This works around the build failure with gcc-10 until there is a patch
for dahdi-tools that is reviewed and acknowledged by upstream.
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
- package two new modules: dlgs and pv_headers
- upstream actually provides three more new modules, but two are missing
dependencies in OpenWrt and one is related to systemd
- use INSTALL_DATA instead of INSTALL_CONF for kamailio configuration
files (not talking about /etc/config/kamailio) as they're default
configuration
- refresh 050-fix-kamailio-utils.patch
- drop upstreamed 150-python3-cross-compile.patch
- libsrutils is no longer packaged as upstream moved it into the core
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
No need to have this symbol around here. libfreetdm is part of the same
package (freeswitch) anyway.
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
A small but important oversight crept into the last asterisk-chan-dongle
update. The iconv library detection was changed to look for iconv()
instead of libiconv().
Reminder: the autoconf macros for finding libraries do _not_ include any
extra headers when testing the compile. So the redefinitions done in
iconv.h provided by any libiconv (stub or regular) aren't visible during
the test the macros execute.
This means that when testing for iconv() the only iconv that can be
detected is the one from libc. Moreover, if the libc doesn't provide
iconv, the check will fail. Below example is with uclibc:
checking for iconv in -liconv... no
configure: error: iconv library missing
Makefile:87: recipe for target '/builder/shared-workdir/build/sdk/build_dir/target-arc_archs_uClibc/asterisk-chan-dongle-asterisk16/asterisk-chan-dongle-2020-05-28-328b2b7d/.configured_a17fb5ef857664f03cd0ce37cc5ea591' failed
So, this changes the patch to look for libiconv(), again.
Also, instead of using AC_CHECK_LIB this uses AC_SEARCH_LIBS, like
upstream does. There apparently was a problem with that in the past, but
now it works fine.
checking for rm... rm
checking for library containing libiconv... -liconv
checking for sqlite3_open in -lsqlite3... yes
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
Currently we disable some sofia tests to prevent build failure. An issue
was actually raised upstream ([1]) where this was discussed.
This cleared up some doubts. For instance it was established that
libcheck is used as a test framework, if the lib is detected. In OpenWrt
libcheck is available, so it may be in staging during the freeswitch
build.
Upstream also mentioned that libcheck is not always compatible with
freeswitch, which depends on the libcheck version.
This commit prevents libcheck detection. This is nicer than explicitly
disabling tests that fail.
[1] https://github.com/signalwire/freeswitch/issues/681
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
Bump dongle plugin to latest git revision. PKG_SOURCE_SUBDIR and
PKG_BUILD_DIR are removed because defaults are fine. PKG_VERSION is
replaced by PKG_SOURCE_DATE, which is in line with OpenWrt
recommendation.
Soft deps are removed as per request in issue #536.
iconv handling is updated. Upstream now lets us specify the include
directory, so that makes the patch simpler. The iconv lib detection is
also updated so that it never picks the libc iconv. This way we get the
same iconv handling independent of the libc in use. Now it'll be always
iconv-stub or iconv-full.
libsqlite3 dep is made explicit. There was always a dep on it, but it
was implicit via asterisk. When the dep is made explicit, the dongle
package picks up on the sqlite3 ABI_VERSION, which is nice.
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
The rtpengine build system uses the Makefiles to generate source files
on the fly. This usually works fine, but when building in parallel,
intermittently the generated files contain garbage (like, lines are
repeated, causing redefinition errors).
As a workaround this commit disables parallel builds.
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
This commit follows upstream, how they include the Debian build system
flags, appending them to their own. To that end OpenWrt's flags are now
saved at the beginning of the Makefiles, and appended at the bottom of
lib/lib.Makefile.
This also sorts out the iptables module compile, to use OpenWrt's
LDFLAGS.
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
We have both liblua and liblua5.3 in $(STAGING_DIR)/usr. But the headers
for liblua5.3 are in (STAGING_DIR)/usr/include/lua5.3. The way things
are now, Asterisk uses the headers from liblua and links against
liblua5.3.
The easiest way out of this is to turn off the lua detection that is
based on version numbers. The next detection looks for a version-less
lua, which works fine.
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
This commit changes handling of configuration and init a bit:
1. Asterisk configuration files are installed with $(INSTALL_DATA). This
way there is no need to chown /etc/asterisk to user asterisk. The
package provides the standard asterisk configuration (no secrets) so
it's OK to install it readable for all. If users put something
sensitive in there they can update permissions how they like.
2. The init script no longer creates directories that the users can
configure in /etc/config/asterisk, which could be all kinds of
strange directories. Instead only the default directories are
created, and only the ones that reside on volatile memory. The other
directories are now created by the package itself.
3. This commit removes the ability to choose a user/group via
/etc/config/asterisk. This makes the init script simpler. If anybody
wants to play around with the user/group, then it's up to them to fix
the permissions.
4. "local" is removed because it's not POSIX.
5. From issue #520 we know that adding the user to the dialout group
does not help with the device permissions for chan_lantiq. This
removes it again.
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
Patches refreshed, no longer needed patches are dropped.
This adds directory "/usr/share/asterisk/firmware/iax" to silence a
run-time warning.
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
Upstream commit 2dced93 "[sofia-sip] Build internal tests" enabled the
building of tests. One of them causes build failures on the build bots:
Making all in s2check
make[11]: Entering directory '/builder/shared-workdir/build/sdk/build_dir/target-arm_cortex-a15+neon-vfpv4_musl_eabi/freeswitch-1.10.3.-release/libs/sofia-sip/s2check'
COMPILE s2tcase.o
s2tcase.c: In function 's2_tcase_add_test':
s2tcase.c:60:27: error: passing argument 2 of '_tcase_add_test' from incompatible pointer type [-Werror=incompatible-pointer-types]
_tcase_add_test(tc, tf, name, signo, 0, start, end);
^~
In file included from s2tcase.c:35:
/builder/shared-workdir/build/sdk/staging_dir/target-arm_cortex-a15+neon-vfpv4_musl_eabi/usr/include/check.h:331:69: note: expected 'const TTest *' {aka 'const struct TTest *'} but argument is of type 'TFun' {aka 'void (*)(int)'}
CK_DLL_EXP void CK_EXPORT _tcase_add_test(TCase * tc, const TTest * ttest,
~~~~~~~~~~~~~~^~~~~
s2tcase.c:60:31: error: passing argument 3 of '_tcase_add_test' makes integer from pointer without a cast [-Werror=int-conversion]
_tcase_add_test(tc, tf, name, signo, 0, start, end);
^~~~
In file included from s2tcase.c:35:
/builder/shared-workdir/build/sdk/staging_dir/target-arm_cortex-a15+neon-vfpv4_musl_eabi/usr/include/check.h:332:47: note: expected 'int' but argument is of type 'const char *'
int _signal, int allowed_exit_value,
~~~~^~~~~~~
s2tcase.c:60:7: error: too many arguments to function '_tcase_add_test'
_tcase_add_test(tc, tf, name, signo, 0, start, end);
^~~~~~~~~~~~~~~
In file included from s2tcase.c:35:
/builder/shared-workdir/build/sdk/staging_dir/target-arm_cortex-a15+neon-vfpv4_musl_eabi/usr/include/check.h:331:27: note: declared here
CK_DLL_EXP void CK_EXPORT _tcase_add_test(TCase * tc, const TTest * ttest,
^~~~~~~~~~~~~~~
cc1: all warnings being treated as errors
I can't reproduce this at home. Not with a master checkout, not with a
current SDK. But the tests are not needed for the OpenWrt packages
anyway, so this commit disables them again.
No rev bump is done because this has no effect on the packages.
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
