- the Savannah source URI doesn't have the 1.0.4 release; switch it to
linphone.org
- even with the source available the build fails, as the source now
defaults to cmake for building. So switch our build process from
autotools to cmake as well.
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
- add CROSS_COMPILE and OS to MAKE_FLAGS; this should help when
compiling on other OSes, e.g. Darwin.
- move RELEASE up (alphabetical sort)
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
On most mipsel targets re fails to build because the re Makefile adds
"-march=mips32" to CFLAGS:
Assembler messages:
Error: -mips32r2 conflicts with the other architecture options, which imply -mips32
src/sip/addr.c:1:0: error: '-mips32r2' conflicts with the other architecture options, which specify a mips32 processor
/**
^
Makefile:92: recipe for target 'build-mipsel/sip/addr.o' failed
Fix this by preventing re from adding to OpenWrt CFLAGS.
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
re and baresip builds for 32bit x86 targets are currently failing:
build-i386/sip/auth.o: In function `sip_auth_encode':
/build/lede-snapshots/i386_pentium/build/sdk/build_dir/target-i386_pentium_musl/re-0.5.6/src/sip/auth.c:279: undefined reference to `__stack_chk_fail_local'
The undefined references occur because there is no linking done to ssp.
Fix this by setting LD to TARGET_CC, so ld is not called directly. This
also means one patch is not needed anymore.
Do the same for rem and restund as they basically all share the same
build system.
Compile-tested for x86, x86_64 and mips.
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
Change from custom make call to using the default Build/Compile routine.
This applies the standard memory corruption mitigation methods, e.g.
FORTIFY_SOURCE and RELRO, to the package if they are enabled (default).
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
Using the CFLAGS from the environment of Build/Compile the optimizations
from libre now appear after OpenWrt CFLAGS, overriding them. Prevent
that by disabling libre's optimization altogether.
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
When using the default Build/Compile routine $LD is set to the actual
linker, so libre's build system is not using ld indirectly via the
compiler anymore. But it feeds -rdynamic to ld instead of
--export-dynamic. Fix that with this patch.
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
With FORTIFY_SOURCE enabled the compiler errors our when libre defines
ssize_t again. Prevent that by adding the proper hint to the header in
question.
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
Remove unneeded variables, whitespaces and trailing slashes, use
specific INSTALL directive instead of "$(CP)".
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
Change from custom make call to using the default Build/Compile routine.
This applies the standard memory corruption mitigation methods, e.g.
FORTIFY_SOURCE and RELRO to the package, if they are enabled (default).
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
- Update from 2.6 to 2.7.1.
- Update sed script because the TARGET_CFLAGS now occur twice in the .pc
file.
- Upgrade dependency from libsrtp to libsrtp2.
- Drop patches that were included upstream.
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
- Currently iksemel doesn't recognize gnutls anymore. Fix that by
substituting the currently used patches with one that also Debian is
using. It allows gnutls detection via pkgconfig.
- Add another patch Debian is using to enable secure gnutls options.
- Update project URL.
- Remove unneeded flags and Build/Prepare customizations.
- Cleanup DEPENDS.
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
Plain git:// doesn't add security. Use https:// instead. See
https://gist.github.com/grawity/4392747.
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
The dependencies in the default section were modified instead of just
the ones for dahdi-cfg. Fix that.
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
- Remove some unnecessary trailing slashes plus fix a whitespace.
- Remove flags that do not seem to have a good reason to be there
(anymore).
- Add sed script to remove $(TARGET_CFLAGS) from pkgconfig file. This
gets rid of the warnings in asterisk13 about COPTS duplication.
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
Copied from Asterisk, sets some sane values. For instance it enables
IPv6 support.
Also it disables DEBUG. With debug enabled 'pjproject enables "assert"
functions which can cause Asterisk to crash unexpectedly' (quote from
Asterisk wiki). This probably fixes issue #140.
With this config_site.h in place the libsrtp dep can be dropped from
select sub packages.
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
- Make "--disable-floating-point" depend on SOFT_FLOAT
- Remove "--disable-ilbc-coder" as configure no longer recognizes it.
- Align args with the ones used by Asterisk (for their bundled pjproject);
this includes adding "--disable-resample".
- With libresample removed remove the dep from asterisk13 accordingly.
- Note: the disabled items are _not_ used by Asterisk anyway.
- Drop speex dep as pjproject does not link against it the way it is
configured.
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
Add patches provided by Asterisk project for pjproject. This fixes the
following vulnerabilities:
- AST-2017-002: Buffer Overrun in PJSIP transaction layer (CVE-2017-9372)
- AST-2017-003: Crash in PJSIP multi-part body parser
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
When opus is found in staging area it will be linked against, adding a
dep that is not in the Makefile. So explicity disable opus.
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
On a Netgear R7000 (ARMv7) without floating point, using SpanDSP with
fixed point to send/receive fax on Asterisk resulted in repeated
failure and intermittent performance. Disabling fixed point fixed
these issues; the software floating point implementation was fine for
real-time use.
Now on a Netgear R7800 (ARMv7) with hardware floating point, the
autoconf auto-detect will say no floating point is available, and
again enables the poor performing fixed point implementation. Ouch!
