How to cope with older library installed in `/usr/lib` by sysadmin - c++

I have recently got an account on a supercomputer grid, and I'm trying to compile my code in theri system. Problem is that program won't link with following errors:
/mnt/opt/tools/slc6/binutils/2.22/bin/ld: warning: libboost_system.so.1.55.0, needed by /mnt/home/jbzdak/tools/boost_1_55//lib/libboost_thread.so, may conflict with libboost_system.so.5
/mnt/opt/tools/slc6/binutils/2.22/bin/ld: /mnt/home/jbzdak/tools/boost_1_55//lib/libboost_thread.so: undefined reference to symbol '_ZN5boost6system15system_categoryEv'
/mnt/opt/tools/slc6/binutils/2.22/bin/ld: note: '_ZN5boost6system15system_categoryEv' is defined in DSO /mnt/home/jbzdak/tools/boost_1_55//lib/libboost_system.so.1.55.0 so try adding it to the linker command line
/mnt/home/jbzdak/tools/boost_1_55//lib/libboost_system.so.1.55.0: could not read symbols: Invalid operation
collect2: error: ld returned 1 exit status
Which is due to the fact that my program needs boost 1.55, and only 1.39 is instlled on the system in /usr/lib64. I have installed my version of boost in local folder, but somehow still system one is loaded first.
Here is excerpt from flags passed to the compiler:
-std=gnu++11 -Werror -Wall -lboost_thread -lboost_filesystem -lboost_system -lboost_iostreams -g -DG4OPTIMISE -Iinclude
-W -Wall -ansi -pedantic -Wno-non-virtual-dtor -Wno-long-long -Wwrite-strings -Wpointer-arith -Woverloaded-virtual -pipe -O2
full listing of flags is here (they should be irrevelant).
Here are revelant config variables:
LIBRARY_PATH /mnt/home/jbzdak/tools/boost_1_55/lib:
CPLUS_INCLUDE_PATH /mnt/home/jbzdak/tools/boost_1_55/include:/mnt/home/jbzdak/tools/geant4.9.6.3/compile/include/Geant4
LD_LIBRARY_PATH /mnt/home/jbzdak/tools/boost_1_55/lib:/mnt/opt/tools/slc6/gcc/4.8.3/lib64: ...
Directory /mnt/home/jbzdak/tools/boost_1_55 contains installed boost library.
I use GCC 4.8.3 with ld 2.22.
I have very little experience with linker errors hence the question. Is there any way to exclude boost libraries in /usr/lib64, or make the linker use locally installed libraries, and and ignore the system one?

I said in a comment:
There's no -L/alternative/location/of/boost/lib shown, so the compiler (linker) doesn't know it needs to look somewhere else for your modern Boost library. You may need -Wl,rpath,/alternative/location/of/boost/lib as well.
And the question was asked:
Why didn't LD_LIBRARY_PATH solve the issue?
Because LD_LIBRARY_PATH is a run-time variable rather than a link-time variable. It affects where the /lib/ld.so.1 (or equivalent) dynamic loader looks for libraries when you run a program, not where the linker looks to find its libraries.

After some additional debugging and asking another question, I found out the root cause of problem. Any -L parameter has precedence over LIBRARY_PATH and somehow -L/usr/lib64 was added (hence it had precedence over my version).
To check what options are sent to gcc pass -v parameter.

Related

udata.cpp: undefined reference to `icudt71_dat'

I am getting an odd ICU related linking error in the now project when building on Ubuntu 22.04.
/usr/bin/ld: /usr/bin/ld: DWARF error: invalid or unhandled FORM value: 0x23
/home/bkey1/vcpkg/installed/x64-linux/debug/lib/libicuuc.a(udata.ao): in function `openCommonData(char const*, int, UErrorCode*)':
udata.cpp:(.text+0x23f7): undefined reference to `icudt71_dat'
/usr/bin/ld: udata.cpp:(.text+0x2458): undefined reference to `icudt71_dat'
The link command is as follows.
usr/bin/cmake -E cmake_link_script CMakeFiles/now.dir/link.txt --verbose=1
/usr/bin/c++ -std=c++2a -Wall -Wextra -Wfloat-equal -Wno-long-long -Wpedantic -funsigned-char -D_GNU_SOURCE=1 -rdynamic CMakeFiles/now.dir/GetStardate.cpp.o CMakeFiles/now.dir/GetTime.cpp.o CMakeFiles/now.dir/GetTimePlatformPOSIX.cpp.o CMakeFiles/now.dir/GetTimePlatformWin32.cpp.o CMakeFiles/now.dir/ISO8601_time.cpp.o CMakeFiles/now.dir/InitLocale.cpp.o CMakeFiles/now.dir/executable_path.cpp.o CMakeFiles/now.dir/now.cpp.o CMakeFiles/now.dir/nowStrings.cpp.o -o now /home/bkey1/vcpkg/installed/x64-linux/debug/lib/libboost_chrono.a /home/bkey1/vcpkg/installed/x64-linux/debug/lib/libboost_filesystem.a /home/bkey1/vcpkg/installed/x64-linux/debug/lib/libboost_locale.a /home/bkey1/vcpkg/installed/x64-linux/debug/lib/libboost_log.a /home/bkey1/vcpkg/installed/x64-linux/debug/lib/libboost_program_options.a /home/bkey1/vcpkg/installed/x64-linux/debug/lib/libboost_regex.a /home/bkey1/vcpkg/installed/x64-linux/debug/lib/libboost_system.a /home/bkey1/vcpkg/installed/x64-linux/debug/lib/libboost_thread.a /home/bkey1/vcpkg/installed/x64-linux/debug/lib/libboost_date_time.a /home/bkey1/vcpkg/installed/x64-linux/debug/lib/libboost_log_setup.a /home/bkey1/vcpkg/installed/x64-linux/debug/lib/libboost_atomic.a /home/bkey1/vcpkg/installed/x64-linux/debug/lib/libicudata.a /home/bkey1/vcpkg/installed/x64-linux/debug/lib/libicui18n.a /home/bkey1/vcpkg/installed/x64-linux/debug/lib/libicuuc.a
First I would like to stress, that there is very little information provided, so the answers such as my own will most likely need to guess what is happening. On the other hand I understand your situation: you cannot include info, that you don't know is relevant to the topic.
Answer:
I would like to draw your attention to the fact, that the symbol icudt71_dat does not appear anywhere in the code directly, but is generated using a macro. (Check .../source/common/unicode/utypes.h) So if the linker complains about not having found such a symbol it most probably means you are linking against a different version of the library than you have a header for. Now I don't know how specifically this could have happened, I would have to see your system, include path, link path etc. However I strongly suggest to revisit both include and link paths. You could perhaps recompile the library, verify, there is no other version recompile and start again. If the header corresponds to the source, it should work.
Aparently you missing icu-devtools package. Install that with command
sudo apt get install icu-devtools

GCC: libstdc++.so: Error adding sybols: file in wrong format

I am trying to compile for a gd32v chip using gcc(the riscv version on the arch community repo).
Compiling seems to work fine, however when trying to link the objects into an elf file, I get the error:
Linking ELF target: main.elf
riscv64-linux-gnu-g++ #_linker_flags -o main.elf ../../bmptk-RISC-V/targets/risc_v/gd32v/gd32vf103xb_boot.o hwlib.o main.o ../../bmptk-RISC-V/targets/risc_v/GD32VF103_standard_peripheral/Source/gd32vf103_rcu.o ../../bmptk-RISC-V/targets/risc_v/GD32VF103_standard_peripheral/Source/gd32vf103_gpio.o ../../bmptk-RISC-V/targets/risc_v/GD32VF103_standard_peripheral/system_gd32vf103.o bmptk_heap_none.o bmptk_fixed_size_stack.o -Os -Tmain.ld
/usr/lib/gcc/riscv64-linux-gnu/10.2.0/../../../../riscv64-linux-gnu/bin/ld: /usr/lib/gcc/riscv64-linux-gnu/10.2.0/../../../../riscv64-linux-gnu/lib/libstdc++.so: error adding symbols: file in wrong format
collect2: error: ld returned 1 exit status
make: *** [../../bmptk-RISC-V/Makefile.inc:1498: main.elf] Error 1
In this make rule, I am using a file '_linker_flags' for my linker flags, to keep the terminal clean during compilation. The contents of this file are as follows:
-march=rv32imac -mabi=ilp32 -Os -fdata-sections -ffunction-sections -I../../bmptk-RISC-V/targets/risc_v/ -I../../bmptk-RISC-V/targets/risc_v/GD32VF103_standard_peripheral -I../../bmptk-RISC-V/targets/risc_v/GD32VF103_standard_peripheral/Include -I../../bmptk-RISC-V/targets/risc_v/RISCV/drivers -I../../bmptk-RISC-V/targets/risc_v -I/usr/include -I/usr/include -I../../hwlib-RISC-V/library -I../../Catch2/single_include -I../../Catch2/single_include/catch2 -I../../boost_1_69_0 -I../../bmptk-RISC-V -I../../bmptk-RISC-V/targets -I../../bmptk-RISC-V/targets/risc_v -I../../bmptk-RISC-V/targets/risc_v/RISCV -I../../bmptk-RISC-V/targets/risc_v/RISCV/drivers -DHWCPP_FAKE_OSTREAM -DBMPTK_TARGET=gd32vf103v -DBMPTK_TARGET_gd32vf103v -DHWLIB_TARGET_gd32vf103v -DHWCPP_TARGET_gd32vf103v -DGF_TARGET_gd32vf103v -DBMPTK_CHIP=gd32vf103v -DBMPTK_CHIP_gd32vf103v -DBMPTK_XTAL= -DBMPTK_BAUDRATE=38400 -DHWLIB_BAUDRATE=38400 -DGODAFOSS_BAUDRATE=38400 -DGF_BAUDRATE=38400 -DBMPTK_VERSION=V04_00_work_in_progress_2020_05_23 -DBMPTK_EMBEDDED -lgcc -Wl,-Map,main.map -Wl,--gc-sections -Wl,-fatal-warnings
I'm not familiar with this error, does anyone know what I would have to look into to fix this?
EDIT:
I asked a teacher at school and they told me that the problem most likely arised from using a mismatching linker and compiler, or that some object files weren't cleaned when calling make. I made sure all objects were deleted before compiling and made sure the compiler and linker were the same.
They should be the same. I am running riscv64-linux-gnu-ld version 2.35 and riscv64-linux-gnu-g++ version 10.2.0. Both are from the arch community repository.
To see exactly the mapping/switches of the libraries supported by your compiler you can use : riscv64-linux-gnu-g++ -print-multi-lib. If you compiler was compiled with multilib enabled you can choose an rv32 libs without hard float otherwise it will not link also since you are compiler for rv32imac.
If your compiler was build without the multlib option you have two option:
Compile with -nostdlib and provide the needed file to the linker crt, libc libgcc ... or you can get a compiler which was build with multilib enabled.

Cross compilation for gpsd shows "unrecognized option"

I am cross-compiling gpsd3.20 on my Ubuntu 16.04 for the ARM architecture. As you may know, gpsd uses Sconsctruct to compile the source codes. During my cross-compilation, the moment when it needs to create the libgps.so it shows an error unrecognized option '-Wl, -Bsymbolic'.
Before posting the question here, I have tried t check my toolchain binaries and I found out that if I run this line manually:
sudo ./arm-v7a-linux-gnueabihf-ld -o test/gpsd-3.20/libgps.so.25.0.0 -pthread -shared -Wl,-Bsymbolic-functions -Wl,-soname=libgps.so.25 test/gpsd-3.20/os_compat.os test/gpsd-3.20/rtcm2_json.os test/rtcm3_json.os test/gpsd-3.20/shared_json.os test/gpsd-3.20/timespec_str.os test/gpsd-3.20/libgpsmm.os -L. -lrt -lm -lrt
The above commands print out the exact error as I mentioned previously. However, if I run the exact command replacing ld with gcc, then there is no any errors.
sudo ./arm-v7a-linux-gnueabihf-gcc -o test/gpsd-3.20/libgps.so.25.0.0 -pthread -shared -Wl,-Bsymbolic-functions -Wl,-soname=libgps.so.25 test/gpsd-3.20/os_compat.os test/gpsd-3.20/rtcm2_json.os test/rtcm3_json.os test/gpsd-3.20/shared_json.os test/gpsd-3.20/timespec_str.os test/gpsd-3.20/libgpsmm.os -L. -lrt -lm -lrt
Upon checking the arm-v7a-linux-gnueabihf-gcc --help, I found out that, gcc support -Wloptions whereas in the arm-v7a-linux-gnueabihf-ld it doesn't support the -Wl options. So now I am not sure how to change the SConstruct file so that it doesn't execute ld instead I want it to execute gcc especially for the libgps.so part.
(can't comment), so as answer: have you tried to set the env.-var.:
export LD=arm-v7a-linux-gnuabihf-gcc
Gcc takes -Wl,XXX and passes XXX to the linker.
I think you've got two combining problems here, though there's some guessing involved without looking into the build itself. First, scons shouldn't be adding the flag when building a library (https://github.com/SCons/scons/issues/3248 - fixed but, I believe, not part of a release). Second, "linking" should probably be done using gcc. If you call gcc to link, it still calls the linker behind the scenes - after dealing with options that are intended for gcc, which -Wl,-Bsymbolic is, it means pass -Bsymbolic on to the linking phase (indicated by -Wl, the 'l' meaning linker). So I'm supposing that the way you've told scons about the cross toolchain isn't quite right either, if it's calling ld directly you're probably going to have other issues as well.

Linking against a c/c++ library

I have some basic questions regarding linking against a C/C++ library. I am trying to understand the difference in using the two different usages -L/usr/local/lib -lm usage and /usr/local/lib/libm.a usage. E.g., when I compile and link an example from the [SUNDIALS] library, both of the following work
gcc -Wall cvRoberts_dns.c -o cvRoberts_dns.exe -I/usr/local/include -L/usr/local/lib/ -lsundials_cvode -lsundials_nvecserial -lm
OR
gcc -Wall cvRoberts_dns.c -o cvRoberts_dns.exe /usr/local/lib/libsundials_cvode.a /usr/local/lib/libsundials_nvecserial.a
However, to compile and link an example from the library [libsbml], the following works
g++ -Wall readSBML.cpp -o readSBML.exe -I/usr/local/include -L/usr/local/lib -lsbml
but the this does not
g++ -Wall readSBML.cpp -o readSBML.exe /usr/local/lib/libsbml.a
If required, I can post the complete error message I get, but the last line of the message is as follows
ld: symbol(s) not found for architecture x86_64
clang: error: linker command failed with exit code 1 (use -v to see invocation)
My questions are as follows:
In the second style of linking (of the first example), there is no information regarding where to find the include files (header files), how does the compiler know the information supplied in -I/usr/local/include which is provided in the first style of the first example?
In the second style of first example there is no /usr/local/lib/libm.a (it actually gives an error message that libm.a cannot be found if I try to include it), then why -lm is required in the first style?
How do I compile the second example in the second style (i.e., using /usr/local/lib/libsbml.a)? I do see that there are files - libsbml.a and libsbml-static.a in the /usr/local/lib folder, but none of them work.
If it helps, I am on an OS X machine.
I would be very thankful if any one could help in this regard.
Just an update - I tried
g++ -Wall readSBML.cpp -o readSBML.exe /usr/local/lib/libsbml.5.dylib
and that compiled and linked just fine.
Thanks
SN
In general
The -L option is meant to find where the libraries themselves are. Each library is a collection of one or more object code (machine language) files. There is no need to find the include files.
The -I option has nothing to with linker, it helps the compiler resolve the header files used in your driver programme( eg Roberts_dns.c). This happens during the pre-processing stage.
In the second style of linking (of the first example), there is no
information regarding where to find the include files (header files),..
If the compilation worked as you expected,it may be because /usr/local/include is in the default include path for gcc. To check the default include path for gcc do gcc -xc -E -v -.
In the second style of first example there is no
/usr/local/lib/libm.a(it actually gives an error message that libm.a
cannot be found if I try to include it), then why -lm is required in
the first style?
In Linux, some libraries like libc.a are directly linked to your execultable by default while libm.a is not. In Mac (your environment), though, libm is directly link to the executable by default. So you don't have to explicitly link it. It is less likely that libm.a is located in /usr/local/lib/. So you got an error. But why link it in the first place?

Linking errors with "-Wl,--no-undefined -Wl,--no-allow-shlib-undefined"

Using the flags "-Wl,--no-undefined -Wl,--no-allow-shlib-undefined" with GCC leads to the following compilation errors on the Travis CI image but not on my machine (both are Ubuntu 12.04 64-bits):
Linking CXX shared library libmocap.so
cd /tmp/_travis/build/src && /usr/bin/cmake -E cmake_link_script CMakeFiles/mocap.dir/link.txt --verbose=1
/usr/bin/g++ -fPIC --coverage -Werror -pedantic -Wno-long-long -Wall -Wextra -Wcast-align -Wcast-qual -Wformat -Wwrite-strings -Wconversion -fvisibility=hidden -Wl,--no-undefined -Wl,--no-allow-shlib-undefined -Wl,--as-needed -shared -Wl,-soname,libmocap.so.0.0.0 -o libmocap.so.UNKNOWN CMakeFiles/mocap.dir/abstract-marker.cc.o CMakeFiles/mocap.dir/abstract-virtual-marker.cc.o CMakeFiles/mocap.dir/color.cc.o CMakeFiles/mocap.dir/link.cc.o CMakeFiles/mocap.dir/marker-set-factory.cc.o CMakeFiles/mocap.dir/marker-set.cc.o CMakeFiles/mocap.dir/marker-trajectory-factory.cc.o CMakeFiles/mocap.dir/marker-trajectory.cc.o CMakeFiles/mocap.dir/marker.cc.o CMakeFiles/mocap.dir/mars-marker-set-factory.cc.o CMakeFiles/mocap.dir/math.cc.o CMakeFiles/mocap.dir/pose.cc.o CMakeFiles/mocap.dir/segment.cc.o CMakeFiles/mocap.dir/string.cc.o CMakeFiles/mocap.dir/trc-marker-trajectory-factory.cc.o CMakeFiles/mocap.dir/virtual-marker-one-point-measured.cc.o CMakeFiles/mocap.dir/virtual-marker-relative-to-bone.cc.o CMakeFiles/mocap.dir/virtual-marker-three-points-measured.cc.o CMakeFiles/mocap.dir/virtual-marker-three-points-ratio.cc.o CMakeFiles/mocap.dir/virtual-marker-two-points-measured.cc.o CMakeFiles/mocap.dir/virtual-marker-two-points-ratio.cc.o
/lib/x86_64-linux-gnu/libc.so.6: undefined reference to `_dl_argv#GLIBC_PRIVATE'
/lib/x86_64-linux-gnu/libc.so.6: undefined reference to `_rtld_global_ro#GLIBC_PRIVATE'
/usr/lib/gcc/x86_64-linux-gnu/4.6/libstdc++.so: undefined reference to `__tls_get_addr#GLIBC_2.3'
/lib/x86_64-linux-gnu/libc.so.6: undefined reference to `_rtld_global#GLIBC_PRIVATE'
/lib/x86_64-linux-gnu/libc.so.6: undefined reference to `__libc_enable_secure#GLIBC_PRIVATE'
collect2: ld returned 1 exit status
make[2]: *** [src/libmocap.so.UNKNOWN] Error 1
Strangely, this does not occur on my machine and hence is a bit difficult to reproduce.
What is the right way to link against the libc and libstdc++ when these flags are enabled?
(as you might guess from the output, I use CMake to generate the compilation command)
The default for the 2nd argument is "--allow-shlib-undefined".
Probably if you choose that option the code will build.
This 2nd argument deals with build time checking where enabling this means checking that the library that you are linking against in turn has its dependencies wired up at build time.. And that is not necessarily the case.
The first argument makes sure that you have not forgotten to state a dependency to a runtime library (which may also be a dependency that a runtime library has to another runtime library).For example if you are calling a function where the implementation is in an example runtime library "libfunc.so" and that library in turn will call a function in another runtime library "libext.so" then by declaring a dependency to both "func" and "ext" the libfunc.so will be generated to internally include a dependency reference to libext.
If you would leave out "--no undefined" and forget to add dependency declarations then the build would still succeed, trusting that your runtime linker would resolve dependencies at runtime. And since the build was successful you might trust that everything will be ok not knowing that the build has deferred responsibility to the runtime linker. But most often the runtime linker is not designed to search for unresolved references but expects to find such dependencies stated in the runtime library. And if no such reference is there you will get a runtime error. Runtime errors are typically much more costly than resolving a compile time error.