Le 11/07/10 18:27, Dominick Samperi a écrit :


On Sun, Jul 11, 2010 at 12:06 PM, Romain Francois
<[email protected] <mailto:[email protected]>> wrote:

    Le 11/07/10 17:51, Dominick Samperi a écrit :

        On Sun, Jul 11, 2010 at 4:58 AM, Romain Francois
        <[email protected] <mailto:[email protected]>
        <mailto:[email protected]
        <mailto:[email protected]>>> wrote:Le

        11/07/10 02:14, Dominick Samperi a écrit :

                Romain,

                I made two small changes that should have no impact at
        all since
                they only
                apply when you build using windows DLL's (not static libs).
                Obviously these
                are not tested by a win-builder submission.


            Why not.


        Because testing this would require PATH to be changed at CRAN to
        include
        the directory containing Rcpp.dll, and this must be done BEFORE
        the package
        build or check process. This would have to be done using a
        Windows batch
        file as part of the initialization. Doing it in configure.win
        will not
        work. Also,
        a real test would require another package (like cxxPack) that
        links against
        Rcpp dynamically.


    As I said, check the runit.client.package.R test that does exactly
    this: checking that a client package works.

    We do test that documented ways to include Rcpp headers and link
    against the user libraries work.

    While on the subject, the current best way to pull header files in
    is to use LinkingTo: Rcpp and not:

    PKG_CPPFLAGS = $(shell $(R_HOME)/bin/Rscript --vanilla -e
    "Rcpp:::CxxFlags()")

    which is what several packages are still using, including cxxPack.


Why are you telling me what I obviously already know?

Why are you not following best practices if you know about them.

Your reply does
not address
any of the points that I raised here about Windows dynamic linking.

I do not have the answer. You pretend you have some answer, back them up with actual evidence.

In
particular,
runit.client.package.R does not address these issues at all (only the STATIC
library libRcpp.a is used).

Yes, but you can alter them as examples. The tests essentially compile another package that use Rcpp, you could adapt them to test whatever other scheme you want ... but I will not do it for you.


Dominick


--
Romain Francois
Professional R Enthusiast
+33(0) 6 28 91 30 30
http://romainfrancois.blog.free.fr
|- http://bit.ly/bc8jNi : Rcpp 0.8.4
|- http://bit.ly/dz0RlX : bibtex 0.2-1
`- http://bit.ly/a5CK2h : Les estivales 2010

_______________________________________________
Rcpp-devel mailing list
[email protected]
https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/rcpp-devel

Reply via email to