There is a key difference between glibc and musl when it comes to
ppc/ppc64 and that is that musl does not support 128-bit long IBM doubles format
it only supports 128-bit long double IEEE format on ppc64 alone.
this change ensures that we account for this change, so far we have been
doing it a bit wrong for ppc/musl case.
(From OE-Core rev: 28cb148c1375ba242addba80b0f68e06d2a4d874)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
libc_cv_ppc_machine and no longer exists in glibc
others are detected correctly in configure. They
perhaps made sense in past when the toolchain build has several
bootstrap stages
(From OE-Core rev: ba418f93fc586bb708fdfade4195f7f3be2e273f)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
recipe for libidl has long been deleted
(From OE-Core rev: 587d0b51b09cd426dc4b2b7646fa063d10ae4f2f)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The standard macros from gettext for iconv include problematic tests which
we've been patching out adhoc. Stop doing this and set results in the site
files instead which is simpler, more maintainable and peforms better too
as an added bonus.
(From OE-Core rev: 5a3bfdc4af18302cf0e3ea5802fdfefaa7235657)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
glibc 2.32 has deprecated sys_siglist [1]
since no is default for musl as well, therefore elevate it to common
site file for all
[1] https://sourceware.org/git/?p=glibc.git;a=commit;h=b1ccfc061feee9ce616444ded8e1cd5acf9fa97f
(From OE-Core rev: 7c19f30849e9a73a7497d8e6c4719a2e57017086)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
They are preventing autotools from checking netinet/sctp.h existence
successfuly which in turn makes configure.ac files with lines similar to
this to fail, even if DEPENDS="lksctp-tools" and netinet/sctp.h is
clearly available under recipe-sysroot/:
AC_CHECK_HEADERS(netinet/sctp.h,,AC_MSG_ERROR(netinet/sctp.h not found))
This is the case for new libosmo-netif versionsi (meta-telephony branch
laforge/nightly), which expects netinet/sctp.h to be available if default
flag --enable-libsctp is used.
Without this patch, do_configure will fail unless following line is used
in the recipe:
CACHED_CONFIGUREVARS += "${@bb.utils.contains('PACKAGECONFIG', 'libsctp', 'ac_cv_header_netinet_sctp_h=yes', '',d)}"
So removing ac_cv_header_netinet_sctp_h would be enough, but I couldn't
find any good information on why the other 2 variables were set in site
config files, hence no good reason to keep them.
(From OE-Core rev: 6a641e744507c8c17fa3d7273cb2e5b512ca0242)
Signed-off-by: Pau Espin Pedrol <pespin@sysmocom.de>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This coreutils (gnulib) test tests for various bugs that only
exist in ancient versions.
It defaults to assuming buggy behaviour with its own implementation
when cross-compiling.
musl and recent glibc (2.29) are not affected.
(From OE-Core rev: 65d38cc1ce5a106c4c2e5068b8440eb6e5a2b33e)
Signed-off-by: André Draszik <git@andred.net>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
autoconf has a test for strtod(), which assumes non-working
when cross-compiling, but it does work in both musl and
recent glibc.
coreutils (gnulib) does some additional tests on top of that,
but assumes working glibc when >= 2.8 when cross compiling.
It doesn't know about musl where the additional tests also
work, though.
(From OE-Core rev: c8429f707d9ac785f295492470ec8b9ff56c043b)
Signed-off-by: André Draszik <git@andred.net>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
As per coreutils' (gnulib's) autotools test, nanosleep()
misbehaves on glibc (2.29), and works fine on musl.
During cross-compile, recent coreutils assume brokenness
when compiling for linux, which pessimises musl.
Set the correct result for musl, and for coherency reasons,
also specify the result for glibc.
(From OE-Core rev: 4522648758dc59f5ece736a0c1c0e95dcc7dafd1)
Signed-off-by: André Draszik <git@andred.net>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This coreutils / gnulib autoconf test is for a broken glibc
implementation of utimes from 2003-07-12 to 2003-09-17.
(From OE-Core rev: a60ca6048cc3a4fef862b128a842c70ac0cd2253)
Signed-off-by: André Draszik <git@andred.net>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
to set the file to the current time
(From OE-Core rev: 530e01e79e9674d0049b183dc23491eb886b066b)
Signed-off-by: André Draszik <git@andred.net>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
recent coreutils (gnulib) assume yes when cross-compiling
for gnu (glibc), but don't know about musl.
For coherence, just set the result to yes for both.
Note that the old coreutils (from meta-gplv2) doesn't
assume anything and instead that recipe hard-codes to yes.
So behaviour with yocto when using meta-gplv2 is actually
better than when using the latest version (when using musl).
This patch rectifies this shortcoming.
(From OE-Core rev: 692fe85264e599eb659456bd2eebf5f12a1cd30f)
Signed-off-by: André Draszik <git@andred.net>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This coreutils (gnulib) test checks for an abort() that existed
in glibc before 2.4.90-10 (in 2006) in certain conditions.
Neither libraries exhibit this problem today.
(From OE-Core rev: 506c3b5adaa2fdffa051d83fb99efc00e432156b)
Signed-off-by: André Draszik <git@andred.net>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
In glibc, getcwd() handles long file names properly, on
musl, getcwd() only works up to PATH_MAX directory depths.
Configuring the autotools (gnulib) test result here allows
coreutils to compile more optimised code for both platforms,
rather than being pessimistic and re-implementing everything
itself.
The difference in behaviour is because both do the kernel
getcwd syscall (which only supports up to PATH_MAX), but
glibc implements fallbacks for longer paths, while musl
doesn't.
(From OE-Core rev: 525e33cf99983ee4bc3cf1822364123551aa7c83)
Signed-off-by: André Draszik <git@andred.net>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
I.e. allocate memory for the pointer returned when the first
argument is NULL.
(From OE-Core rev: fe148da36af2dc086e05e2ebc1c088f1b5485de2)
Signed-off-by: André Draszik <git@andred.net>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
I.e. shared text that is being executed.
(From OE-Core rev: 1370742d20aaa1a17f3e242ea9bb54df944b6416)
Signed-off-by: André Draszik <git@andred.net>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
calloc (N, S) returns non-NULL when N*S is zero,
and returns NULL when N*S overflows.
(From OE-Core rev: ba2bc4b6529013a303dd67abe944ceb619e7466c)
Signed-off-by: André Draszik <git@andred.net>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
their mkstemp() passes all tests from the gnulib m4 macro
gl_FUNC_MKSTEMP.
(From OE-Core rev: 2b0905d25fa295fc868bf6df1da89283b9dc1f45)
Signed-off-by: André Draszik <git@andred.net>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
I.e. a memcmp() that is 8bit clean (like glibc).
(From OE-Core rev: d938ca1b8bc715047d771907c6ea4a1c9aa6594f)
Signed-off-by: André Draszik <git@andred.net>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
I.e. a chown() that works when given -1 as uid / gid
(From OE-Core rev: 226b8b2b1b565ef019b442ad3bf1f5d86704e1d2)
Signed-off-by: André Draszik <git@andred.net>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Fixed when powerpc64:
$ bitbake <image> -ctestsdk
[snip]
checking whether printf supports %p... configure: error: cannot run test program while cross compiling
See `config.log' for more details.
test_cvs (oeqa.sdk.buildcvs.BuildCvsTest) ... FAIL
[snip]
The cvs_cv_func_printf_ptr is already in powerpc32-linux.
(From OE-Core rev: cbba73baf94cd0ee99d010abab79140cd5e4b99e)
Signed-off-by: Robert Yang <liezhi.yang@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
These variables force runtime tests during configure
they are already cached for other architectures
(From OE-Core rev: 7a0e456d9dcfe9d9d9a0dbd24a6083c8d40516ff)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
coreutils configure script needs to know whether /proc/uptime is
available, but this is not possible in a cross-compilation
context. This leads to an uptime program that fails to work on the
target, as it has been compiled without /proc/uptime reading support.
This commit fixes that by telling coreutils at configure time that
/proc/uptime will be available on the target (which seems to be a
reasonable assumption on Linux systems).
This commit is made with great inspiration from Thomas Petazzoni's
patch to buildroot to fix the same issue.
(From OE-Core rev: 199b661c0699eb31df5bc73f2e3dce6a2db4585d)
Signed-off-by: Sean Nyekjaer <sean.nyekjaer@prevas.dk>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The value was hardcoded from the time it couldn't be computed, which
is no longer the case. After C99 'bool' is only defined if stdbool.h
is included, it's implementation defined and not required to be 1, so
caching it doesn't make sense and certain recipes whoose code test
ac_cv_sizeof_bool fail to build.
(From OE-Core rev: 15af2d527d582ef181d6b9c042844aa89f991f0b)
Signed-off-by: Ioan-Adrian Ratiu <adrian.ratiu@ni.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
There are tools in the wild where we need this to be a full path, even
if that doesn't make sense in most other cases due to libtool issues.
Allow those cases to override the default value as currently its near
impossible to do so.
(From OE-Core rev: 65652f419a4c8578121f1f67d43f23ce4eae5a37)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This is the cache variable used by AC_FUNC_MMAP, which is possibly one of the
worst autoconf macros to exist.
Apart from being a runtime test which silently claims that mmap() is broken when
cross-compiling, this is basically to verify that mmap() actually works, because
SVR4.0 (released 1988) was broken. Thirty years later, everyone has a working
mmap().
common-glibc already has an assignment, so add a corresponding assignment to
common-musl and remove it from the machine-specific files.
(From OE-Core rev: 93dd7c87cef4fd9c22a09857fb55218c8be87b5b)
Signed-off-by: Ross Burton <ross.burton@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Add linux site info for the MicroBlaze architecture. This is based on
the site info from meta-xilinx and additional options based on config
options that other *-linux architecture site info files define.
(From OE-Core rev: def626a46efdeeaf867a27464aed463d265d777c)
Signed-off-by: Nathan Rossi <nathan@nathanrossi.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
With the commit afa9f769d62034d4443dfe929422d1d591adf709 some nios2 builds
(uboot, core-image-minimal, etc) were broken due to db trying to use ARM instructions
in mutexes. The reason was db "configure" used the cached entry from nios2-linux
(which was incorrect). So the remedy was to remove the incorrect cached entry
and let db "configure" figure out which is the proper mutex to use.
(From OE-Core rev: c17c6ba906425d4035b8e044c8bd8bd68c47ef74)
Signed-off-by: Juro Bystricky <juro.bystricky@intel.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Depending on whether large file support is enabled, the size of ino_t can
be 4 or 8 on 32 bit x86. Drop the value from the site cache and allow the
system to vary it depending on configuration. Very few other arches actually set
this so its likely obsolete and doesn't need to be hardcoded anymore.
Also drop the 64 bit and x32 variants since autodetection of this value appears
to work fine and this avoids any confusion or sizing errors depending on large
file support.
(From OE-Core rev: 31ae16003cac6c8cf587c98d0c58e9f21690cb40)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Some of these are clearly dead, e.g. one binutils patch reverts the effects
of the earlier one.
This also removes the uclibc site files. We now have mechanisms to allow these
to be extended from another layer should someone ever wish to do that.
(From OE-Core rev: e01e7c543a559c8926d72159b5cd55db0c661434)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Add support for MIPS Release 6 ISA
(From OE-Core rev: fcb67508be00cdd22181d6c9e4c3d29dfa578b45)
Signed-off-by: Zubair Lutfullah Kakakhel <Zubair.Kakakhel@imgtec.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The fix for [YOCTO #5935] was applied for mips64 but not for mips64el
Patch it for mips64el
For description of issue, check OE-Core 7a5b6b96
(From OE-Core rev: 9b8d7f9fc10c862b78ebc669a7b47e9cb1142d87)
Signed-off-by: Zubair Lutfullah Kakakhel <Zubair.Kakakhel@imgtec.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
If we do not cache is then configure in distcc will try
to compute it using AC_TRY_RUN which will give some result
on x86-64 host when target is x86-64 but it generally is wrong
for cross compilation. We therefore cache it for all linux
targets.
This issue is highlighted by clang when compiling distcc
where it figures that its cross compiling and therefore triggers
it to include its own implementation of snprintf() which does
not go well with clang, gcc compiles it fine thats why we
never saw the problem thus far.
(From OE-Core rev: e940dfcb5ad4017e5fe616c583253439603656db)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The size of off_t depends on whether large file support is enabled through
preprocessor flags, so we can't cache this.
(From OE-Core rev: 25bd4e9483a7d2c16a460b4f363e91b5b943bb58)
Signed-off-by: Ross Burton <ross.burton@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
ccache's configure tries to do a runtime test to
determine hw_cv_func_snprintf_c99 and hw_cv_func_vsnprintf_c99
which wont work in OE.
c99 versions of these functions are provided on all
supported libc implementations in OE-Core
(From OE-Core rev: dbf2ac39913429a0d2a8ce129cebe4469d6a2c22)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
There are variables which are used in all target specific
sitefiles. Move it to common-linux, so it can be effective
for all targets. Usually they will vary based upon libc
e.g. musl does not have process shared mutexes so apr_cv_process_shared_works
should be no for it. For glibc though it should be yes but
existing behaviour is to use 'no' so its left as it is.
(From OE-Core rev: 2c5135f2106842e1b5ef880a08dbd4e50c9e0d04)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
If you profile gettext do_configure you notice it has some "sleep 1"
type events occurring. This patch ensures we cache the right values to
avoid those pointless delays there and in any other configure scripts
using the same macros.
(From OE-Core rev: ae49c16816e23fcfdcfb88d2d763e91be78f9dc0)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Add nios2-linux info, pulled from OE-Classic [1] as of commit
fabd8e6d07d3cd0cc93c2a0fc804f8c8f316c649 .
[1] https://github.com/openembedded/openembedded.git
(From OE-Core rev: 8fa526cd6ede52c71d8aa09482431ce656860c42)
Signed-off-by: Marek Vasut <marex@denx.de>
Signed-off-by: Ross Burton <ross.burton@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
We do not want bash to use internal code for getcwd() because
it can fail in certain filesystem types. Trust that the function
in glibc is not "broken".
(From OE-Core rev: 72539c68e4a6ef1497af3f707107c3a69ab37bbd)
Signed-off-by: Joe Slater <jslater@windriver.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Pull wordsize related items from arm-common and put in
arm-32. Leave them as they were. Copy arm-32 to arm-64
and comment out all variables. Re-define to correct
values leaving items which do not appear to be used
undefined.
(From OE-Core rev: 99d43a1554b2daeb7aba3c4582ccec758e6ec4a5)
Signed-off-by: Joe Slater <jslater@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
We need this to ensure there are some correct values when building with
meta-darwin. This matches the addition of a common-mingw site file in
the core.
(From OE-Core rev: 8c2ade6f48dcc522c68f798286d1c43bd2c1c8e7)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The expat mingw build breaks because ix86-common defines
ac_cv_func_mmap_fixed_mapped=yes, causing the build to think that
mmap(), mman.h actually exist. Fix this by always forcing the variable
to "no", which configure would have concluded without such interference.
(From OE-Core rev: 9add830b85222b57774acc35539e678d9ca8efb2)
Signed-off-by: Richard Tollerton <rich.tollerton@ni.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
