In CMake, we can use find_dependency() in an package -config.cmake file to "forwards the correct parameters for QUIET and REQUIRED which were passed to the original find_package() call." So, naturally we'll want to do that instead of calling find_package() in such files.
Also, for dependency on a threads library, CMake offers us the FindThreads module, so that we write include(FindThreads), prepended by some preference commands, and get a bunch of interesting variables set. So, that's preferable to find_package(Threads).
And thus we have a dilemma: What to put in -config.cmake files, for a threads library dependency? The former, or the latter?
Following a discussion in comments with #Tsyarev, it seems that:
find_package(Threads) includes the FindThreads module internally.
... which means it "respects" the preference variables affecting FindThreads behavioe.
so it makes sense, functionally and aesthetically, to just use find_package() in your main CMakeLists.txt and find_dependency() in -config.cmake.
How can I specify where my libs live when building examples, or even writing new projects that use rust-sfml? I currently have both SFML and CSFML installed in /usr/local/{include,lib,share} as opposed to /usr/{include,lib,share}.
Is there a flag in cargo I can apply that will set the include/library path accordingly? Something like cargo build --lib=/usr/local/lib --include=/usr/local/include.
here is the rust-sfml page.
Each library that links against C libraries needs to support the appropriate linker parameters, including finding the library itself. It appears that there is an open pull request that adds support for specifying the install path to rust-sfml.
Specifically, Cargo allows specifying a build script. The build script communicates to Cargo by printing items to standard out. Two of those values that are interesting in this case are:
rustc-link-lib— indicates that the specified value should be passed to the compiler as a -l flag.
rustc-link-search — indicates the specified value should be passed to the compiler as a -L flag.
In Autoconf, when you want to check for the existence of some header, you do:
AC_CHECK_HEADERS([foo.h])
Now, let's suppose that I have some package, sponge, of which pkg-config knows, and I want to find out if it has a header named spongefoo.h.
I can do the following:
PKG_CHECK_MODULES([SPONGE])
CPPFLAGS="$CPPFLAGS $SPONGE_CFLAGS"
AC_CHECK_HEADERS([spongefoo.h])
This would work, but the Autoconf/Automake documentation says (my additions in brackets):
"Sometimes package developers are tempted to set user variables such as 'CFLAGS' [and, I add, 'CPPFLAGS'] because it appears to make their job easier. However, the package itself should never set a user variable, [...] To get around this problem, Automake introduces an automake-specific shadow variable [named AM_CPPFLAGS]"
However, AC_CHECK_HEADERS() and other test macros don't know about Automake's shadow variables.
So how do I properly check that a package has some header?
(BTW, AC_CHECK_HEADER() gets a 4'th argument, "INCLUDES", but this is a verbatim '"#include"' text, not a '"-I..."' switch, so it doesn't help me much.)
The requirement to temporarily set a user variable like CPPFLAGS to some value and run a compilation test while preserving the original value is actually done quite often. This is usually accomplished in Autoconf using a pattern similar to the following:
ac_save_CPPFLAGS=$CPPFLAGS
CPPFLAGS="$CPPFLAGS $SPONGE_CFLAGS"
AC_CHECK_HEADERS([spongefoo.h])
CPPFLAGS=$ac_save_CPPFLAGS
Note that the ac_ shell variable prefix is informally reserved for Autoconf itself, you might want to use your own variable name like proj_save_CPPFLAGS.
We using an automated build system which downloads and compiles source. The only interface I have to control the behaviour of the compilation is by setting ENV VARs and the arguments given to `./configure'.
The issue is that the 'configure' script (of the particular source I'm compiling) checks for a system header file, which if found, adversely affects the compilation process. (the compilation process will avoid compiling libraries which it believes are already installed on the local system when the above mentioned system header file is found.)
Since this is an automated process, I cannot modify the 'configure' script in anyway, and as mentioned can only specify the environment variables and arguments passed to `configure'. The configure script uses the AC_CHECK_HEADERS macro to generate the code to do the check for the system file. Is there anyway to avoid a check of a specific system file from the configure arguments?
The troublesome header file is in the path /usr/include/pcap/.
Thanks
Well there's a few things you could try:
remove foo.h from AC_CHECK_HEADERS and always build the library
use AC_CHECK_HEADER for foo.h and check for /usr/include/pcap/foo.h and don't AC_DEFINE(HAVE_FOO_H) if /usr/include/pcap/foo.h is there.
you could use AC_ARG_ENABLE or AC_ARG_WITH to turn off the offending test on a host-by-host basis via arguments to configure. So the answer to that question is yes.
All of these assume you can modify configure.ac and regenerate configure. If you can't do that you might have to modify configure (in an automated fashion, of course).
I need to conditionally compile some code based on the presence of a library. Seems like this should be easy with autoconf/automake but I can't figure it out.
For example, if there is a PNG library present, I want to include code to use it. My configure.ac has:
AC_CHECK_LIB([png], [png_create_write_struct_2])
and my Makefile.am has:
if USE_LIBPNG
libdev_la_SOURCES += png.c
endif
(which adds png.c to the list of sources for libdev so it gets compiled).
An automake conditional like USE_LIBPNG requires the conditional be defined in configure.ac, so i need:
AM_CONDITIONAL([USE_LIBPNG], [test SOMETHINGOROTHER])
The question is, what can test SOMETHINGOROTHER be? What does AC_CHECK_LIB define that I can test for?
AC_CHECK_LIB's default behavior is to define a symbol (in config.h) which can be used in source code, but that doesn't help the Makefile since the AM_CONDITIONAL needs a shell test
I tried overriding the default AC_CHECK_LIB behavior like so:
AC_CHECK_LIB([png], [png_create_write_struct_2], [HAS_LIBPNG=1])
after which I could test for it:
AM_CONDITIONAL([USE_LIBPNG], [test "x$HAS_LIBPNG" = "x1"])
This is ugly, but works for the Makefile... but creates a new problem: since it discards the original AC_CHECK_LIB behavior, and I no longer get a symbol added to config.h, which I need.
I must be missing something basic, or possible Doing It Wrong. Have been digging around for hours and found no answer.
Anyone?
If the library you're checking for supplies a .pc file for use with pkg-config, then you're much better off using PKG_CHECK_MODULES to get the correct flags. libpng does:
(in configure.ac)
PKG_CHECK_MODULES([libpng], [libpng12])
This gives you access to the variables $(libpng_CFLAGS) and $(libpng_LIBS) which you will want to add to Makefile.am (probably in AM_CFLAGS/AM_CXXFLAGS and LDADD, or target-specific versions thereof).
It will also cause configure to fail with an error if libpng12.pc isn't found. If you want configure to continue, you'll need to supply the third and fourth arguments to PKG_CHECK_MODULES, which are ACTION-IF-FOUND and ACTION-IF-NOT-FOUND:
(in configure.ac)
PKG_CHECK_MODULES([libpng], [libpng12], [HAVE_LIBPNG=1], [HAVE_LIBPNG=0])
Now, if you need an automake conditional, you can do something like:
(in configure.ac)
AM_CONDITIONAL([USE_LIBPNG], [test "$HAVE_LIBPNG" -eq 1])
If you also need the preprocessor definition, you could use AC_DEFINE like so:
(in configure.ac)
AS_IF([test "$USE_LIBPNG" -eq 1], [AC_DEFINE([USE_LIBPNG], [1], [Define if using libpng.])])
Possibly nicer is to set the definition in Makefile.am:
(in Makefile.am)
AM_CPPFLAGS =
if USE_LIBPNG
AM_CPPFLAGS += -DUSE_LIBPNG
endif
This will clutter your command line, though, whereas AC_DEFINE can put the definition in a header if you use AC_CONFIG_HEADERS. I guess this doesn't really matter if you use AM_SILENT_RULES([yes]) or don't care about your command line being neat (and let's be honest, automake generates some pretty gnarly command lines anyway).
A note on good autoconf style
It is considered poor form to build optional support based on whether or not a check succeeded (see this gentoo doc for details). Here's how I'd code optional support for libpng:
(in configure.ac)
# This is because the first PKG_CHECK_MODULES call is inside a conditional.
PKG_PROG_PKG_CONFIG
AC_ARG_WITH([libpng],
[AS_HELP_STRING([--with-libpng],
[support handling png files #<:#default=check#:>#])],
[],
[with_libpng=check])
AS_CASE(["$with_libpng"],
[yes], [PKG_CHECK_MODULES([libpng], [libpng12], [HAVE_LIBPNG=1])],
[no], [],
[PKG_CHECK_MODULES([libpng], [libpng12], [HAVE_LIBPNG=1], [HAVE_LIBPNG=0])])
AM_CONDITIONAL([USE_LIBPNG], [test "$with_libpng" != no -a "$HAVE_LIBPNG" -eq 1])
(in Makefile.am)
if USE_LIBPNG
AM_CPPFLAGS += -DUSE_LIBPNG
AM_CFLAGS += $(libpng_CFLAGS)
LDADD += $(libpng_LIBS)
libdev_la_SOURCES += png.c
endif
If your library doesn't have a .pc file
For completeness, here's how I'd check for a library that didn't have a .pc file. I'll skip over the details of following good autoconf style. AC_CHECK_LIB sets a cache variable, so you can test that instead of replacing the ACTION-IF-FOUND of AC_CHECK_LIB:
(in configure.ac)
AC_CHECK_LIB([png], [png_create_write_struct_2])
# Then test:
AS_IF([test "$ac_cv_lib_png_png_create_write_struct_2" = yes], [HAVE_LIBPNG=1], [HAVE_LIBPNG=0])
# Or set conditional:
AM_CONDITIONAL([USE_LIBPNG], [test "$ac_cv_lib_png_png_create_write_struct_2" = yes])
IMHO, you should only do it this way if you have no other option.
I will disagree mildly with Jack on his recommendation to use PKG_CHECK_MODULES. It is probably best to avoid using that. But I will agree with Jack on avoiding assigning to LIBS in the 3rd argument to AC_CHECK_LIB. Life is easier if you let AC_CHECK_LIB use the default settings.
Although AC_CHECK_LIB does not define a shell variable indicating whether or not the library was found, you can do this in configure.ac:
AM_CONDITIONAL([USE_LIBPNG],[grep HAVE_LIBPNG confdefs.h > /dev/null])
Arguably, this is relying on internal autoconf details, but in practice will work reliably.
Thanks for the replies.
Jack: I'm trying for maximum portability, so can't assume the libraries were installed as part of a package (they're not on my own box!), which means the no-other-option solution you suggested is what I already tried-- setting a shell variable manually-- but also manually performs the extra steps that would have been done by AC_CHECK_LIB: prepending the library to LIBS and defining HAVE_LIBxxx.
There was a catch though: autoheader complains about the bare AC_DEFINE:
autoheader: warning: missing template: HAVE_LIBPNG
autoheader: Use AC_DEFINE([HAVE_LIBPNG], [], [Description])
I'd be nice if autoheader worked in the future, so I had to change AC_DEFINE to the full monty:
AC_CHECK_LIB([png], [png_create_write_struct_2],
[HAS_LIBPNG=1
LIBS="-lpng $LIBS"
AC_DEFINE([HAVE_LIBPNG], 1, [Define to 1 if you have the `png' library (-lpng)])])
This works, but I don't much like having to duplicate the default behavior of AC_CHECK_LIB.
William: Yes I could grep for the symbol definition in confdefs.h, that also works.
Both solutions have their pros and cons (what doesn't?). Not sure which way I'll go, but it's nice to have options.
Thanks again.