error Rlog1p #330

yfyang86 · 2021-03-16T15:08:42Z

On

R version 4.0.3 (2020-10-10)
Platform: aarch64-apple-darwin20.0.0 (64-bit)
Apple clang version 12.0.0 (clang-1200.0.32.29)
Target: arm64-apple-darwin20.3.0

The original source code fails with the message:

../inst/include/armadillo_bits/eop_aux.hpp:98:115: error: no member named 'Rlog1p' in namespace 'std'; did you mean simply 'Rlog1p'?

which shows a conflict of std::log1p with #define log1p Rlog1p in Rmath.h

On ``` R version 4.0.3 (2020-10-10) Platform: aarch64-apple-darwin20.0.0 (64-bit) Apple clang version 12.0.0 (clang-1200.0.32.29) Target: arm64-apple-darwin20.3.0 ``` The original source code fails with the message: ``` ../inst/include/armadillo_bits/eop_aux.hpp:98:115: error: no member named 'Rlog1p' in namespace 'std'; did you mean simply 'Rlog1p'? ``` which show a conflict of `std::log1p` with `#define log1p Rlog1p` in `Rmath.h`

eddelbuettel · 2021-03-16T15:13:20Z

Can you provide a short code example that triggers it?

Can you state which RcppArmadillo version you used? Can you maybe try the one in the Rcpp drat repo too (which is based on Armadill 10.3-rc*)?

Also, we may want to try to deal with this (if it all) at the #define level before sourcing any Armadillo files.

eddelbuettel · 2021-03-16T15:15:41Z

inst/include/armadillo_bits/eop_aux.hpp

@@ -22,6 +22,8 @@
 //! use of the SFINAE approach to work around compiler limitations
 //! http://en.wikipedia.org/wiki/SFINAE

+#undef log1p
+


We almost surely do not want to unconditionally undefine something in the upstream sources.

This one is a workaround. I will make another PR if I succeed in changing the configure file instead of modifying the upstream armadillo code.

Here is my environment information:

> sessionInfo() R version 4.0.3 (2020-10-10) Platform: aarch64-apple-darwin20.0.0 (64-bit) Running under: macOS 11.2 Matrix products: default BLAS: /Users/xxx/miniforge3/envs/py38/lib/R/lib/libRblas.dylib LAPACK: /Users/xxx/miniforge3/envs/py38/lib/R/lib/libRlapack.dylib locale: [1] C/UTF-8/C/C/C/C attached base packages: [1] stats graphics grDevices utils datasets methods base loaded via a namespace (and not attached): [1] compiler_4.0.3 tools_4.0.3

with (on Mac M1):

Configured with: --prefix=/Applications/Xcode.app/Contents/Developer/usr --with-gxx-include-dir=/Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX.sdk/usr/include/c++/4.2.1 Apple clang version 12.0.0 (clang-1200.0.32.29) Target: arm64-apple-darwin20.3.0 Thread model: posix InstalledDir: /Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin

and

> packageVersion("Rcpp") [1] ‘1.0.6’

I use the the latest CRAN/Github master version to install RcppArmadillo , then it reports an error

arm64-apple-darwin20.0.0-clang++ -std=gnu++11 -I"/Users/yifanyang/miniforge3/envs/py38/lib/R/include" -DNDEBUG -I'/Users/yifanyang/miniforge3/envs/py38/lib/R/library/Rcpp/include' -D_FORTIFY_SOURCE=2 -isystem /Users/yifanyang/miniforge3/envs/py38/include -mmacosx-version-min=11.0 -I/Users/yifanyang/miniforge3/envs/py38/include -I../inst/include -fPIC -ftree-vectorize -fPIC -fPIE -fstack-protector-strong -O2 -pipe -stdlib=libc++ -fvisibility-inlines-hidden -fmessage-length=0 -isystem /Users/yifanyang/miniforge3/envs/py38/include -fdebug-prefix-map=/Users/runner/miniforge3/conda-bld/r-base-split_1615277992527/work=/usr/local/src/conda/r-base-4.0.3 -fdebug-prefix-map=/Users/yifanyang/miniforge3/envs/py38=/usr/local/src/conda-prefix -c RcppArmadillo.cpp -o RcppArmadillo.o In file included from RcppArmadillo.cpp:21: In file included from ../inst/include/RcppArmadillo.h:31: In file included from ../inst/include/RcppArmadilloForward.h:49: In file included from ../inst/include/armadillo:374: ../inst/include/armadillo_bits/eop_aux.hpp:97:124: error: no member named 'Rlog1p' in namespace 'std'; did you mean 'log1p'? template<typename eT> arma_inline static typename arma_integral_only<eT>::result log1p (const eT x) { return eT( std::log1p(double(x)) ); } ~~~~~^~~~~ log1p /Users/yifanyang/miniforge3/envs/py38/lib/R/include/Rmath.h:75:15: note: expanded from macro 'log1p' #define log1p Rlog1p ^ /Users/yifanyang/miniforge3/envs/py38/bin/../include/c++/v1/cmath:435:9: note: 'log1p' declared here using ::log1p; ^ In file included from RcppArmadillo.cpp:21: In file included from ../inst/include/RcppArmadillo.h:31: In file included from ../inst/include/RcppArmadilloForward.h:49: In file included from ../inst/include/armadillo:374: ../inst/include/armadillo_bits/eop_aux.hpp:98:115: error: no member named 'Rlog1p' in namespace 'std'; did you mean simply 'Rlog1p'? template<typename eT> arma_inline static typename arma_real_only<eT>::result log1p (const eT x) { return std::log1p(x); } ^~~~~~~~~~ Rlog1p ../inst/include/armadillo_bits/eop_aux.hpp:99:86: note: 'Rlog1p' declared here template<typename eT> arma_inline static typename arma_cx_only<eT>::result log1p (const eT& x) { arma_ignore(x); return eT(0); } ^ /Users/yifanyang/miniforge3/envs/py38/lib/R/include/Rmath.h:75:15: note: expanded from macro 'log1p' #define log1p Rlog1p ^ In file included from RcppArmadillo.cpp:21: In file included from ../inst/include/RcppArmadillo.h:31: In file included from ../inst/include/RcppArmadilloForward.h:49: In file included from ../inst/include/armadillo:433: ../inst/include/armadillo_bits/fn_misc.hpp:170:21: error: no member named 'Rlog1p' in namespace 'std'; did you mean simply 'Rlog1p'? return (log_a + std::log1p(std::exp(negdelta))); ^~~~~~~~~~ Rlog1p /Users/yifanyang/miniforge3/envs/py38/lib/R/include/Rmath.h:74:9: note: 'Rlog1p' declared here double Rlog1p(double); ^ 3 errors generated. make: *** [/Users/yifanyang/miniforge3/envs/py38/lib/R/etc/Makeconf:181: RcppArmadillo.o] Error 1 ERROR: compilation failed for package ‘RcppArmadillo’

eddelbuettel · 2021-03-16T15:17:46Z

You can also try to define #undef ARMA_HAVE_LOG1P before your includes. Many options, let's get a reproducible example first as this may be compiler / platform dependent.

eddelbuettel · 2021-03-16T15:33:46Z

I am sure we can sort this out. Don't run off modifying configure, please, or waste time in other ways. Let's just analyse a little more.

Can you please provide a short file with either Rcpp::cppFunction() in a one-or-two-liner or for Rcpp::sourceCpp() to trigger this? RcppArmadillo has well over 800 reverse depends on CRAN and is used in a number of ways, and this has not come up yet. It could be as simple as re-ordering header files too. But we need to know more before we get into action.

(And that is precisely why we say here over at Rcpp to first file an issue to discuss matters before going off on pull requests which may just get rejected.)

yfyang86 · 2021-03-16T15:43:01Z

Here is an working example, with log1p in C++11 cmath ref:

> code = ' void simple_test(){std::cout << "log1p(0) = " << std::log1p(0) << std::endl;}'
> cppFunction(plugins=c("cpp11"), code)
file8a824422791b.cpp:7:51: error: no member named 'Rlog1p' in namespace 'std'; did you mean simply 'Rlog1p'?
 void simple_test(){std::cout << "log1p(0) = " << std::log1p(0) << std::endl;}
                                                  ^~~~~~~~~~
                                                  Rlog1p
/Users/yifanyang/miniforge3/envs/py38/lib/R/include/Rmath.h:74:9: note: 'Rlog1p' declared here
double  Rlog1p(double);
        ^
1 error generated.

Update: I delete the misleading void error.

eddelbuettel · 2021-03-16T16:00:01Z

Sorry, don't see it:

> Rcpp::cppFunction("double issue330(double x) { return std::log1p(x); }")
> issue330(1.234)
[1] 0.803794
>

Ubuntu 20.10, g++ 10.2. Also passes with clang++-11.

To me, you have not yet provided a reproducible example.

eddelbuettel · 2021-03-16T16:00:40Z

I see miniforge in your output; maybe you set yourself up for an interaction with other headers. I would not know. CRAN also does not use conda.

eddelbuettel · 2021-03-16T16:09:43Z

I can also merrily call ::log1p() as the Rcpp sources do too:

> Rcpp::cppFunction("double issue330r(double x) { return ::log1p(x); }")
> issue330r(1.234)
[1] 0.803794
>

So maybe for your Conda use you need to #undef HAVE_LOG1P ? Maybe you try and experiment in a simpler "clean room" setup via Docker. I still see no issue in our sources.

yfyang86 · 2021-03-16T16:11:22Z

I see miniforge in your output; maybe you set yourself up for an interaction with other headers. I would not know. CRAN also does not use conda.

May that's the reason. I tried on other platforms and got no error. I will try to make a clean install (of R) tomorrow and figure out why. I think it might be the way the compiler expands the macro causes the issue. Hence maybe this one is only an issue on Mac M1 miniforge3's R version (the tool chain on new Mac is a disaster for developers). I will close this one and report in other thread if there is anything interesting there.

eddelbuettel · 2021-03-16T16:14:37Z

We should still try to get to it (not here but maybe as a Conda bug report) as it seems that a define gets in the way.

kevinushey · 2021-03-16T16:53:27Z

The relevant part of Rmath.h:

https://github.com/wch/r-source/blob/75a1b9791926ba24e18b9f2f58634e8233898216/src/include/Rmath.h0.in#L72-L76

That is,

#if !defined(HAVE_WORKING_LOG1P)
/* remap to avoid problems with getting the right entry point */
double  Rlog1p(double);
#define log1p Rlog1p
#endif

So the issue is most likely that your installed copy of R believes (correctly or incorrectly?) that your platform doesn't have log1p defined, and so R is trying to stuff its own definition in.

Does compilation succeed if you set -DHAVE_WORKING_LOG1P?

yfyang86 · 2021-03-16T17:32:29Z

Does compilation succeed if you set -DHAVE_WORKING_LOG1P?

I also traced #if !defined(HAVE_WORKING_LOG1P) (version4.0.x) in Rmath.h, and will test why configure file (to be precisely, the conda R environment, not the configure file) fails to detect it.

Manually change the PKG_CXXFLAGS flag in RcppArmadillo/src/Makevars.in into:

PKG_CXXFLAGS = -I../inst/include @OPENMP_FLAG@ -DHAVE_WORKING_LOG1P

could compile the package successfully. That's why

I will try to make a clean install (of R) tomorrow and figure out why.

R-4.0 is installed through a standard native ARM64 conda. For anyone with the same issue, I record the conda environment here:

            shell level : 2
       user config file : /Users/yifanyang/.condarc
 populated config files : /Users/yifanyang/miniforge3/.condarc
          conda version : 4.9.2
    conda-build version : not installed
         python version : 3.9.2.final.0
       virtual packages : __osx=11.2=0
                          __unix=0=0
                          __archspec=1=arm64
       base environment : /Users/yifanyang/miniforge3  (writable)
           channel URLs : https://conda.anaconda.org/conda-forge/osx-arm64
                          https://conda.anaconda.org/conda-forge/noarch
          package cache : /Users/yifanyang/miniforge3/pkgs
                          /Users/yifanyang/.conda/pkgs
       envs directories : /Users/yifanyang/miniforge3/envs
                          /Users/yifanyang/.conda/envs
               platform : osx-arm64
             user-agent : conda/4.9.2 requests/2.25.1 CPython/3.9.2 Darwin/20.3.0 OSX/11.2
                UID:GID : 501:20
             netrc file : None
           offline mode : False

eddelbuettel · 2021-03-16T17:45:05Z

Manually change the PKG_CXXFLAGS flag in RcppArmadillo/src/Makevars.in into:

You probably want to add it to CXXFLAGS (or CPPFLAGS) in either your own ~/.R/Makevars or your installation-wide one in R's etc/ directory. That's what I do for, say, some Boost flags and like. Be mindful that you may need it several times for the different C++ standards: 11, 14, 17, ...

isuruf · 2021-03-17T04:10:05Z

So the issue is most likely that your installed copy of R believes (correctly or incorrectly?) that your platform doesn't have log1p defined, and so R is trying to stuff its own definition in.

This is an issue with conda-forge's R package. Please open an issue in https://github.com/conda-forge/r-base-feedstock and we'll get it fixed ASAP.

yfyang86 · 2021-03-17T12:44:52Z

Update Here on conda-forge. I recompiled R from the source, and this issue is solved.

eddelbuettel · 2021-03-17T12:49:11Z

Good to know you're set; would probably still be good to help the conda folks get the package right.

Sue9104 · 2023-03-28T07:01:04Z

Binary packages are available.
install.packages("RcppArmadillo", type="mac.binary") works for me.

mmtmr · 2023-04-09T20:28:48Z

Binary packages are available.
install.packages("RcppArmadillo", type="mac.binary") works for me.

This is the only solution that works for me, thank you!

eddelbuettel requested changes Mar 16, 2021

View reviewed changes

yfyang86 closed this Mar 16, 2021

yfyang86 mentioned this pull request Mar 17, 2021

R on ARM Mac M1:load failed for ‘stats’ and Rlog1p error conda-forge/r-base-feedstock#163

Closed

1 task

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

error Rlog1p #330

error Rlog1p #330

yfyang86 commented Mar 16, 2021

eddelbuettel commented Mar 16, 2021 •

edited

Loading

eddelbuettel Mar 16, 2021

yfyang86 Mar 16, 2021

eddelbuettel commented Mar 16, 2021

eddelbuettel commented Mar 16, 2021

yfyang86 commented Mar 16, 2021 •

edited

Loading

eddelbuettel commented Mar 16, 2021

eddelbuettel commented Mar 16, 2021

eddelbuettel commented Mar 16, 2021

yfyang86 commented Mar 16, 2021

eddelbuettel commented Mar 16, 2021

kevinushey commented Mar 16, 2021

yfyang86 commented Mar 16, 2021

eddelbuettel commented Mar 16, 2021 •

edited

Loading

isuruf commented Mar 17, 2021

yfyang86 commented Mar 17, 2021

eddelbuettel commented Mar 17, 2021

Sue9104 commented Mar 28, 2023

mmtmr commented Apr 9, 2023

error Rlog1p #330

error Rlog1p #330

Conversation

yfyang86 commented Mar 16, 2021

eddelbuettel commented Mar 16, 2021 • edited Loading

eddelbuettel Mar 16, 2021

Choose a reason for hiding this comment

yfyang86 Mar 16, 2021

Choose a reason for hiding this comment

eddelbuettel commented Mar 16, 2021

eddelbuettel commented Mar 16, 2021

yfyang86 commented Mar 16, 2021 • edited Loading

eddelbuettel commented Mar 16, 2021

eddelbuettel commented Mar 16, 2021

eddelbuettel commented Mar 16, 2021

yfyang86 commented Mar 16, 2021

eddelbuettel commented Mar 16, 2021

kevinushey commented Mar 16, 2021

yfyang86 commented Mar 16, 2021

eddelbuettel commented Mar 16, 2021 • edited Loading

isuruf commented Mar 17, 2021

yfyang86 commented Mar 17, 2021

eddelbuettel commented Mar 17, 2021

Sue9104 commented Mar 28, 2023

mmtmr commented Apr 9, 2023

eddelbuettel commented Mar 16, 2021 •

edited

Loading

yfyang86 commented Mar 16, 2021 •

edited

Loading

eddelbuettel commented Mar 16, 2021 •

edited

Loading