Reputation: 6245
My main development platform is Gentoo on Linux. However, recently I tried to build my program on the fresh VM install on Debian.
My program contains of main binary and couple of dll/so/dylib libraries. One of the libraries depends on the presence of unixODBC/iODBC.
I was told by unixODBC maintainers to use odbc_config
script to identify the build parameters.
When I build on Gentoo - everything works fine. There is no problems.
However, when I build on Debian - the build fails because apparently Debian does not produce odbc_config script and instead in this case rely on pkg-config.
So, I need to add a test in configure.ac to check for odbc_config script presence and pass it along to one of the so files generation (lets call it libodbc_lib
project).
Could someone please help me with this?
EDIT:
Is this correct to be put in configure.ac:
AC_CHECK_PROG(ODBC,odbc_config,yes)
if test x"${ODBC}" == x"yes" ; then
ODBC_CFLAGS = `odbc_config --cflags`
ODBC_LIBS = `odbc_config --libs` -lodbcinst
else
ODBC_CFLAGS = `pkg-config odbc --cflags`
ODBC_LIBS = `pkg-config odbc --libs` -lodbcinst
fi
AC_SUBST(ODBC_CFLAGS)
AC_SUBST(ODBC_LIBS)
If it is - how do I use ODBC_FLAGS/ODBC_LIBS in my subproject?
EDIT2:
Based on this answer I used the following code:
In the main configure.ac:
AC_CHECK_PROG(ODBC,odbc_config,yes)
if test x"${ODBC}" == x"yes" ; then
ODBC_CFLAGS = `odbc_config --cflags`
ODBC_LIBS = `odbc_config --libs` -lodbcinst
else
ODBC_CFLAGS = `pkg-config odbc --cflags`
ODBC_LIBS = `pkg-config odbc --libs` -lodbcinst
fi
AC_SUBST(ODBC_CFLAGS)
AC_SUBST(ODBC_LIBS)
In the libodbc_lib/Makefile.am:
libodbc_lib_la_CXXFLAGS = -I../../dbinterface \
-DUNICODE \
-DUNIXODBC \
-I@ODBC_CFLAGS@
libodbc_lib_la_LDFLAGS = -L../dbinterface \
-ldbinterface \
@ODBC_LIB@
I regenerated configure, run it successfully and then tried running make
.
I got following error:
CXXLD libodbc_lib.la
/usr/lib/gcc/x86_64-pc-linux-gnu/11.3.0/../../../../x86_64-pc-linux-gnu/bin/ld: cannot find @ODBC_LIB@: No such file or directory
What I did wrong?
EDIT3:
After fixing the missing S
, I got following compile commands:
make[2]: Entering directory '/home/igor/dbhandler/Debug/libodbc'
/bin/sh ../libtool --tag=CXX --mode=compile g++ -DHAVE_CONFIG_H -I. -I/home/igor/dbhandler/libodbc -I.. -I../../dbinterface -DUNICODE -DUNIXODBC -I@IODBC_CFLAGS@ -g -O0 -MT libodbc_lib_la-database_odbc.lo -MD -MP -MF .deps/libodbc_lib_la-database_odbc.Tpo -c -o libodbc_lib_la-database_odbc.lo `test -f 'database_odbc.cpp' || echo '/home/igor/dbhandler/libodbc/'`database_odbc.cpp
libtool: compile: g++ -DHAVE_CONFIG_H -I. -I/home/igor/dbhandler/libodbc -I.. -I../../dbinterface -DUNICODE -DUNIXODBC -I@IODBC_CFLAGS@ -g -O0 -MT libodbc_lib_la-database_odbc.lo -MD -MP -MF .deps/libodbc_lib_la-database_odbc.Tpo -c /home/igor/dbhandler/libodbc/database_odbc.cpp -fPIC -DPIC -o .libs/libodbc_lib_la-database_odbc.o
libtool: compile: g++ -DHAVE_CONFIG_H -I. -I/home/igor/dbhandler/libodbc -I.. -I../../dbinterface -DUNICODE -DUNIXODBC -I@IODBC_CFLAGS@ -g -O0 -MT libodbc_lib_la-database_odbc.lo -MD -MP -MF .deps/libodbc_lib_la-database_odbc.Tpo -c /home/igor/dbhandler/libodbc/database_odbc.cpp -o libodbc_lib_la-database_odbc.o >/dev/null 2>&1
mv -f .deps/libodbc_lib_la-database_odbc.Tpo .deps/libodbc_lib_la-database_odbc.Plo
/bin/sh ../libtool --tag=CXX --mode=link g++ -I../../dbinterface -DUNICODE -DUNIXODBC -I@IODBC_CFLAGS@ -g -O0 -L../dbinterface -ldbinterface -o libodbc_lib.la -rpath /usr/local/lib libodbc_lib_la-database_odbc.lo
libtool: link: g++ -fPIC -DPIC -shared -nostdlib /usr/lib/gcc/x86_64-pc-linux-gnu/11.3.0/../../../../lib64/crti.o /usr/lib/gcc/x86_64-pc-linux-gnu/11.3.0/crtbeginS.o .libs/libodbc_lib_la-database_odbc.o -L../dbinterface -ldbinterface -L/usr/lib/gcc/x86_64-pc-linux-gnu/11.3.0 -L/usr/lib/gcc/x86_64-pc-linux-gnu/11.3.0/../../../../lib64 -L/lib/../lib64 -L/usr/lib/../lib64 -L/usr/lib/gcc/x86_64-pc-linux-gnu/11.3.0/../../../../x86_64-pc-linux-gnu/lib -L/usr/lib/gcc/x86_64-pc-linux-gnu/11.3.0/../../.. -lstdc++ -lm -lc -lgcc_s /usr/lib/gcc/x86_64-pc-linux-gnu/11.3.0/crtendS.o /usr/lib/gcc/x86_64-pc-linux-gnu/11.3.0/../../../../lib64/crtn.o -g -O0 -Wl,-soname -Wl,libodbc_lib.so.0 -o .libs/libodbc_lib.so.0.0.0
libtool: link: (cd ".libs" && rm -f "libodbc_lib.so.0" && ln -s "libodbc_lib.so.0.0.0" "libodbc_lib.so.0")
libtool: link: (cd ".libs" && rm -f "libodbc_lib.so" && ln -s "libodbc_lib.so.0.0.0" "libodbc_lib.so")
libtool: link: ar cru .libs/libodbc_lib.a libodbc_lib_la-database_odbc.o
libtool: link: ranlib .libs/libodbc_lib.a
libtool: link: ( cd ".libs" && rm -f "libodbc_lib.la" && ln -s "../libodbc_lib.la" "libodbc_lib.la" )
make[2]: Leaving directory '/home/igor/dbhandler/Debug/libodbc'
I still the variable name there and not their values.
Is it normal?
Upvotes: 0
Views: 483
Reputation: 37747
As UnixODBC upstream does ship and install *.pc
files, I would expect that file to be both present and correct and therefore I would ignore any *-config
scripts. The pkg-config system is quite well thought out and works even for quite weird cross compilation environments. The *.pc
mechanism works well on Linux, on FreeBSD, on OSX, cross-compiling for Windows on Linux, to name a few.
A well-written _config
program written in portable shell could do the same by basically reproducing much of the pkg-config
logic in portable shell for each and every _config
script, hopefully correctly.
However, odbc_config
is not a portable shell script. It is a binary executable, i.e. it will regularly break for cross-compiling, as the system you build on will usually not be able to run programs like odbc_config
which are built to run on the system you are building for.
And even if the flags from the *.pc
files were unsuitable for a very unusual build environment: Using PKG_CHECK_MODULES
defines appropriate _CFLAGS
and _LIBS
variables for the configure script, so even in a very unusual build environments one can always override whatever the *.pc
file might contain by calling configure
like
../configure ODBC_CFLAGS='-I/weird/stuff -DWEIRD_STRING="moo"' ODBC_LIBS='-L/very/weird/libxyz -lodbc'
So... using odbc_config
has no advantages, upstream already provides a odbc.pc
file so it is always present, so why not just always use odbc.pc
?
So, in configure.ac
(if builds without odbc.pc
present should fail, otherwise you will have to do some AC_DEFINE
and/or AM_CONDITIONAL
to conditionally build with or without ODBC support) do
m4_pattern_forbid([PKG_CHECK_MODULES])dnl
PKG_CHECK_MODULES([ODBC], [odbc])
and in any subdirectory (what you call "subproject") Makefile.am
or Makefile-files
where you need to link somehting against libodbc
, put, depending on whether you are building an executable
bin_PROGRAMS += foobar
[…]
foobar_CPPFLAGS += $(ODBC_CFLAGS)
foobar_LDADD += $(ODBC_LIBS)
or a (libtool) library
lib_LTLIBRARIES += libfoo.la
[…]
libfoo_la_CPPFLAGS += $(ODBC_CFLAGS)
libfoo_la_LIBADD += $(ODBC_LIBS)
That should work for all native and cross-compile builds in properly set up build environments, and people can still override odbc_CFLAGS
and odbc_LIBS
in case of problems.
Of course, you can always AC_CHECK_PROG
or AC_PATH_PROG
or AC_CHECK_TOOL
or AC_PATH_TOOL
together with an AC_ARG_VAR
for the odbc_config
program and then define and AC_SUBST
an _CFLAGS
and _LIBS
variable set to the output of $ODBC_CONFIG --cflags
and $ODBC_CONFIG --libs
, respectively, and then then use the _CFLAGS
and _LIBS
vars in Makefile.am
/Makefile-files
as above.
However, that is a lot of code to write, and with a lot of special cases to consider, and if you have to ask about how to do this you will probably get a lot more wrong than if you just just use PKG_CHECK_MODULES
.
You can always add something later if the PKG_CHECK_MODULES
route actually does not work for a use case and which cannot be fixed within the pkg-config framework. Until that time (if it ever happens), I would recommend to just use the simple PKG_CHECK_MODULES
method and probably be done.
Upvotes: 1
Reputation: 180048
So, I need to add a test in configure.ac to check for odbc_config script presence and pass it along to one of the so files generation
Autoconf has AC_PATH_PROG()
for checking for a program in the executable search path. You would of course use AC_SUBST()
to define one or more output variables by which to convey the results to the generated makefiles.
But no, coming back around to my comment on the answer to one of your previous questions, what you ought to do is not have configure
forward information about the executable, but rather for it to determine the needed flags itself and forward them, via one or more output variables. If you continue to use odbc_config
, at least conditionally, then that means having configure
run it and capture the output. You should not inject shell command substitutions into your compilation commands.
And if you substitute a different mechanism, whether conditionally or exclusively, then similarly for that. (That's what your other answer describes with respect to pkg-config
.)
Upvotes: 0