Skip to content

[R-pkg-devel] long user + system time with CRAN special checks clang 19

4 messages · Mauricio Vargas Sepulveda, Ivan Krylov, Tomas Kalibera

#
Dear R Community,

First of all, happy holidays.

I sent a package to CRAN and it passes all checks except this one:

   https://win-builder.r-project.org/incoming_pretest/cpp11tesseract_5.3.5_20241220_042113/specialChecks/clang19/outputs.txt

   checking cpp11tesseract_5.3.5.tar.gz ...

   Timings:
                    user system elapsed
   cpp11tesseract 18.624  1.701  22.045

I submitted a 2nd time after reducing the examples to the bare minimum and doing a copy and paste of the vignette outputs to provide a pre-rendered result, and it shows the same.

Is there any way to be able to replicate this error locally? I tried Win Builder and the R Consortium docker images, but being this a nonspecific error it is hard to replicate.

Best wishes,

??
Mauricio "Pach?" Vargas Sep?lveda
PhD Student, Political Science
University of Toronto
#
? Fri, 20 Dec 2024 08:00:08 +0000
Mauricio Vargas Sepulveda <m.sepulveda at mail.utoronto.ca> ?????:
It's not about the timings. 22 seconds for a full package installation
is quite tame compared to some C++ behemoths we have on CRAN (how about
3700 seconds?). The problem is that the resulting shared library fails
to load afterwards:
<https://win-builder.r-project.org/incoming_pretest/cpp11tesseract_5.3.5_20241220_042113/specialChecks/clang19/package/00install.out>

Unfortunately, it's still a puzzle why the function
tesseract::TessBaseAPI::GetVariableAsString(char const*, std::string*)
const fails to load from libtesseract.
#
? Fri, 20 Dec 2024 11:30:04 +0300
Ivan Krylov via R-package-devel <r-package-devel at r-project.org> ?????:
That's because libtesseract.so is linked to a different C++ standard
library. Compare:
c++filt says it's:

tesseract::TessBaseAPI::GetVariableAsString(char const*, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> >*) const

The symbol actually exported by libtesseract.so is named differently:

$ nm -gDP /usr/lib/x86_64-linux-gnu/libtesseract.so.5 | grep GetVariableAsString
_ZNK9tesseract11TessBaseAPI19GetVariableAsStringEPKcPNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE T a4f40 16

c++filt says:

tesseract::TessBaseAPI::GetVariableAsString(char const*, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >*) const

What should a package do when it's compiled by clang++ with the libc++
headers and library, but all system dependencies had been compiled with
g++ and linked with libstdc++? How to check packages using the clang
compiler without compiling the rest of the system from source using
clang? How do other C++ packages solve this problem?
#
On 12/20/24 10:04, Ivan Krylov via R-package-devel wrote:
It would say it is the responsibility of the check system/machine that 
the toolchain and the libraries used to build R packages are compatible, 
so also use the same C++ library. There is nothing R packages should do 
about this. R packages only need to make sure their code is portable and 
that they do not download any pre-compiled code.

Tomas