[Rcpp-devel] Rcpp 0.8.4

[Romain Francois]

Le 16/07/10 15:26, Dominick Samperi a écrit :
>
>
> On Fri, Jul 16, 2010 at 4:12 AM, Romain Francois
> <romain at r-enthusiasts.com <mailto:romain at r-enthusiasts.com>> wrote:
>
>     Le 16/07/10 01:18, Dominick Samperi a écrit :
>
>         Currently Rcpp 0.8.4 builds cleanly at CRAN under Windows only
>
>
>     This is as close to nonsense as it can be, please stop this quest of
>     yours to advertise problems that do not exist and RTFM
>
>
> I am only reporting what CRAN shows

Then we don't see the same thing. Here is the cran test results page for 
Rcpp: http://cran.r-project.org/web/checks/check_results_Rcpp.html

The three warnings on top are R-devel only, they are due to a more rigid 
test of links used in Rd pages. The development version of Rcpp has been 
fixed to take that into account.

The errors you see on solaris are due to new code. We have limited 
bandwidth to test on solaris/suncc and no actual machine to do it, 
therefore we sometimes have to resort to use cran checks as a way to 
find errors. In that instance, the developer responsible for this code 
has been alerted of the issue, and I'm sure he will find a workaround. 
If someone has access to a solaris/suncc combo, then I'd like to hear 
about it so that we can send them release candidates each time we want 
to release Rcpp.

The errors on OSX are related to version 0.8.3, because 0.8.4 has not 
yet been processed, it always take more time. However, we have 
information that Rcpp 0.8.4 works on OSX. The error in 0.8.3 was due to 
a name clash between Rcpp and a macro that is defined only in the ppc 
architecture, which my machine does not have.

Anyway "Currently Rcpp 0.8.4 builds cleanly at CRAN under Windows only" 
is just wrong.

> (does it publish nonsense?). My only
> "quest" is to
> prevent crashes of my work every time Rcpp does a release. I decided to
> use Rcpp because
> I thought we could cooperate on testing.
>
>     The author of that package chose to not follow the documented way to
>     pull in the headers , which is as I told you before to add
>
>
> The "documented way" is a moving target, and the only defense is to test
> before doing a release.
> (both unit testing and integration testing).

The recommendation to use LinkingTo: Rcpp came out with Rcpp 0.8.0, 
three months ago. You released your package 5 times in the meantime, and 
you seem to choose to not follow advice. In the future, I might not 
waste my time and just not advise you of anything.

> Since you "forgot" to add the GNU make system dependency I guess you
> could use a little
> review of the "documented way" as well.

Well, the email points out the "Rcpp-package" vignette, which contains 
the full story. I can write the document, but I cannot force you to read 
it, this has to come from you.

-- 
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