get-developers: add option to get developers for files

The DEVELOPERS file lists file patterns, but the get-developers script
only supports getting packages and CPU architectures. Some files are
neither of those, e.g. defconfigs, support files, package infras, ...

Add a '-f' option that allows to give a list of files for which to get
the developer(s).

Note that the DEVELOPERS file contains directories as well as files,
and the directories are not expanded into individual files. Therefore,
we have to use f.startswith(devfile) to match a directory. This assumes
that the directory name ends with /, otherwise 'package/python' would
also match 'package/python-foo'. Since we make sure that directories do
end with / in the DEVELOPERS file, and since false positives are not
much of an issue in this case, this isn't a problem.

Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

package/sysdig: fix incorrect module name

Latest version of sysdig introduced a variable for the kernel module
name. This patch updates that variable into Makefile.in.

Signed-off-by: Angelo Compagnucci <angelo.compagnucci@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

python-twisted: bump to version 17.5.0

Also fix/add runtime dependencies: Automat and hyperlink modules.

Signed-off-by: Yegor Yefremov <yegorslists@googlemail.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

python-automat: new package

Signed-off-by: Yegor Yefremov <yegorslists@googlemail.com>
[Arnout: properly wrapped Config.in help text]
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

python-m2r: new package

Host variant is needed to generate long description
for python-automat package.

Signed-off-by: Yegor Yefremov <yegorslists@googlemail.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

python-mistune: add host variant

Host variant is required by host-python-m2r.

Signed-off-by: Yegor Yefremov <yegorslists@googlemail.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/kodi-pvr-mythtv: bump version to 4.15.3

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/tor: bump version to 0.3.0.10

Release notes: https://blog.torproject.org/blog/tor-03010-released

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

python-netifaces: bump to version 0.10.6

Signed-off-by: Yegor Yefremov <yegorslists@googlemail.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/sysdig: bump to version 0.17.0

This patch bumps sysdig to version 0.17.0 and updates the
dependencies accordingly.

According to [1], the minimum GCC version is 4.8 so encode that as
well.

[1] https://www.sysdig.org/wiki/how-to-install-sysdig-from-the-source-code/

Signed-off-by: Angelo Compagnucci <angelo.compagnucci@gmail.com>
[Arnout: also bump gcc dependency to 4.8]
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/jq: install to staging

Install the libraries to staging to be usable by other packages.

Signed-off-by: Angelo Compagnucci <angelo.compagnucci@gmail.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/libb64: new package

This patch adds libb64, a library of ANSI C routines for fast
encoding/decoding data into and from a base64-encoded format.

The package contains a static library, headers, and an executable.
The latter, however, requires C++ and offers no advantages over busybox
or coreutils base64, so it is not installed. Therefore, nothing is
installed to target.

Signed-off-by: Angelo Compagnucci <angelo.compagnucci@gmail.com>
[Arnout:
- properly wrap Config.in
- move from Crypto to Other menu
- don't install to target
- don't make headers executable
- don't add -D to install with multiple source files
- extend commit log]
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/python-pillow: bump to version 4.2.1

Signed-off-by: Angelo Compagnucci <angelo.compagnucci@gmail.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

libcgi: bump to version 1.1

Cleanup the libcgi package,
by using https://github.com/rafaelsteil/libcgi as an upstream.
Remove old patch and use cmake to build new version.

Signed-off-by: Joe Lin <xlcwu.taiwan@gmail.com>
[Mario:
- Use github URL in Config.in
- Use upstream sha256 hash]
Signed-off-by: Mario J. Rugiero <mrugiero@gmail.com>
[Arnout:
- Add reference to upstream hash file
- Remove unneeded LIBCGI_SUPPORTS_IN_SOURCE_BUILD = NO]
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/libhttpparser: bump version to 2.7.1

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/mono: bump to version 5.0.1.1 and monolite to version 1050001000

This patch bumps to mono 5.0.1.1 and updates the unit tests patch
accordingly.

The latest version of mono carries a bit of changes in the monolite
package: the version string changes and from now on, monolite should
be installed in a subdirectory with the exact version string as a name.
Therefore, mono and monolite are bumped in a single commit.

Patch 0004, although applied in upstream master, is not yet in this
(stable) release.

Signed-off-by: Angelo Compagnucci <angelo.compagnucci@gmail.com>
[Arnout: squashed the patches, removed /usr part from HOST_DIR]
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/stella: bump version to 5.0.1

Removed 0003-Use-gnu-11-standard-to-prevent-errors-on-PPC.patch, since
stella now uses -std=c++14 that builds fine when using PPC altivec
vectorization.

Removed 0004-gcc7.patch, backported from upstream.

Update toolchain dependencies to gcc 4.9 since codebase now uses C++14 features.

Signed-off-by: Sergio Prado <sergio.prado@e-labworks.com>
[Arnout: also remove 0004-gcc7.patch]
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

Merge branch 'master' into next

This is in preparation of bumping the stella package, which is also
modified in master.

package/fontconfig: bump version to 2.12.4

Removed patch applied upstream:
https://cgit.freedesktop.org/fontconfig/commit/?id=1ab5258f7c2abfafcd63a760ca08bf93591912da

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/erlang-jiffy: bump version to 0.14.11

Removed ERLANG_JIFFY_SET_VERSION_HOOK, not needed anymore after
https://github.com/davisp/jiffy/commit/d596d890bcc7d74d34d185854873ba678bbf237b

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/mksh: bump version to R55

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/minizip: bump version to 1.1

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/memcached: bump version to 1.5.0

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/mdadm: bump version to 4.0

This version of mdadm can optionally use the corosync and libdlm
packages. However, it checks for their presence by directly probing the
host /usr/include, which fails for cross-compilation:

http://git.neil.brown.name/?p=mdadm.git;a=commitdiff;h=7716570e6d906e7326f83d466f6ba73009649d03#patch2
http://git.neil.brown.name/?p=mdadm.git;a=commitdiff;h=d15a1f72bd92bc4724ee94b2ae8132633ffeb72b#patch1

Since Buildroot doesn't package either corosync nor libdlm, disable
them explicitly with -DNO_COROSYNC -DNO_DLM.

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/mcelog: bump version to 153

Removed patch applied upstream:
https://git.kernel.org/pub/scm/utils/cpu/mce/mcelog.git/commit/?id=524ed1cd16de5adcb4a5015777d3dfc63afda1fe

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/lzo: bump version to 2.10

Added locally computed sha256 hash.

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/lzip: bump version to 1.19

Switched to sha256 hash provided by upstream.

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/lvm2: bump version to 2.02.173

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/luvi: bump version to 2.7.6

Switched to github helper because upstream does not provide the tarball
anymore.

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/luv: bump version to 1.9.1-1

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/libuv: bump version to 1.13.1

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/logsurfer: use hashes provided by upstream

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/logrotate: bump version to 3.12.3

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/log4cpp: bump version to 1.1.3

Added sha256 hash, removed patch applied upstream:
https://sourceforge.net/p/log4cpp/codegit/ci/fe4ed0

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

fs/squashfs: honour the number of parallel jobs

By default, mksquashfs uses all the CPUs available, like us with
BR2_JLEVEL=0.

However, by setting BR21_JLEVEL to non-zero, a user volutarily limits
the number of jobbs allowed to run concurrently (for various reasons,
like this is not a priority task on the machine; or like the
autobuilders, to spread the load accross multiple runs without
overloading the machine).

So, propagate tthe parallelisn settings down to mksquashfs, to repect
the user's will.

Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Clayton Shotwell <clayton.shotwell@rockwellcollins.com>
Tested-by: Clayton Shotwell <clayton.shotwell@rockwellcollins.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/libatomic_ops: bump version to 7.4.6

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/{mesa3d, mesa3d-headers}: bump version to 17.1.6

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/sqlite: bump version to 3200000 (3.20.0)

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

expat: bump to version 2.2.3

Drop the XML_POOR_ENTROPY workaround. Upstream commit fd9581a34e5665
(Never require XML_POOR_ENTROPY for "./configure && make") fixes this
issue.

Signed-off-by: Baruch Siach <baruch@tkos.co.il>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

ficl: fix build with gcc 7

gcc 7 with -Os (optimize for size) takes the liberty to remove the code
of inline functions entirely - which is indeed allowed according to the
C99 standard, cfr. this [1] very complete StackOverflow answer. This
leads to undefined function references at link time.

The proper approach is to patch the code to use 'static inline'
instead. However, for various reasons such a patch is not entirely
trivial, and upstream has not moved since 7 years, so instead take the
simpler approach of restoring the gcc original inline behaviour with
-fgnu89-inline to fix this issue.

Fixes http://autobuild.buildroot.org/results/d72/d72a7e9775a10bc884b9953dd2eba3ad32100cca

[1] https://stackoverflow.com/questions/6312597/is-inline-without-static-or-extern-ever-useful-in-c99

Signed-off-by: Francois Perrad <francois.perrad@gadz.org>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/stella: fix gcc7 support

Fixes
http://autobuild.buildroot.net/results/6ed/6edd7950c402dbf433f3904193c645ebc8531c7e/

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

Revert "package/skeleton-custom: also check for missing directories"

This partially reverts commit cf6d321e9d5c9dec63e42466fec4bf5de5853bd3.

The underlying reason for doing the checks is to ensure that a
merged-usr setting is properly enforced, even for custom skeletons.

Before that patch, a custom skeleton where both /bin and /usr/bin were
missing was accepted; but then the first package that intalled something
in /bin would create it as a directory, thus breaking the merged-usr
situation. Ditto sbin and lib, of course.

cf6d321 was created to detect that situation, and the fix was to require
that both directories do exist in the custom skeleton, so that we new
the check for consistency were OK, in all cases.

However, that broke existing skeletons which where missing both
directories, and some people are shinning about it...

The crux of the problem is that Buildroot should be responsible for
creating the directories or the symliks when they are missing, and only
fail if the existing ones are incorrect, but not impose the burden on
the user.

A situation where the problem arises is when a skeleton is shared
between various builds, some using a merged-usr while other do not.

We fix that by reverting the offending changes change, back to the
previous behaviour. We keep the stderr redirection in stat calls, and
the variable renaming.

A proper fix to create the missing directories can be added later.

Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Arnout Vandecappelle <arnout@mind.be>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

uclibc: allow thread_db to be built with C++

gdb 7.12+ is compiled and linked with C++, therefore the thread_db
headers need to use extern "C", to be usable for C++ code.

Fixes:
 Issue when building gdbserver for m68k-coldfire:

thread-db.o: In function `thread_db_init()':
thread-db.c:(.text+0x5b6): undefined reference to `td_ta_new(ps_prochandle*, td_thragent**)'
thread-db.c:(.text+0x61e): undefined reference to `td_thr_get_info(td_thrhandle const*, td_thrinfo*)'
thread-db.c:(.text+0x632): undefined reference to `td_symbol_list()'
...

Signed-off-by: Waldemar Brodkorb <wbx@openadk.org>
Tested-by: Romain Naour <romain.naour@gmail.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

pkg-cmake.mk: set pkg-config env. vars. in the host configure environment

This change is fixing the bug reported in [1].

Cmake may run pkg-config to find the dependencies when configuring a
package. Because of the value of PATH, and it will use the Buildroot's
pkg-config wrapper, which, by default, is configured (via some
environment variables) to find the target dependencies.

When configuring a host package using cmake, to prevent cmake from
wrongly solving dependencies from the target tree (when the
*-config.cmake files are using pkg-config) instead of looking for them
in the Buildroot's host tree or in the host system itself, we need to
set the environment variables altering the pkg-config behavior in the
cmake configure environment.

So, this change is fixing the cmake host-packages configuration step,
by properly setting the pkg-config environment variables to their values
for finding host dependencies.

Before this patch:

  make O=/opt/br/abo/cmake-host-find-root-path libxml2 host-mariadb{-dirclean,-configure} && echo $?
  [...]
  >>> host-mariadb 10.1.25 Configuring
  (mkdir -p /opt/br/abo/cmake-host-find-root-path/build/host-mariadb-10.1.25/ && cd /opt/br/abo/cmake-host-find-root-path/build/host-mariadb-10.1.25/ && rm -f CMakeCache.txt && PATH="/opt/br/abo/cmake-host-find-root-path/host/bin:/opt/br/abo/cmake-host-find-root-path/host/sbin:/home/sam/.local/bin:/sbin:/usr/sbin:/bin:/usr/bin"  /usr/bin/cmake /opt/br/abo/cmake-host-find-root-path/build/host-mariadb-10.1.25/ -DCMAKE_INSTALL_SO_NO_EXE=0 -DCMAKE_FIND_ROOT_PATH="/opt/br/abo/cmake-host-find-root-path/host" -DCMAKE_FIND_ROOT_PATH_MODE_PROGRAM="BOTH" -DCMAKE_FIND_ROOT_PATH_MODE_LIBRARY="BOTH" -DCMAKE_FIND_ROOT_PATH_MODE_INCLUDE="BOTH" -DCMAKE_INSTALL_PREFIX="/opt/br/abo/cmake-host-find-root-path/host" -DCMAKE_C_FLAGS="-O2 -I/opt/br/abo/cmake-host-find-root-path/host/include" -DCMAKE_CXX_FLAGS="-O2 -I/opt/br/abo/cmake-host-find-root-path/host/include" -DCMAKE_EXE_LINKER_FLAGS="-L/opt/br/abo/cmake-host-find-root-path/host/lib -Wl,-rpath,/opt/br/abo/cmake-host-find-root-path/host/lib"
  -DCMAKE_ASM_COMPILER="/usr/bin/as" -DCMAKE_C_COMPILER="/usr/bin/gcc" -DCMAKE_CXX_COMPILER="/usr/bin/g++"  -DCMAKE_C_COMPILER_ARG1="" -DCMAKE_CXX_COMPILER_ARG1=""  -DCMAKE_COLOR_MAKEFILE=OFF -DBUILD_DOC=OFF -DBUILD_DOCS=OFF -DBUILD_EXAMPLE=OFF -DBUILD_EXAMPLES=OFF -DBUILD_TEST=OFF -DBUILD_TESTS=OFF -DBUILD_TESTING=OFF  -DWITH_SSL=bundled )
  -- Running cmake version 3.8.2
  [...]
  -- Found PkgConfig: /opt/br/abo/cmake-host-find-root-path/host/bin/pkg-config (found version "0.28")
  [...]
  -- Found LibXml2: /opt/br/abo/cmake-host-find-root-path/host/arm-buildroot-linux-gnueabihf/sysroot/usr/lib/libxml2.so (found version "2.9.4")
  [...]
  0

After this patch is applied:

  make O=/opt/br/abo/cmake-host-find-root-path libxml2 host-mariadb{-dirclean,-configure} && echo $?
  [...]
  >>> host-mariadb 10.1.25 Configuring
  (mkdir -p /opt/br/abo/cmake-host-find-root-path/build/host-mariadb-10.1.25/ && cd /opt/br/abo/cmake-host-find-root-path/build/host-mariadb-10.1.25/ && rm -f CMakeCache.txt && PATH="/opt/br/abo/cmake-host-find-root-path/host/bin:/opt/br/abo/cmake-host-find-root-path/host/sbin:/home/sam/.local/bin:/sbin:/usr/sbin:/bin:/usr/bin" PKG_CONFIG="/opt/br/abo/cmake-host-find-root-path/host/bin/pkg-config" PKG_CONFIG_SYSROOT_DIR="/" PKG_CONFIG_LIBDIR="/opt/br/abo/cmake-host-find-root-path/host/lib/pkgconfig:/opt/br/abo/cmake-host-find-root-path/host/share/pkgconfig" PKG_CONFIG_ALLOW_SYSTEM_CFLAGS=1 PKG_CONFIG_ALLOW_SYSTEM_LIBS=1  /usr/bin/cmake /opt/br/abo/cmake-host-find-root-path/build/host-mariadb-10.1.25/ -DCMAKE_INSTALL_SO_NO_EXE=0 -DCMAKE_FIND_ROOT_PATH="/opt/br/abo/cmake-host-find-root-path/host" -DCMAKE_FIND_ROOT_PATH_MODE_PROGRAM="BOTH" -DCMAKE_FIND_ROOT_PATH_MODE_LIBRARY="BOTH" -DCMAKE_FIND_ROOT_PATH_MODE_INCLUDE="BOTH" -DCMAKE_INSTALL_PREFIX="/opt/br/abo/cmake-host-find-roo
 t-path/host" -DCMAKE_C_FLAGS="-O2 -I/opt/br/abo/cmake-host-find-root-path/host/include" -DCMAKE_CXX_FLAGS="-O2 -I/opt/br/abo/cmake-host-find-root-path/host/include" -DCMAKE_EXE_LINKER_FLAGS="-L/opt/br/abo/cmake-host-find-root-path/host/lib -Wl,-rpath,/opt/br/abo/cmake-host-find-root-path/host/lib" -DCMAKE_ASM_COMPILER="/usr/bin/as" -DCMAKE_C_COMPILER="/usr/bin/gcc" -DCMAKE_CXX_COMPILER="/usr/bin/g++"  -DCMAKE_C_COMPILER_ARG1="" -DCMAKE_CXX_COMPILER_ARG1=""  -DCMAKE_COLOR_MAKEFILE=OFF -DBUILD_DOC=OFF -DBUILD_DOCS=OFF -DBUILD_EXAMPLE=OFF -DBUILD_EXAMPLES=OFF -DBUILD_TEST=OFF -DBUILD_TESTS=OFF -DBUILD_TESTING=OFF  -DWITH_SSL=bundled )
  -- Running cmake version 3.8.2
  [...]
  -- Found PkgConfig: /opt/br/abo/cmake-host-find-root-path/host/bin/pkg-config (found version "0.28")
  [...]
  -- Found LibXml2: /usr/lib/libxml2.so (found version "2.9.4")
  [...]
  0

[1] http://lists.busybox.net/pipermail/buildroot/2017-August/199776.html

Reported-by: "Sigalas, Antonios (Nokia - GR/Athens)" <antonios.sigalas@nokia.com>
Cc: "Sigalas, Antonios (Nokia - GR/Athens)" <antonios.sigalas@nokia.com>
Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Samuel Martin <s.martin49@gmail.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

faad2: security bump to version 2.8.1

Fixes: CVE-2017-9218, CVE-2017-9219, CVE-2017-9220, CVE-2017-9221,
CVE-2017-9222, CVE-2017-9223, CVE-2017-9253, CVE-2017-9254,
CVE-2017-9255, CVE-2017-9256, CVE-2017-9257

http://seclists.org/fulldisclosure/2017/Jun/32

Switch to .tar.bz2 to save some bandwidth.

Add autoreconf since unfortunately upstream tarball does not ship the
configure script.

Cc: Gustavo Zacarias <gustavo@zacarias.com.ar>
Signed-off-by: Baruch Siach <baruch@tkos.co.il>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

nvme: remove the udev optional dependency

Upstream removed the udev dependent code in commit 4cbe5888f4bc (nvme-cli:
fabrics: remove libudev dependency).

Cc: Mamatha Inamdar <mamatha4@linux.vnet.ibm.com>
Signed-off-by: Baruch Siach <baruch@tkos.co.il>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

nvme: fix libuuid test when cross compiling

Use $(LD) to test whether libuuid is installed, so that we check the target,
not the host.

Fixes:
http://autobuild.buildroot.net/results/e26/e263ffbf8947374c2aac73ab42edcbf4d9a3600a/
http://autobuild.buildroot.net/results/53b/53b2d1e0c3cdeafb980389bcfd8a10b6ca48fbc2/

Cc: Mamatha Inamdar <mamatha4@linux.vnet.ibm.com>
Signed-off-by: Baruch Siach <baruch@tkos.co.il>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/snappy: remove pthread handling for static builds

Commit 0e2ee0e6e8076 (package.snappy: fix static builds with threads)
added LIBS=-pthread to configure options to fix static build with
threads on some architectures. This works for autotools packages. Then,
commit d26282b3aa14cb (package/snappy: bump version, switch to cmake)
switched to cmake that does not understand the 'LIBS' parameter.

As pointed out by Thomas[1] the special handling of libpthread for
static builds is not needed anymore: it was needed only to build the
tests, and these are no longer built. Note that packages that link
statically with snappy will still need to link with -pthread. Only
two in-tree packages currently depend on snappy: lua-csnappy depends
on !BR2_STATIC_LIBS and leveldb adds -pthread from its build system.

Tested with the following musl & uclibc defconfigs which also fix build
errors:
http://autobuild.buildroot.net/results/39f/39f9fc4300ddca19e8a533765b46838003752000/
http://autobuild.buildroot.net/results/886/8860b9259f7a3aecb2ddbfa787be048072bd347b/

[1] http://lists.busybox.net/pipermail/buildroot/2017-July/199549.html

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Acked-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
[Baruch: improved commit message]
Signed-off-by: Baruch Siach <baruch@tkos.co.il>
[Arnout: improved commit message more, based on Yann's input]
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

bind: fix configure in case lmdb devel files are present on the host

Fix configure failure in case lmdb devel files are present on the host
by adding --without-lmdb option (reported [1] and fix tested [2],[3] by
grunpferd@netscape.net).

Fixes:

  checking for lmdb library... yes
  checking for library containing mdb_env_create... no
  configure: error: found lmdb include but not library.

[1] http://lists.busybox.net/pipermail/buildroot/2017-August/199945.html
[2] http://lists.busybox.net/pipermail/buildroot/2017-August/199963.html
[3] http://lists.busybox.net/pipermail/buildroot/2017-August/199964.html

Signed-off-by: Peter Seiderer <ps.report@gmx.net>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/jack2: backport two build fixes

0005-gcc7.patch fixes
http://autobuild.buildroot.net/results/c06/c0610325d7785dfa51c5d36775623ca8fa517f24/

0006-fix-ftbfs-with-clang.patch
fixes the subsequent build error:
common/memops.c.31.o: In function `sample_move_dither_rect_d16_sSs':
memops.c:(.text+0x4dc): undefined reference to `fast_rand'

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

glibc: add patch to fix build on x86

This fixes the following build issues:

In file included from ../sysdeps/x86_64/multiarch/varshift.c:19:0,
                 from ../sysdeps/i386/i686/multiarch/varshift.c:1:
../sysdeps/x86_64/multiarch/varshift.h: In function '__m128i_shift_right':
../sysdeps/x86_64/multiarch/varshift.h:26:1: error: SSE vector return without SSE enabled changes the ABI [-Werror=psabi]
 {

[...]

.../lib/gcc/i686-buildroot-linux-gnu/7.1.0/include/tmmintrin.h:136:1: error: inlining failed in call to always_inline '_mm_shuffle_epi8': target specific option mismatch

The patch has been taken from glibc's patchwork at
https://patchwork.sourceware.org/patch/21003/.

Fixes bug #10156.

Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Tested-by: Romain Naour <romain.naour@gmail.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

linux-headers: aarch64: Prevent selecting unsupported versions

Versions older than Linux v3.7 do not support the aarch64 architecture
so disable them, for reference see https://kernelnewbies.org/Linux_3.7

Without this patch these defconfigs fail to build

BR2_aarch64=y
BR2_KERNEL_HEADERS_3_2=y

BR2_aarch64=y
BR2_KERNEL_HEADERS_3_4=y

with error messages like this:

make[1]: Entering directory '/home/buildroot/br5_next/output/build/linux-headers-3.2.89'
Makefile:567: /home/buildroot/br5_next/output/build/linux-headers-3.2.89/arch/arm64/Makefile: No such file or directory

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

linux: bump default to version 4.12.5

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

linux-headers: bump 4.{4, 9, 12}.x series

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

linux-headers: bump 4.1.x series

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

qemu-sparc64: update kernel

This fixes '__multi3' issue when building the kernel.

Signed-off-by: Waldemar Brodkorb <wbx@openadk.org>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

qemu-sparc: update kernel

Fixes gcc 7.1.0 compile issues.

Signed-off-by: Waldemar Brodkorb <wbx@openadk.org>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

uclibc: fix microblaze runtime error

With binutils 2.29 a runtime error is seen:
Kernel panic - not syncing: Attempted to kill init! exitcode=0x00000100

Backport upstream patch to fix the issue.

Signed-off-by: Waldemar Brodkorb <wbx@openadk.org>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

linux: drop full stop from prompt text

We don't add a full stop at the end of the prompt text.

Signed-off-by: Baruch Siach <baruch@tkos.co.il>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

mutt: fix build with BR2_PACKAGE_SKELETON_NONE

BR2_PACKAGE_SKELETON_NONE does not provide the /var subdirectory which
breaks symlink creation in that directory. Make sure that /var exists
before creating the symlink.

Fixes:
http://autobuild.buildroot.net/results/32d/32dff0df759baf1120e4f11335e735eb0708b237/

Signed-off-by: Baruch Siach <baruch@tkos.co.il>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

qt5base: fix examples compile without opengl support

Fixes [1]:

  make[1]: Entering directory '.../build/qt5base-5.9.1/examples/opengl'
  cd hellowindow/ && ( test -e Makefile || .../build/qt5base-5.9.1/bin/qmake -o Makefile .../build/qt5base-5.9.1/examples/opengl/hellowindow/hellowindow.pro ) && make -f Makefile
  make[2]: Entering directory '.../build/qt5base-5.9.1/examples/opengl/hellowindow'
  .../host/bin/sparc-linux-g++ -c -pipe -D_LARGEFILE_SOURCE -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 -Os --sysroot=.../host/sparc-buildroot-linux-uclibc/sysroot -O2 -O3 -fno-exceptions -Wall -W -D_REENTRANT -fPIC -DQT_NO_EXCEPTIONS -D_LARGEFILE64_SOURCE -D_LARGEFILE_SOURCE -DQT_NO_DEBUG -DQT_GUI_LIB -DQT_CORE_LIB -I. -I../../../include/QtGui/5.9.1 -I../../../include/QtGui/5.9.1/QtGui -I../../../include -I../../../include/QtGui -I../../../include/QtCore/5.9.1 -I../../../include/QtCore/5.9.1/QtCore -I../../../include/QtCore -I.moc -I../../../mkspecs/devices/linux-buildroot-g++ -o .obj/hellowindow.o hellowindow.cpp
  In file included from hellowindow.cpp:51:0:
  hellowindow.h:96:5: error: ‘QOpenGLShaderProgram’ does not name a type
       QOpenGLShaderProgram *m_program;
       ^~~~~~~~~~~~~~~~~~~~
  hellowindow.h:97:5: error: ‘QOpenGLBuffer’ does not name a type
       QOpenGLBuffer m_vbo;
       ^~~~~~~~~~~~~
  hellowindow.cpp: In constructor ‘Renderer::Renderer(const QSurfaceFormat&, Renderer*, QScreen*)’:
  hellowindow.cpp:62:40: error: invalid use of incomplete type ‘class QOpenGLContext’
       m_context = new QOpenGLContext(this);
                                          ^
  In file included from ../../../include/QtGui/qsurfaceformat.h:1:0,
                   from ../../../include/QtGui/../../src/gui/kernel/qsurface.h:45,
                   from ../../../include/QtGui/qsurface.h:1,
                   from ../../../include/QtGui/../../src/gui/kernel/qwindow.h:51,
                   from ../../../include/QtGui/qwindow.h:1,
                   from ../../../include/QtGui/QWindow:1,
                   from hellowindow.h:51,
                   from hellowindow.cpp:51:

[1] http://autobuild.buildroot.net/results/001/001bc930014fca8c2a8b7a390350b4e1a15fa375/build-end.log

Signed-off-by: Peter Seiderer <ps.report@gmx.net>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

linux: bump cip version to v4.4.75-cip6

Signed-off-by: Angelo Compagnucci <angelo.compagnucci@gmail.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

DEVELOPERS: add Joshua Henderson for qt5wayland

Add myself and include the qt5/qt5wayland package.

Signed-off-by: Joshua Henderson <joshua.henderson@microchip.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

netatalk: bump to version 3.1.9

This version is compatible with LibreSSL.

Signed-off-by: Adam Duskett <Adamduskett@outlook.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

glibc: add patch fixing the build with binutils 2.29

Fixes the following build failure, which occurs on all architectures:

/tmp/cc6APNmN.s: Assembler messages:
/tmp/cc6APNmN.s: Error: `loc1@GLIBC_2.17' can't be versioned to common symbol 'loc1'
/tmp/cc6APNmN.s: Error: `loc2@GLIBC_2.17' can't be versioned to common symbol 'loc2'
/tmp/cc6APNmN.s: Error: `locs@GLIBC_2.17' can't be versioned to common symbol 'locs'
../o-iterator.mk:9: recipe for target '/tmp/output/build/glibc-2.25/build/misc/regexp.os' failed

Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

gdb: drop support for Microblaze

Microblaze support was never upstreamed in gdb. We currently use a gdb
from the Xilinx Github repository, but this gdb is based on 7.6, and
has never been updated in the last 4 years. There are no other active
branches at https://github.com/Xilinx/gdb/branches.

Xilinx has a slightly newer gdb, based on 7.7, available at
https://github.com/Xilinx/meta-xilinx/tree/master/recipes-microblaze/gdb. However,
it's apparently only available in the form of stack of big patches
(https://github.com/Xilinx/meta-xilinx/tree/master/recipes-microblaze/gdb/files).

Currently, gdb on Microblaze fails to build in various conditions:

 * Against glibc, with "error: conflicting types for 'ps_lgetfpregs'"

 * Against musl, with "error: unknown type name 'elf_gregset_t'"

While those issues can probably be fixed, the fact that there is no
active upstream significantly reduces the incentive to fix those
problems.

Therefore, let's drop support for gdb on Microblaze entirely.

Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Acked-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Acked-by: Romain Naour <romain.naour@gmail.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

dbus: add upstream patch to fix startup hang with with expat >= 2.2.1

After c0ad6ded018ffbc33f7f5 expat: security bump to version 2.2.1
the system can hang on startup under certain circumstances.

This happens when:
  * we use systemd as init system
  * the random nonblocking pool takes a while to initialize
    * this apparently doesn't happen on qemu, so this would not have
      been caught by the runtime testing infrastructure
    * it also doesn't seem to happen when network booting

For a more detailed description of the bug see here:
https://bugs.freedesktop.org/show_bug.cgi?id=101858

The patch should be in next dbus version 1.10.24

Set DBUS_AUTORECONF = YES because configure.ac is changed.

Signed-off-by: Marcus Hoffmann <m.hoffmann@cartelsol.com>
[Arnout: add upstream commit sha + Marcus's Sob to the patch]
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

package/skeleton-common: fix recursive variable

This fixes a recursive variable definition introduced by the refactoring
in commit 120307520f69f801bf05df1d2fd3433c6d498551. Simply rename the
hook to remove the recursion.

Fixes https://bugs.busybox.net/show_bug.cgi?id=10146

Signed-off-by: Stefan Becker <chemobejk@gmail.com>

azmq: Add new package

The azmq library provides Boost Asio style bindings for ZeroMQ. This
library is built on top of ZeroMQ's standard C interface and is intended
to work well with C++ applications which use the Boost libraries in
general, and Asio in particular.

Signed-off-by: RJ Ascani <rj.ascani@gmail.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

instrumentation: extend packages-file-list.txt with symlinks and directories

Using the package instrumentation hooks, a file packages-file-list.txt is
created containing the list of files installed by a package with their size.

Due to the use of 'find -type f', symbolic links to files are not included
in this file list. Since the original purpose of this file was to calculate
the total size of a package and symbolic links have file size 0, this was
not a problem.

However, if packages-file-list.txt is reused for other purposes, for example
to get a complete list of files installed by a package regardless of size,
symbolic links to files are important too.

Likewise, to get a complete view of what a package installs, directories
should be included too.

Update the instrumentation hook accordingly.
Although for files an md5sum is taken, we cannot do this for directories.
Instead, mimic the output of md5sum with a fake hash string.

Note: for directories that are used by several packages, e.g. /etc/init.d/,
the package that created the directory will be treated as the 'owner' of
that directory.  This gives a somewhat distorted view of ownership.
Similarly, the package size reported by 'make graph-size' (based on
packages-file-list.txt) will include the 'size' of a directory for that
owner, which is reported as e.g. 4096 bytes.

Signed-off-by: Thomas De Schampheleire <thomas.de_schampheleire@nokia.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

instrumentation: extract duplication to get list of installed files

Before and after the building of each package, the instrumentation hooks are
run. One of these hooks obtains the list of files installed by a package.

The code to obtain this list is currently duplicated in the start and end
part of the hook. While the amount of duplication is currently small, a
subsequent patch will make more changes to this code, increasing the
duplication.

Therefore, split off into a helper function.

Signed-off-by: Thomas De Schampheleire <thomas.de_schampheleire@nokia.com>
Reviewed-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

qt5wayland: new package

QtWayland is a Qt 5 module that wraps the functionality of Wayland.

Cc: Julien Corjon <corjon.j@ecagroup.com>
Cc: Brent Sink <brents_3@hotmail.com>
Cc: Naoki Matsumoto <n-matsumoto@melcoinc.co.jp>
Signed-off-by: Joshua Henderson <joshua.henderson@microchip.com>
[Arnout:
 - select wayland instead of depending on it;
 - propagate jscore dependency from qt5declarative
 - re-order select/depends statements
 - rewrite comment to only refer to GL backend (jscore is an arch
   dependency)
 - rewrap help text of BR2_PACKAGE_QT5WAYLAND_COMPOSITOR]
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

imx6q-sabresd: Add a qt5 defconfig variant

Introduce imx6q-sabresd_qt5_defconfig that supports the opensource
Etnaviv graphical stack.

This defconfig provides a way to quickly test some graphical applications,
such as kmscube, qt5, glmark2.

Currently kernel mainline exhibits issues when running cpufreq as ondemand
governor on mx6, so add a linux fragment that disables such option for the
time being.

Signed-off-by: Fabio Estevam <festevam@gmail.com>
[Arnout:
 - Add Fabio to DEVELOPERS
 - Rename linux config fragment to linux_qt5.fragment
 - Add comment to linux_qt5.fragment why it is needed
 - Add explicit CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE=y to it
 - Reorder defconfig entries to they match imx6q-sabresd_defconfig
 - Remove strace and host-uboot-tools, neither seems to be needed
 - Drop BR2_PACKAGE_QT5BASE_LICENSE_APPROVED, it no longer exists
 - Replace BR2_TARGET_ROOTFS_EXT2_BLOCKS with BR2_TARGET_ROOTFS_EXT2_SIZE
 - Regenerate .gitlab-ci.yml]
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>

Update for 2017.08-rc1

Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

python-systemd: bump to version 234

Signed-off-by: Adam Duskett <aduskett@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

CHANGES: final update before 2017.08-rc1

Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

package/kodi: add support for more architectures

For the autoconf-based build system it was necessary to allow each
supported architecture separatly:
https://github.com/xbmc/xbmc/blob/Krypton/m4/xbmc_arch.m4

For the CMake build system this is not necessary anymore so we only
blacklist archs in Config.in known to be broken.

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
[Thomas: slightly tweak the BR2_PACKAGE_KODI_ARCH_SUPPORTS definition.]
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

CHANGES: update with gcc 7.x graphite optimization bug fixed

Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

systemd: bump version to 234

Other changes:
- Removed the following patches due to being merged upstream:
    - resolved-bugfix-of-null-pointer-p-question-dereferencing.patch
    - resolved-simplify-alloc-size-calculation.patch
    - resolved-do-not-allocate-packets-with-minimum-size.patch
- Updated the following patches to work with 234:
    - fix-getty-unit.patch
    - build-check-for-ln-relative.patch
    - fix-am-path-libgcrypt-no-found.patch
- Updated ordering of remaining patches.
- Reformatted remaining patches as git style patches.
- Updated Upstream-Status comment in "build-check-for-ln-relative.patch" to
  "Denied [No desire to support building on old distributions]"

Signed-off-by: Adam Duskett <aduskett@gmail.com>
Tested-by: Marcus Hoffmann <m.hoffmann@caretelsol.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

let-me-create: fix build with musl

Add a patch to fix build with musl C library.

Fixes:
  http://autobuild.buildroot.net/results/af946fa6fe05ee265e4ac97742b15afeb0cea1ab

Signed-off-by: Rahul Bedarkar <rahulbedarkar89@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

configs/socrates_cyclone5: new defconfig

Configuration contains:
  - building uboot (and spl) from altera's github
  - booting from SD card
  - working ethernet, serial, i2c, fpga_manager
  - minimal rootfs with busybox
  - ready to flash SD card image

Signed-off-by: Michał Łyszczek <michal.lyszczek@bofc.pl>
[Thomas:
 - add entry to DEVELOPERS file
 - refresh .gitlab-ci.yml
 - fix commit title]
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

package/binutils: fix assertion in 2.29 xtensa ld

Section sorting code in xtensa ld was incorrectly cleaned up in release
2.29, resulting in the following assertion when --sort-section,alignment
option is used:

  BFD (GNU Binutils) 2.29 assertion fail elf32-xtensa.c:3965

Backport the fix that restores original section sorting.

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

package/binutils: fix crash caused by buggy xtensa overlay

In some xtensa configurations there may be system/user registers in
xtensa-modules with negative index. ISA initialization for such config
may clobber heap and result in program termination.
Don't update lookup table entries for register with negative indices.

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

CHANGES: update with the cloog removal

Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

cloog: remove package

This host-only package was used to enable graphite support in GCC 4.9,
but Buildroot no longer supports Graphite optimizations in GCC 4.9
(only GCC >= 5.x is supported). Therefore, we can remove the cloog
package.

Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

isl: bump to version 0.18

Now that we no longer support Graphite optimizations in gcc 4.9 (which
required ISL 0.14.1), we can bump ISL to a newer version, which allows
to support Graphite optimizations in GCC >= 5.x.

The comment about the version being limited to 0.14.1 because of cloog
is no longer relevant, so it is removed.

Our libtool patch now applies properly, so we can remove
ISL_LIBTOOL_PATCH = NO.

Signed-off-by: Andrey Yurovsky <yurovsky@gmail.com>
[Thomas:
 - remove ISL_LIBTOOL_PATCH = NO
 - improved commit log]
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

gcc: support grahite only for GCC 5 or above

GCC 4.9.x requires an old version of ISL to support graphite, and this
old version of ISL is not sufficient to support graphite in modern GCC
versions.

Since GCC 4.9.x is getting older, and to keep things simple, we stop
supporting graphite on old GCC < 5.x. This way, we will be able to
bump ISL to a version that is suitable for modern GCC versions.

In addition, this allows to drop the dependency on cloog, which was
only needed for GCC 4.9.

Signed-off-by: Andrey Yurovsky <yurovsky@gmail.com>
[Thomas: rework commit log.]
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

CHANGES: update with recent changes

Update with the skeleton rework details, and a webkitgtk bug that is
now fixed.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

support/testing: add runtime testing for read-only systemd

We add the 3 following combinations:

  - basic systemd, read-only, network w/ ifupdown
  - basic systemd, read-only, network w/ networkd
  - full systemd, read-only, network w/ networkd

The tests just verify what the /sbin/init binary is, and that we were
able to grab an IP address. More tests can be added later, for example
to check each systemd features (journal, tmpfiles...)

Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
[Arnout: regenerate .gitlab-ci.yml]
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

system: make systemd work on a read-only rootfs

When the rootfs is readonly, systemd will expect /var to be writable.
Because we do not really have a R/W filesystem to mount on /var, we make
it a tmpfs [*], and use the systemd-tmpfiles feature to populate it with
"factory" defaults.

We obtain those factory defaults by redirecting /var to that location at
build time, using a symlink /var -> /usr/share/factory which is the
location in which systemd-tmpfiles will look for when instructed to
"recursively copy" a directory.

With a line like:

    C /var/something - - - -

it will look for /usr/share/factory/something and copy it (recursively
if it is a directory) to /var/something, but only if it does not already
exist there.

We also mark this copy with the exclamation mark, as it is only safe to
copy on boot, not when changing targets.

To be noted: the real format for such lines are:

    C /var/something - - - - /from/where/to/copy/something

But if the source is not given, then it is implicitly taken from
/usr/share/factory (which in our case is as-good a location as whatever
else, so we use it, and thus we need not specify the source of the
copy).

Note that we treat symlinks a little bit specially, by creating symlinks
to the factory defaults rather than copying them.

Finally, /var at build time is a symlink, but at runtime, it must be a
directory (so we can mount the tmpfs over there). We can't change that
as a target-finalize hook, because:

  - some packages may want to set ownership and/or access rights on
    files or directories in /var, and that only happens while assembling
    the filesystem images; changing /var from a symlink to a (then
    empty) directory would break this;

  - /var would be a directory on sub-sequent builds (until the next
    "make clean").

Instead, we use the newly-introduce pre- and post-rootfs command hooks,
to turn /var into a directory before assembling the image, and back to a
symlink after assembling the image.

[*] People who want the factory-defaults only on first boot will have
    to tweak the fstab to mount something else than a tmpfs on /var.

Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Reviewed-by: Romain Naour <romain.naour@gmail.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

fs: add pre- and post-command hooks

In some cases, the directory structure we want in the filesystem is not
exactly what we have in target/

For example, when systemd is used on a read-only rootfs, /var must be a
tmpfs. However, we may have packages that install stuff in there, and
set important rights (via the permission-table). So, at build time, we
need /var to be a symlink to the remanent location (/usr/share/factory)
while at runtime we need /var to be a directory.

One option would have been to have /var as a real directory even during
build time, and in a target-finalize hook, move everything out of there
and into the "factory" location. However, that's not possible because
it's too early: some packages may want to set ownership and/or acces
rights on directories or files in /var, and this is only done in the
fakeroot script, which is called only later during the assembling of the
filesystem images.

Also, there would have been no way to undo the tweak (i.e. we need to
restore the /var symlink so that subsequent builds continue to work) if
it were done as a target-finalize hook.

The only solution is to allow packages to register pre- and post-hooks
that are called right before and right after the rootfs commands are
executed, and inside in the fakeroot script.

We can however not re-use the BR2_ROOTFS_POST_FAKEROOT_SCRIPT feature
either because it is done before the filesystem command, but there is
nothing that is done after. Also, we don't want to add to, and modify a
user-supplied variable.

So, we introduce two new variables that packages can set to add the
commands they need to run to tweak the filesystem right at the last
moment.

Those hooks are not documented on-purpose; they are probably going to
only ever be used by systemd.

Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Reviewed-by: Romain Naour <romain.naour@gmail.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

system: separate sysv and systemd parts of the skeleton

For systemd, we create a simple /etc/fstab with only an entry for /, as
systemd otherwise automatically mounts what it needs where it needs it.

systemd does not like that the content of /var be symlinks to /tmp,
especially journald that starts before /tmp is mounted, and thus the
journal files are hidden from view, which causes quite a bit of fuss...

Instead, move the current /var to a sysv-only skeleton.

systemd at install time will create the /var content it needs, so we
just create an empty /var for systemd.

systemd would create /home and /srv at runtime if they are missing, but
it is better to create them right now, to simplify supporting systemd on
a RO filesystem in the (near) future.

Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Reviewed-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

package/skeleton: introduce sysv, systemd and none specific skeletons

Currently, we use the same skeleton for sysv-like init systems and
systemd, even though systemd has some peculiarities that makes our
default skeleton unfit.

So, we'll need to provide different skeletons (really, only part of
it) for sysv-like and systemd. In addition, in order to support the
"no init system" (BR2_INIT_NONE) use case, we introduce a "none"
skeleton.

Introduce three new skeleton packages, aptly named skeleton-sysv,
skeleton-systemd and skeleton-none. All three are providers of the
skeleton virtual package, in lieu of the skeleton-common package,
which is now a simple dependency of all three new skeletons.

Those packages are empty for now. In followup changes:
  - sysv-specific stuff will be moved out of skeleton-common and into
    skeleton-sysv;
  - systemd-specific stuff will be added to skeleton-systemd.

Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
[Arnout:
 - merge with the patch that enables the BR2_INIT_NONE case
 - simplify the BR2_PACKAGE_SKELETON_COMMON_ONLY select logic]
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
[Thomas:
 - remove the BR2_PACKAGE_SKELETON_COMMON_ONLY logic, and instead
   introduce a separate skeleton-none package for the BR2_INIT_NONE]
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

package/skeleton-common: simplify staging install

We don't really care what's going in staging, as long as it is properly
setup for merged/non-merged /usr, especially for the lib/ directory.

So we can just copy the skeleton as-is.

This simplifies maintenance, should we ever need to tweak the layout:
we'd just have to do it once in the skeleton directory to have it
propagated to both target and staging.

Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Reviewed-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

package/skeleton: make it a virtual package

We now have two packages that can act as a skeleton, skeleton-common,
also known as our default skeleton, and skeleton-custom.

This means that the skeleton package can be a standard virtual package
now.

Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

skeleton-common: rename SKELETON_ROOT_PASSWORD to SKELETON_COMMON_ROOT_PASSWORD

Package variables should always be prefixed with the package name. For
SKELETON_COMMON_ROOT_PASSWORD, this means we set the variable a second
time based on the value it previously had. Fortunately, this is fine
for make, even for recursively expanded variables.

This also allows to simplify the condition of the empty password - it's
not needed to set it again to empty if it already was empty.

Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

package/skeleton: split out into skeleton-common

Move all the handling of the default skeleton into a new package,
skeleton-common.

We don't name it skeleton-default, because it will be further split
later, into a skeleton for sysv and another for systemd, with some parts
still common between the two. So just name it skeleton-common right now;
this will save us a rename later.

While we're at it, also assign to SKELETON_COMMON_TARGET_FINALIZE_HOOKS
instead of directly to the global FINALIZE_HOOKS. Therefore, we don't
need to do all of that in a condition BR2_PACKAGE_SKELETON_COMMON==y.

Note: it would be technically sound to move the skeleton files together
within a sub-directory of the skeleton-common package. However, we refer
the user to those files, from various locations (manual, packages). It
will indeed be easier for the user to find those files in
system/skeleton/ rather than in package/skeleton-common/skeleton/

Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Reviewed-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
[Arnout: remove the mkdir $(STAGING_DIR)/usr/include which was removed
         in skeleton.mk in master.]
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

package/skeleton-custom: simplify target/staging install

For target install, we don't need to check the merged-usr symlinks
any more, it was already checked before.

For staging, instead of creating directories, just copy the same
skeleton, which was already checked to be correct.

Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
[Arnout: split off in a separate patch + wrote commit message]
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

package/skeleton-custom: also check for missing directories

In addition to checking the symlinks in the merged usr case, also check
whether /bin, /lib, /sbin and their /usr/... counterparts exist in the
non-merged case.

Missing directories are collected in the variable
SKELETON_CUSTOM_MISSING_DIRS. For symmetry and because it's more
logical, rename SKELETON_CUSTOM_NOT_MERGED_USR to
SKELETON_CUSTOM_NOT_MERGED_USR_DIRS.

When a directory is indeed missing, "stat" will print an error.
Buildroot *also* prints an error, which is clearer. So remove the error
from stat by redirecting it to /dev/null.

Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
[Arnout: split off in a separate patch + wrote commit message]
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

package/skeleton-custom: rework the merged_usr and building conditions

skeleton-custom.mk contains a huge condition for BR_BUILDING and for
BR2_ROOTFS_MERGED_USR. This makes the code difficult to follow, so
rework the condition a bit so that only the parts that really need to
be conditional are indeed conditional.

Note that there is no overhead in needlessly evaluation the "stat"
commands. Indeed, the assignments use late evaluation so the "stat"
is only executed when evaluating the condition - when skeleton-custom
is not selected, stat is never called.

Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
[Arnout: split off in a separate patch + wrote commit message]
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>