In order to avoid confusion of the flags-gathering pkg-config
result with the actual test for the availability of "check",
change the pkg-config output variable prefix from CHECK to TESTS.
Put the extra libraries into SR_EXTRA_LIBS instead of LIBS.
Create an SR_CHECK_LIBS macro to make that easy. Substitute
SR_EXTRA_LIBS into libsigrok.pc, too.
Place a copy of ax_cxx_compile_stdcxx_11.m4 from the Autoconf
macro archive into our private m4/ directory. This is cleaner
than trying to parse M4 file versions etc. Plus, the macro is
now always available.
Introduce the SR_CHECK_COMPILE_FLAGS macro and use it to check
for additional compiler flags. Put the accepted flags into the
separate substitution variable SR_EXTRA_CFLAGS.
With this and the preceding changes, bug #578 should now be fixed.
Use the SR_ARG_ENABLE_WARNINGS macro to configure and check for
the availability of compiler warning flags. Maintain separate
sets of warning flags for C and C++.
The configure option --enable-warnings=[min|max|fatal|no] can
be used to set the compiler warning level for all languages.
The default level is "max".
Place custom Autoconf macros which other sigrok modules may
re-use into a separate file m4/sigrok.m4. Also, introduce new
macros for defining the package and library versions, and for
gathering compiler warning flags.
It seems that contrary to what the documentation says, leaving
the category argument to m4_warn() empty is not allowed. Use
the "unsupported" category for lack of a better choice.
It can sometimes happen that .git/HEAD or .git/refs/head/*, which
are added as config.status dependencies during configure, do not
exist anymore at build time. For instance, when the current branch
is deleted after switching to a different one.
Wrap the dependencies inside $(wildcard ...) to avoid this problem.
Note that this is a GNU make feature. However, it should be fine
as it is only used for git builds. Even if a non-GNU make is used,
the construct will hopefully just expand to nothing.
Replace DRIVER() and DRIVER2() by a single SR_DRIVER() macro.
Derive the names of shell variables and preprocessor defines
programatically to cut down on repetition.
It seems automake automatically adds the directory containing
the generated version.h to the include path. Use nostdinc to
disable default includes altogether.
Append the git revision hash to the libsigrok package version,
unless HEAD exactly matches a release tag. Note that this does
not affect the version known to autoconf -- e.g. source tarballs
created by make dist will not receive a revision suffix.
Changes to git HEAD automatically trigger a reconfiguration.
Uncommitted changes do not, which is why I left out the -dirty
suffix.
Make it so that $(datadir) is resolved at make time, as per
autotools recommendations. Note that $datadir is not fully
resolved at configure time to begin with, i.e. part of it
already was evaluated at make time.
Use the proper tool for the job and make libsigrok/version.h
a secondary configuration header, so that autoconf's AC_DEFINE
machinery can be used to generate it. Note that the header
template is still hand-written, enabling fine control of the
content.
Move the include flags for files in the source tree from
configure.ac to Makefile.am where they belong. Also use
AM_CPPFLAGS instead of CFLAGS/CXXFLAGS to make sure the
files in the build/source tree are always picked up first.
Also, remove the include/libsigrok sub-directory from the
search path, thereby making the <libsigrok/> prefix mandatory
when building libsigrok itself. This matches the convention
already imposed on users of the library.
With gcc 5.1 released and defaulting to std=gnu11, the code will be compiled
according to different standards depending on the compiler version so we
should better specify explicitly what standard we are targetting.
C11 is now quite mature, it is supported in the just release Debian stable
(gcc 4.9) and also in old-stable (gcc 4.7), so there should be no reason to
use anything more ancient.
We also should have no reason to need any non-standard GNU extension.
So using only C11 + POSIX sounds like the best option right now.
The driver for BayLibre ACME depends on Linux-specific sysfs
interfaces to ina226 and tmp435 devices. Exclude it for different
targets.
Signed-off-by: Bartosz Golaszewski <bgolaszewski@baylibre.com>
Daniel Elstner
Daniel Glöckner
Uwe Hermann
Aurelien Jacobs
Vladislav Ivanov
Bartosz Golaszewski