I have an installation of DPDK and I'm trying to compile code with it. This worked on my WSL machine, however on the test server, with the same install of dpdk, I'm getting the error:
/usr/local/include/rte_mempool.h: error: unknown type name 'ssize_t'
I've noticed this header does not include /sys/types.h, but it also doesn't include that on the machine that this works on. I don't know where it's supposed to come from, but it's coming from somewhere.
How do I get these headers to be aware of ssize_t?
As noted in a comment:
The compiler options include -std=c99 -O3 -march=native -I/usr/local/include -include rte_config.h, and a whole bunch of -l options (dpdk's make structure adds these in). This prompted me to run gcc --version and the working one is Ubuntu gcc 9.3.0 and the broken one is gcc 5.4.0. Seems like it might be an incompatibility between dpdk and the gcc installed.
As mentioned in the comments by #JonathanLeffier, the root cause of the issue is including sys/types.h when gcc option --std=c99 is passed. The easiest fix without modifying the DPDK or sample code is to include the path to types.h as part of cflags.
If the native build is for x86_64 target, follow these steps:
execute find /usr/include/ -name types.h to identify the right file for local build (this is because the current cflags has -march=native)
modify the CFLAGS from -std=c99 -O3 -march=native -I/usr/local/include -include rte_config.h to -std=c99 -O3 -march=native -I/usr/local/include -include rte_config.h --include=/usr/include/[target-machine]/sys/types.h
Note: In my humble suggestion please use pkg-config for populating the right CFLAGS and LDFLAGS for both shared and static binary.
Related
I'm running a Debian (Buster) container and my goal is to compile a small program I wrote which relies on libgit2. First, I was installing libgit2 via the libgit2-dev package and my Makefile had the following:
gcc -O2 -fpic -shared -I /usr/local/include -lgit2 -o output.so my_app.c
However, I'd rather have a "cleaner" environment and install libgit2 via the libgit-27 which, AFAIK, only installs the shared object binary instead of also including the development files like libgit2-dev does.
Using find I can find where the .so file is installed into:
$ find / -name "*git2*" -print 2>/dev/null
/usr/lib/x86_64-linux-gnu/libgit2.so.0.27.7
/usr/lib/x86_64-linux-gnu/libgit2.so.27
/usr/share/doc/libgit2-27
/var/lib/dpkg/info/libgit2-27:amd64.list
/var/lib/dpkg/info/libgit2-27:amd64.symbols
/var/lib/dpkg/info/libgit2-27:amd64.md5sums
/var/lib/dpkg/info/libgit2-27:amd64.shlibs
/var/lib/dpkg/info/libgit2-27:amd64.triggers
and I've been trying several combinations of linking this .so with gcc like:
gcc -O2 -fpic -shared -L /usr/lib/x86_64-linux-gnu/ -libgit2.so.27 -o output.so my_app.c
but so far I always get the following error:
my_app.c:1:10: fatal error: git2.h: No such file or directory
#include <git2.h>
^~~~~~~~
compilation terminated.
I understand this is a glaring lack of knowledge on how C compilation works. My two questions are:
Is it possible to compile my program by just relying on the libgit2-27 Debian Buster package instead of libgit2-dev? If not, why?
If yes, an example and explanation would be appreciated!
I've used linuxbrew to install gcc 5.3 on a machine on which I don't have sudo access. I now want to link with X11:
> gcc test.c -lX11
ld: cannot find -lX11
I've checked that libX11.so exists in /usr/lib64/ which is on the compiler's LIBRARY_PATH. If I use the system's gcc it works fine, but I need a newer version to compile my actual program.
use -L flag, like this -L/usr/lib64, or you can specify full path to library like this gcc test.c /usr/lib64/libX11.so
According to this comment by a linuxbrew developer,
linuxbrewed gcc removes /usr/lib64 from the library path because mixing system libraries with brewed libraries creates havoc.
The solution is to brew install linuxbrew/xorg/xorg.
I'm trying to compile the following project on a remote server.
I've git cloned the project on a folder called 'scode'.
The project requires glib2 and gsl libraries. Since I'm trying to compile on a remote server, I do not have sudo privileges. So I can't use a tool to install glib2 and gsl for me.
As a result, I've manually compiled both gsl and gslib2 under the folders 'scode/gsl' and 'scode/glib'.
I've had to modify the Makefile and add absolute paths to these directories as -I options.
Nonetheless, when I try to compile the final executable. I get the following error:
[dyuret#psglogin scode]$ make
gcc -O3 -D_GNU_SOURCE -Wall -std=c99 -I. -I /home-2/dyuret/scode/gsl
-I /home-2/dyuret/scode/glib/ pkg-config --cflags glib-2.0 scode.o svec.o pkg-config --libs glib-2.0 -lm -lgsl -lgslcblas -o scode
//home-2/dyuret/scode/glib/glib/libglib-2.0.la: file not recognized:
File format not recognized
collect2: error: ld returned 1 exit status make: * [scode] Error 1
I've researched the issue a bit. This link looks informative but I can't quite decipher what the author is saying, as I'm not that experienced with compilers, libtools and the compilation flow in general.
Any help would be much appreciated. I've already spent some time on this issue and I haven't been able to make much progress.
It sounds as if what you did in order to compile the libraries in non-default (i.e. non-system) locations was maybe wrong.
For packages using autoconf (i.e. that have a configure script in the source root) you're supposed to use the --prefix option to ./configure to set the target location where you want the package installed.
With packages building shared libraries, it's often essential to do the make install step, which it sounds as if you maybe didn't do.
Sorry for being vague, these things are a bit complicated.
Someone at my group helped me with the problem. Here're the steps he roughly carried out:
(1) Manually installed glib and additional libraries at $HOME directory - i.e. $HOME/lib, $HOME/include.
(1.1) I think he did this by './configure prefix=$HOME', 'make', 'make install'.
(2) Got rid of `pkg_config` usage, which was causing the problem I outlined originally. Here are his new CLFAGS and LIBS variables:
CFLAGS=-O3 -D_GNU_SOURCE -Wall -std=c99 -I. -I$$HOME/include -I$$HOME/include/glib-2.0 -I$$HOME/lib/glib-2.0/include
LIBS=-lglib-2.0 -lm -lgsl -lgslcblas -L$$HOME/lib -L/usr/local/cuda/lib64 -lcudart
After this, the code compiled without additional problems.
When I run the following command from a makefile on 64-bit Red Hat Enterprise Linux 5.0 using GCC 4.2.3:
gcc -c -ansi -pedantic -O0 -fPIC -I. -I.. -Iheader_files/include "source_file.c"
I get the following error:
cc1: error: unrecognized command line option "-lang-c"
Superficially, the problem is that "-lang-c" is no longer a valid option on newer versions of GCC. However, the deeper question is: Why does GCC receive a "-lang-c" option that wasn't in the original command?
Additional background:
The current installation doesn't have any GCC "specs" files that I can find
Running "gcc -dumpspecs" produces a long list of defaults, but "-lang-c" isn't among them.
"-lang-c" does not appear to be in any environment variables that I know of that influence GCC.
Any help would be appreciated.
Regards,
Mark Biesiada
Make sure that your gcc driver program is the same version as your installed GCC.
Add the -v option to your compile command to check the versions and where the options are coming from.
When I compile a simple Hello World! program that uses the sscanf function on my local Debian lenny x64, it works. But when I upload the same program to the server running CentOS x86, it will not work. If I do not use sscanf, then the program works on both computers.
gcc -std=c99 -O2 -pipe -m32
If I compile it with sscanf but without -std=c99, then it works on both computers.
gcc -O2 -pipe -m32
What is the problem with sscanf and c99 on CentOS x86 ? I thought that compiling with the -m32 flag would work on all Linuxes ? (I have limited access to the CentOS server, so I do not have access to error messages.)
Probably the CentOS box is using an old version of glibc. Since the nonstandard GNU extensions to their scanf implementation ended up making glibc conflict with c99, they added a nasty hack of redirecting *scanf to __isoc99_*scanf when -std=c99 is in use; if your copy of glibc is missing the __isoc99_sscanf symbol, the program will then fail to run.
Static linking, or linking to a different libc without ugly backwardsness-compatibility hacks, would solve the problem.
Are you uploading the binary or the source and then recompiling? If you are uploading the binary, you are probably running into a library compatibility issue between Debian and CentOS.
If that is the case, upload the source only and recompile on CentOS.
If you do not have permission to compile # CentOS, then try compiling a static binary. You can use dietlibc which makes smaller binaries than glibc or try EGLIBC which is the default C library that Debian will use starting Debian "squeeze".
I came up with the similar problem, it works # Ubuntu 64-bit, but the compile fails # CenseOS 64-bit (REHL5 desktop):
the error message is:
undefined reference to `__isoc99_sscanf#GLIBC_2.7'
when i copied the executable file compiled #Ubuntu to REHL5, and run it another error appeared:
elf file os abi invalid
it is compiled without flag -std=c99, i'm a newbie at C, and looking forword some workarounds, ex. add some flag.
Makefile:
CC=gcc
CCFLAGS= -Wall -O2 -DLINUX -I../include
demos:linuxdemo.c
$(CC) $(CCFLAGS) -o demoA linuxdemo.c -L../lib -lsense4 -lusb
$(CC) $(CCFLAGS) -o demoSO linuxdemo.c -lusb -lsense4
clean:
rm -f demoA
rm -f demoSO
You need to update your glibc to 2.7
download the rpm package from here:
http://archive.fedoraproject.org/pub/archive/fedora/linux/releases/8/Everything/x86_64/os/Packages/
needs:
libc-common-2.7-2.x86_64.rpm
glibc-headers-2.7-2.x86_64.rpm
glibc-devel-2.7-2.x86_64.rpm
glibc-2.7-2.x86_64.rpm
command:
rpm -Uvh --aid --nodeps glibc-common-2.7-2.x86_64.rpm
rpm -Uvh --aid --nodeps glibc-headers-2.7-2.x86_64.rpm
rpm -Uvh --aid --nodeps glibc-devel-2.7-2.x86_64.rpm
rpm -Uvh --aid --nodeps glibc-2.7-2.x86_64.rpm