[R-pkg-devel] Question about Clang 17 Error

Thu Oct 12 04:51:55 CEST 2023

Update: I submitted a new version of the package, but it did not fix the
issue. The package has now been archived and I do not have access to the
error log output anymore from r-devel-linux-x86_64-fedora-clang.

I did reproduce CRAN's configuration in a VM using the information provided
by CRAN for r-devel-linux-x86_64-fedora-clang. I still cannot reproduce the
error and at this point I believe that there is a chance that CRAN's
machine is misconfigured.

The specific error happens after rbedrock has been compiled and linked
successfully. The specific error is that the symbol
_ZNSt3__122__libcpp_verbose_abortEPKcz cannot be found when rbedrock.so is
loaded.This symbol was introduced into libc++ in Clang 15.0. What I believe
to be happening to cause the error is that Clang++ 17 is adding a reference
to this symbol when compiling and linking rbedrock.so but the dynamic
linker is loading an older version of libc++.so when trying to load
rbedrock.so and the symbol is not found.

If this is the cause, then I think that the CRAN machine needs to configure
the dynamic linker to use the Clang++ 17 libc++.so, or add the proper
command line options to R's config variables.

It's possible that the CRAN's r-devel-linux-x86_64-fedora-clang machine is
fine and I've missed something, and I would be happy if someone could help
me figure out what it is.

Also, a new issue cropped up when 0.3.1 was tested on the
r-oldrel-macos-x86_64 machine. /usr/bin/ar seems to have failed to produce
an archive. The other Mac versions did fine, so I'm not sure if this is a
random error or something related to my package. The error log is here:
https://www.r-project.org/nosvn/R.check/r-oldrel-macos-x86_64/rbedrock-00install.html

If anyone can help me resolve this, I'd appreciate it.

On Wed, Sep 27, 2023 at 2:54 PM Reed A. Cartwright <racartwright using gmail.com>
wrote:

> Is there any way to submit packages directly to the CRAN's clang17 setup?
> I can enable verbose output for CMake and compare the output, but I'd
> rather not clog up the CRAN incoming queue just to debug a linker error?
>
> On Wed, Sep 27, 2023 at 2:43 PM Simon Urbanek <simon.urbanek using r-project.org>
> wrote:
>
>> It looks like a C++ run-time mismatch between what cmake is using to
>> build the static library and what is used by R. Unfortunately, cmake hides
>> the actual compiler calls so it's hard to tell the difference, but that
>> setup relies on the correct sequence of library paths.
>>
>> The rhub manually forces -stdlib=libc++ to all its CXX flags
>>
>> https://urldefense.com/v3/__https://github.com/r-hub/rhub-linux-builders/blob/master/fedora-clang-devel/Makevars__;!!IKRxdwAv5BmarQ!bAZgiOQaK4hd5BTk_Ldx9IEHgzHKVbC-uMkvYv5GOVkZDvbedcGwS8dQ4MWXRjukFfds7UpiR9NDZfEoUCWeoVnCfrDa$
>> so it is quite different from the gannet tests-clang-trunk setup (also
>> note the different library paths), but that's not something you can do
>> universally in the package, because it strongly depends on the toolchain
>> setup.
>>
>> Cheers,
>> Simon
>>
>>
>> > On 28/09/2023, at 9:37 AM, Reed A. Cartwright <racartwright using gmail.com>
>> wrote:
>> >
>> > I was unable to reproduce the error on the rhub's clang17 docker image.
>> >
>> > I notice that the linking command is slightly different between systems.
>> > And this suggests that I need to find some way to get CRAN to pass
>> -stdlib
>> > flag at the linking stage.
>> >
>> > CRAN:
>> > /usr/local/clang17/bin/clang++ -std=gnu++17 -shared
>> > -L/usr/local/clang/lib64 -L/usr/local/clang17/lib
>> -L/usr/local/gcc13/lib64
>> > -L/usr/local/lib64 -o rbedrock.so actors.o bedrock_leveldb.o dummy.o
>> init.o
>> > key_conv.o nbt.o random.o subchunk.o support.o -L./leveldb-mcpe/build
>> > -pthread -lleveldb -lz
>> >
>> > RHUB:
>> > clang++-17 -stdlib=libc++ -std=gnu++14 -shared -L/opt/R/devel/lib/R/lib
>> > -L/usr/local/lib -o rbedrock.so actors.o bedrock_leveldb.o dummy.o
>> init.o
>> > key_conv.o nbt.o random.o subchunk.o support.o -L./leveldb-mcpe/build
>> > -pthread -lleveldb -lz -L/opt/R/devel/lib/R/lib -lR
>> >
>> > On Wed, Sep 27, 2023 at 11:36 AM Gábor Csárdi <csardi.gabor using gmail.com>
>> > wrote:
>> >
>> >> You might be able to reproduce it with the clang17 container here:
>> >>
>> >>
>> https://urldefense.com/v3/__https://r-hub.github.io/containers/__;!!IKRxdwAv5BmarQ!a5vkX68B5unua6_Zsh92b99AXfbJiewU7Mp0nqAKE0JDT8v3g2d08JZ8Yq_0ubp0j4GeTWWLjLVAN-FoLqhhk9c$
>> >> You can either run it directly or with the rhub2 package:
>> >>
>> >>
>> https://urldefense.com/v3/__https://github.com/r-hub/rhub2*readme__;Iw!!IKRxdwAv5BmarQ!a5vkX68B5unua6_Zsh92b99AXfbJiewU7Mp0nqAKE0JDT8v3g2d08JZ8Yq_0ubp0j4GeTWWLjLVAN-FoxPbUQlE$
>> >>
>> >> Gabor
>> >>
>> >> On Wed, Sep 27, 2023 at 8:29 PM Reed A. Cartwright
>> >> <racartwright using gmail.com> wrote:
>> >>>
>> >>> My package, RBedrock, is now throwing an error when compiled against
>> >>> Clang17. The error log is here:
>> >>>
>> >>>
>> >>
>> https://urldefense.com/v3/__https://www.stats.ox.ac.uk/pub/bdr/clang17/rbedrock.log__;!!IKRxdwAv5BmarQ!a5vkX68B5unua6_Zsh92b99AXfbJiewU7Mp0nqAKE0JDT8v3g2d08JZ8Yq_0ubp0j4GeTWWLjLVAN-FoNhThuZA$
>> >>>
>> >>> The important part is
>> >>> """
>> >>> Error: package or namespace load failed for ‘rbedrock’ in
>> dyn.load(file,
>> >>> DLLpath = DLLpath, ...):
>> >>> unable to load shared object
>> >>>
>> >>
>> '/data/gannet/ripley/R/packages/tests-clang-trunk/rbedrock.Rcheck/00LOCK-rbedrock/00new/rbedrock/libs/rbedrock.so':
>> >>>
>> >>>
>> >>
>> /data/gannet/ripley/R/packages/tests-clang-trunk/rbedrock.Rcheck/00LOCK-rbedrock/00new/rbedrock/libs/rbedrock.so:
>> >>> undefined symbol: _ZNSt3__122__libcpp_verbose_abortEPKcz
>> >>> Error: loading failed
>> >>> """
>> >>>
>> >>> From what I can gather through googling, this error can be caused by
>> >> using
>> >>> the C linker when one of the dependent libraries is a C++ library.
>> >>>
>> >>> I cannot tell if this is an issue with my package (likely) or CRAN's
>> >>> clang17 setup (less likely).
>> >>>
>> >>> Background about the package: rbedrock is written in C but links
>> against
>> >> a
>> >>> C++ library (Mojang's leveldb fork)  via the library's C-API
>> functions. I
>> >>> use a dummy .cpp file in the source directory to trigger R into using
>> the
>> >>> C++ linker. That does still seem to be happening according to the log.
>> >>>
>> >>> Has anyone seen this before and know where I should start looking to
>> fix
>> >> it?
>> >>>
>> >>> Thanks.
>> >>>
>> >>> --
>> >>> Reed A. Cartwright, PhD
>> >>> Associate Professor of Genomics, Evolution, and Bioinformatics
>> >>> School of Life Sciences and The Biodesign Institute
>> >>> Arizona State University
>> >>> ==================
>> >>> Address: The Biodesign Institute, PO Box 876401, Tempe, AZ 85287-6401
>> USA
>> >>> Packages: The Biodesign Institute, 1001 S. McAllister Ave, Tempe, AZ
>> >>> 85287-6401 USA
>> >>> Office: Biodesign B-220C, 1-480-965-9949
>> >>> Website:
>> >>
>> https://urldefense.com/v3/__http://cartwrig.ht/__;!!IKRxdwAv5BmarQ!a5vkX68B5unua6_Zsh92b99AXfbJiewU7Mp0nqAKE0JDT8v3g2d08JZ8Yq_0ubp0j4GeTWWLjLVAN-Fo7waq1VI$
>> >>>
>> >>>        [[alternative HTML version deleted]]
>> >>>
>> >>> ______________________________________________
>> >>> R-package-devel using r-project.org mailing list
>> >>>
>> >>
>> https://urldefense.com/v3/__https://stat.ethz.ch/mailman/listinfo/r-package-devel__;!!IKRxdwAv5BmarQ!a5vkX68B5unua6_Zsh92b99AXfbJiewU7Mp0nqAKE0JDT8v3g2d08JZ8Yq_0ubp0j4GeTWWLjLVAN-FocAOfF7A$
>> >>
>> > ______________________________________________
>> > R-package-devel using r-project.org mailing list
>> >
>> https://urldefense.com/v3/__https://stat.ethz.ch/mailman/listinfo/r-package-devel__;!!IKRxdwAv5BmarQ!bAZgiOQaK4hd5BTk_Ldx9IEHgzHKVbC-uMkvYv5GOVkZDvbedcGwS8dQ4MWXRjukFfds7UpiR9NDZfEoUCWeoZEUmVSQ$
>> >
>>
>>

	[[alternative HTML version deleted]]