>        Not a chance.
>
>        And that is not because we are meanies but because \proglang{R}
>     and Visual
>        Studio simply do not get along. As \pkg{Rcpp} is all about extending
>        \proglang{R} with \proglang{C++} interfaces, we are bound by the
>     available
>        toolchain.  And \proglang{R} simply does not compile with Visual
>        Studio. Go complain to its vendor if you are still upset.
>
>     So in short: no VC++.
>
>
> It is possible to build Rcpp using VC++, but it is not easy, and getting the
> DLL's to work properly is difficult. Furthermore, VC++ is not supported by
> CRAN so packages that depend on VC++ will have to be built using the
> MinGW tool chain before they can be released to CRAN. It should also
> be noted that VC++ Express does not officially support 64bit builds (though
> this is possible with some work, at least with the 2008 version).
>
> That said, it can be helpful to test using different compilers because the
> GNU compilers tend to be lenient (though the latest versions are advertised
> to follow the standards more closely).
>
> For example, here is an issue that was picked up by the VC++ compiler but
> not by other compilers. On Line 171 of XPtr.h there is a template
> function definition
> where one of the parameters is given a default value, but two trailing
> parameters
> are not.

This was definitely an omission, thanks for picking it up.

I've set "prot" and "tag" to R_NilValue in the svn version, as this was 
the intention.

> The C++ standard says if one parameter is given a default
> value, then all
> parameters after this one must also be given default values. I don't
> know of any
> rule that says this does not apply to template functions, maybe there is
> one,
> in which case this is a problem with VC++.
>
> The easy fix is to set the default value R_NilValue for the last two
> parameters.
> This results in errors due to ambiguous calls that end up calling the wrong
> constructor. This occurs on Line 240 of Module.cpp, and on Line 200 of
> RcppCommon.cpp. A simple work-around is to replace
> clxp(cl) with clxp(cl,true) in the first case, and p(v) with p(v,true) in
> the second. I have not checked the logic to see if this might cause other
> problems.
>
> Besides this issue, the changes needed (in Rcpp 0.8.5)
> to compile using VC++ are as follows:
>
> 1. Date.cpp: replace include of <unistd.h> with:
>
> //#include <unistd.h>        // for read and close on Solaris
> #ifdef _MSC_VER
> // POSIX open,read,write deprecated as of VC++ 2005.
> // Use ISO conformant _open,_read,_write instead.
> #include <io.h>
> #define open _open
> #define write _write
> #define close _close
> #define read _read
> #define snprintf _snprintf
> #else
> #include <unistd.h> // for Solaris, e.g.
> #endif
>
> 2. RcppDatetime.cpp: redefine snprintf as follows for MSVC:
>
> #ifdef _MSC_VER
> #define snprintf _snprintf
> #endif
>
> 3. Insert missing header file stdint.h into Rcpp/inst/include (attached).
>
> Dominick

I don't have bandwidth right now to test any of this. As you say VC++ is 
not supported by CRAN anyway.

Have you tried building client packages with VC++, as this is more 
likely to start getting problematic at that stage. Due to templates 
lazyness, most of the code of Rcpp is not actually compiled in Rcpp, but 
if/when needed by the client code.

Romain

-- 
Romain Francois
Professional R Enthusiast
+33(0) 6 28 91 30 30
http://romainfrancois.blog.free.fr
|- http://bit.ly/aAyra4 : highlight 0.2-2
|- http://bit.ly/94EBKx : inline 0.3.6
`- http://bit.ly/aryfrk : useR! 2010