RcppGSLExample fails to install on rhub (platform = "solaris-x86-patched-ods")

[pcarbo]

I was unable to find any instances of this issue being reported elsewhere (e.g., in R-package-devel), and I am a bit surprised by that, so I do apologize if this is a duplicate issue report.

I tested the provided "RcppGSLExample" package in the "solaris-x86-patched-ods" platform made available via rhub, and I got an error during the package installation step. I'm attaching the full 00install.out output; here is a brief summary of the relevant error messages, indicating issues with multiply defined labels:

/opt/developerstudio12.6/bin/CC -std=c++11 -library=stdcpp,CrunG3 -G -L/opt/csw/lib -o RcppGSLExample.so RcppExports.o colNorm.o -L/opt/csw/li\
b -lgsl -lgslcblas -lm
ld: fatal: symbol 'SEXPREC* Rcpp::wrap<gsl_vector>(const gsl_vector&)' is multiply-defined:
        (file RcppExports.o type=FUNC; file colNorm.o type=FUNC);
ld: fatal: symbol 'SEXPREC* Rcpp::wrap<gsl_vector_float>(const gsl_vector_float&)' is multiply-defined:
        (file RcppExports.o type=FUNC; file colNorm.o type=FUNC);
ld: fatal: symbol 'SEXPREC* Rcpp::wrap<gsl_vector_int>(const gsl_vector_int&)' is multiply-defined:
        (file RcppExports.o type=FUNC; file colNorm.o type=FUNC);

00install.out.gz

[eddelbuettel]

Yes, but in general many more things fail on Solaris than on the other OSs so I am not sure what I could possibly do for you here -- not having access to Solaris myself (besides RHub) etc

[eddelbuettel]

(Unless we did a simple mistake here, didn't inline the definitions properly and include the header multiple times. But for simple errors like that g++ and clang++ are also reasonably attentive....)

[pcarbo]

Thanks for the quick response @eddelbuettel—let me see if I can come up with a fix; if I do, I will report it here. At last resort, I will ask CRAN to grant us an exception for our package (which depends on RcppGSL).

[eddelbuettel]

I see. I just checked another small package of mine depending in RcppGSL too, and it is reported 'healthy' at CRAN including Slowlaris so this does not appear to affect all...

https://cloud.r-project.org/web/checks/check_results_RcppZiggurat.html

Ditto with mvabund which is kinda the grand-daddy of all of these:
https://cloud.r-project.org/web/checks/check_results_mvabund.html

[pcarbo]

A solution that works—although not elegant at all—is to combine the RcppExports.cpp and colNorm.cpp into one .cpp file. When I do this, the ERRORs disappear.

[eddelbuettel]

I took another look and there were arguably four inline statements missing at the end of RcppGSLForward.h:

modified   inst/include/RcppGSLForward.h
@@ -215,11 +215,11 @@ namespace Rcpp {
     template<> inline SEXP wrap(const gsl_matrix_view&);         
     template<> inline SEXP wrap(const gsl_matrix_const_view&);
 
-    template <typename T> SEXP wrap(const ::RcppGSL::vector<T>&);
-    template <typename T> SEXP wrap(const ::RcppGSL::matrix<T>&);
+    template <typename T> inline SEXP wrap(const ::RcppGSL::vector<T>&);
+    template <typename T> inline SEXP wrap(const ::RcppGSL::matrix<T>&);
     
-    template <typename T> SEXP wrap(const ::RcppGSL::vector_view<T>&);
-    template <typename T> SEXP wrap(const ::RcppGSL::matrix_view<T>&);
+    template <typename T> inline SEXP wrap(const ::RcppGSL::vector_view<T>&);
+    template <typename T> inline SEXP wrap(const ::RcppGSL::matrix_view<T>&);
     
 }
 

I am surprised that g++ on Linux is so much more forgiving. Trouble is testing the error from RcppGSLExample requires an updated package first...

Edit: Noticed I did already go all the way to test the client package during tests: test_client_package.R. So now I am surprised you are bitten by this as the test should have shown it too. Hm.

[eddelbuettel]

And I may just have fixed it.

"Bad" borked Solaris build from maybe an half-an-hour ago:

RcppGSL 0.3.7.1: ERROR

Build ID: RcppGSL_0.3.7.1.tar.gz-c580dfc86e7c4dad9464437195d5ddea
Platform: Oracle Solaris 10, x86, 32 bit, R-release, Oracle Developer Studio 12.6
Submitted: 2 minutes 26.7 seconds ago
Build time: 2 minutes 23.4 seconds

ERRORS:

• checking whether package ‘RcppGSL’ can be installed ... ERROR
  Installation failed.
  See ‘/export/home/Xv3Utay/RcppGSL.Rcheck/00install.out’ for details.

See the full build log:
HTML: https://builder.r-hub.io/status/RcppGSL_0.3.7.1.tar.gz-c580dfc86e7c4dad9464437195d5ddea
Text: https://builder.r-hub.io/status/original/RcppGSL_0.3.7.1.tar.gz-c580dfc86e7c4dad9464437195d5ddea
Artifacts: https://artifacts.r-hub.io/RcppGSL_0.3.7.1.tar.gz-c580dfc86e7c4dad9464437195d5ddea

"Good" one on new code I will commit now:

RcppGSL 0.3.7.1: OK

Build ID: RcppGSL_0.3.7.1.tar.gz-9ded1fed1a7344e6a6aa1b15bd8dfa81
Platform: Oracle Solaris 10, x86, 32 bit, R-release, Oracle Developer Studio 12.6
Submitted: 2 minutes 59.6 seconds ago
Build time: 2 minutes 58 seconds

See the full build log:
HTML: https://builder.r-hub.io/status/RcppGSL_0.3.7.1.tar.gz-9ded1fed1a7344e6a6aa1b15bd8dfa81
Text: https://builder.r-hub.io/status/original/RcppGSL_0.3.7.1.tar.gz-9ded1fed1a7344e6a6aa1b15bd8dfa81
Artifacts: https://artifacts.r-hub.io/RcppGSL_0.3.7.1.tar.gz-9ded1fed1a7344e6a6aa1b15bd8dfa81

[eddelbuettel]

Package 0.3.8 is now on CRAN and I tried to say 'Thank you' to you on Twitter too.

[pcarbo]

Awesome! I am equally surprised that this error does not show up on other systems. I will see if the new RcppGSL version fixes our issue with the mashr package.

[eddelbuettel]

So am I! I was thinking about this on the weekend, and given that RcppGSL has been there for years I can only suspect that it is a more recent compiler feature.

But whatever 🍰 is that a real and genuine bug got squashed! And it is good to see RcppGSL having some tires kicked. I always liked the package, and it is not so widely used so lovely to see it coming to use across town :) (It also got better a few years when I realized how to improve the ctor/dtor pairing around the inner C objects). So thanks again for the very helpful issue ticket, and cheers from a few miles away.

[pcarbo]

Our packages have relied very heavily on Rcpp, RcppArmadillo, et al, and this is the first hicccup we have encountered — an (overdue) thank you to you!

[eddelbuettel]

Yes, I sort-of keep a list of Rcpp using packages (which once started as a short list...) so I know of some of the StephensLab goodies. And we can blame the "hickup" on the road less traveled -- RcppGSL is (for whichever reason) used a lot less than RcppArmadillo (and I partly blame having to link to GSL, partly because some content is more esoteric....)