Recently, the autoconf macros for libtool started using '/usr/bin/file'
to determine the type of library that is generated by the toolchain.
Packages that use this recent version of the libtool autoconf macros
will fail in a rather dramatic way when /usr/bin/file is not present
on the host: the package will still build but no shared library is
generated, which in turn may cause build failures in other packages
that link with it.
For example, libpng's configure determines that it is not possible to
build a shared library on MIPS64 because the expected output from 'file'
is not present. Therefore, only a static libpng.a is built. Later,
bandwithd links with -lpng but it doesn't use the pkg-config's
Private-Libs (because it's not linking statically) and it doesn't have
access to the NEEDED reference from the shared library. Therefore, it
doesn't link with zlib and fails with
pngrutil.c:(.text+0x55c): undefined reference to `inflate'
We cant use host-file because it is itself an autotools package and is
itself using libtool, so this would be a chicken-n-egg problem. Besides,
the libtool script really wants to call /usr/bin/file, so it would not
even find our host-file anyway.
So, just require that '/usr/bin/file' is present on the host.
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Arnout Vandecappelle <arnout@mind.be>
Cc: Baruch Siach <baruch@tkos.co.il>
Reviewed-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Currently, we require a host with an UTF-8 locale as soone as we build a
Buildroot toolchain with support for locales. This means that we do
erquire such a locale when building a toolchain with either uClibc,
glibc or musl.
However, glibc and musl do not require such a locale to be present.
Use the new option to check if an UTF8 locale is needed on the host,
rather than deriving that from toolchain settings.
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
ExtUtils::MakeMaker is one of the Perl Core modules usually packaged in
Perl package for a Debian/Ubuntu based system.
For a Fedora based system, each Perl Core modules have their own RPM
package. So install only Perl package is not enough.
Fixes:
>>> host-libxml-parser-perl 2.41 Configuring
[...]
perl `which perl` Makefile.PL
Can't locate ExtUtils/MakeMaker.pm in @INC (you may need to install the ExtUtils::MakeMaker module)
Add a new Perl module check in dependency.sh.
Signed-off-by: Romain Naour <romain.naour@gmail.com>
Cc: François Perrad <francois.perrad@gadz.org>
Reviewed-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
[Thomas:
- Check for Thread::Queue, not Thread:Queue.
- Use 'printf' instead of 'echo -e', since printf is POSIX, but not
'echo -e'.]
Signed-off-by: Vicente Olivert Riera <Vincent.Riera@imgtec.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
The screen is cluttered when we build for 32 bit target and 32 bit gcc
is missing.
~/buildroot$ make
[...]
/usr/bin/ld: cannot find crt1.o: No such file or directory
/usr/bin/ld: cannot find crti.o: No such file or directory
/usr/bin/ld: skipping incompatible /usr/lib/gcc/x86_64-linux-gnu/4.9/libgcc.a when searching for -lgcc
/usr/bin/ld: cannot find -lgcc
/usr/bin/ld: skipping incompatible /usr/lib/gcc/x86_64-linux-gnu/4.9/libgcc_s.so when searching for -lgcc_s
/usr/bin/ld: cannot find -lgcc_s
/usr/bin/ld: cannot find -lc
/usr/bin/ld: skipping incompatible /usr/lib/gcc/x86_64-linux-gnu/4.9/libgcc.a when searching for -lgcc
/usr/bin/ld: cannot find -lgcc
/usr/bin/ld: skipping incompatible /usr/lib/gcc/x86_64-linux-gnu/4.9/libgcc_s.so when searching for -lgcc_s
/usr/bin/ld: cannot find -lgcc_s
/usr/bin/ld: cannot find crtn.o: No such file or directory
collect2: error: ld returned 1 exit status
[...]
Your Buildroot configuration needs a compiler capable of building 32 bits binaries.
The final note is enough, and adding 2>/dev/null to the gcc test
invocation is also more consistent with the rest of the script. The
patch makes the '/usr/bin/ld:' and 'collect2:' lines go away.
Signed-off-by: Jens Stimpfle <debian@jstimpfle.de>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
When g++ is not installed, a misleading error message turns up because
of a bad combination of an unquoted shell variable and control flow.
~/buildroot$ make
You may have to install 'g++' on your build machine
/home/testuser/buildroot/support/dependencies/dependencies.sh: 136: [: -lt: unexpected operator
[Thomas:
- fixed commit log, as per the suggestion of Yann E. Morin.
- don't change existing empty new lines, suggested by Yann.
- use positive logic in the newly added test, suggested by Yann.]
Signed-off-by: Jens Stimpfle <debian@jstimpfle.de>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
When we set LD_LIBRARY_PATH when building our host tools, we append any
pre-existing value to our custom path:
LD_LIBRARY_PATH="$(HOST_DIR)/usr/lib:$(LD_LIBRARY_PATH)"
But then if LD_LIBRARY_PATH was previously empty, we end up with an
LD_LIBRARY_PATH that ends with a colon.
Also, when we check that an existing LD_LIBRARY_PATH does not contain
CWD, we previously did not look for a zero-length prefix.
Since 'man ld.so' says of LD_LIBRARY_PATH:
A colon-separated list of directories in which to search for ELF
libraries at execution-time. Similar to the PATH environment
variable.
And POSIX states about PATH:
A zero-length prefix is a legacy feature that indicates the current
working directory.
And bash also recognises a zero-length prefix to search in CWD:
A zero-length (null) directory name in the value of PATH indicates
the current directory.
We may thus end up on a system where a zero-length prefix in
LD_LIBRARY_PATH is interpreted as CWD.
Do not append the previous LD_LIBRARY_PATH if it was empty, and check
for a zero-length prefix when checking dependencies.
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Cc: Peter Korsgaard <jacmet@uclibc.org>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Signed-off-by: Vincent Stehlé <vincent.stehle@laposte.net>
Acked-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Remove the specific check that was done in dependencies.sh to use the
generic one that were introduced by the previous patch.
Also, introduce, BR2_NEEDS_HOST_JAVAC and BR2_NEEDS_HOST_JAR as it is
needed by classpath.
Signed-off-by: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Reviewed-by: Thomas De Schampheleire <thomas.de.schampheleire@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Avoid copy/pasting the same block of code to check if a program is
available on the host machine.
Signed-off-by: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Reviewed-by: Thomas De Schampheleire <thomas.de.schampheleire@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
The current check is broken, as it only checks if the user's login
shell is bash, not what the system shell is.
Mimick the sequence found in the top-level Makefile to search for
bash, except for the fallback case, where we explicitly check that
'sh' is bash, by checking if it sets $BASH, so we know the fallback
case, in the top-level Makefile, to use 'sh' will indeed use bash.
Remove superfluous semi-colons ';' at the end of lines, they are
not needed in a shell script (this is not C!)
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Cc: Peter Korsgaard <jacmet@uclibc.org>
Cc: Andrew Barnes <andy@outsideglobe.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Change BR2_HOST_NEEDS_JAVA to BR2_NEEDS_HOST_JAVA as it makes more
sense.
The host doesn't need Java but Buildroot needs the host to have
Java in order to build the package that select this option.
Signed-off-by: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
XBMC needs Java on the host in order to build, because it uses a
code-generator which is built in two phases: In the first phase SWIG is used
to parse C++ header files that define the API. SWIG outputs an XML file
that contains a complete description of the structure of the API. In the
second phase, the XML file is ingested by a Groovy (Java) program that then
creates C++ code that forms the bridge to the scripting language (Python).
The second phase is why we need java on the host.
You can learn more at the XBMC's wiki:
http://wiki.xbmc.org/index.php?title=Codegeneration#How_it_works
In order to check that, this patch introduce this mechanism in
dependencies.sh, and it also defines the variable in Config.in
[Peter: fix error message]
Signed-off-by: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
To make the naming consistent (all user-visible options should be
prefixed with BR2_).
An entry is added to Makefile.legacy to warn users who have set
BUILDROOT_CONFIG but not BR2_CONFIG.
Still export BUILDROOT_CONFIG but pointing to some phony value, to
make sure that scripts that still use it fail in a predictable way.
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Tested-by: Jérémy Rosen <jeremy.rosen@openwide.fr>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Cleanup mixed indents and remove commented lines.
Signed-off-by: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
If the grub package is selected it also selects
BR2_HOSTARCH_NEEDS_IA32_COMPILER. This triggers a test in dependencies.sh
to verify the host compiler can build 32 bit executables. Currently this
test does not set any output for the compiler which causes a stray a.out
to be create outside the output directory. This patch sets the compiler
output to /dev/null so no a.out is created but the test is still performed.
Signed-off-by: Martin Bark <martin@barkynet.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
'printf' was introduced because it is more portable than 'echo -e'. But
when the escape sequences are just newlines we can just as well use
plain 'echo' (and remove the newline escape sequences).
This looks cleaner than having some lines with echo and some with
printf.
Signed-off-by: Bjørn Forsman <bjorn.forsman@gmail.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
support/dependencies/dependencies.sh uses #!/bin/sh shebang. It is not
guaranteed that /bin/sh provides an 'echo' implementation that
understands the '-e' flag (interpret backslash escape chars). For
example, dash doesn't.
'printf' is more portable (it must interpret backslash escape chars,
according to POSIX), so use that.
NOTE: Before the previous commit, the dependencies.sh script used
/bin/echo instead of the shell built-in. That's probably why this hasn't
come up before.
Signed-off-by: Bjørn Forsman <bjorn.forsman@gmail.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Buildroot fails to run on NixOS because it has no /bin/echo or
/bin/grep. Instead of relying on absolute paths, rely on tools to be
available in PATH. This should work for all systems.
Signed-off-by: Bjørn Forsman <bjorn.forsman@gmail.com>
Acked-by: Thomas De Schampheleire <thomas.de.schampheleire@gmail.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Some bootloaders (such as Grub) need to be built for x86 (i.e IA32)
even if the target architecture is x86-64. However, when the target
architecture is x86-64, the cross-compiler generated by Buildroot is
not able to generate 32 bits code.
To solve this, we will rely on the host compiler being a x86 + x86-64
compiler. Therefore, this commit introduces the
BR2_HOSTARCH_NEEDS_IA32_COMPILER option, which tells the dependency
checking logic to verify that the host compiler is indeed capable of
building x86 32 bits code.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
This is the list needed to run the Linaro pre-built toolchain
on a 64-bit Ubuntu 13.10 system.
Signed-off-by: Frank Hunleth <fhunleth@troodon-software.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Since the addition of root password setting support in buildroot, there have
been a few bug reports in this area ([1], [2]). In these cases, the system
mkpasswd did either not work, or did not provide the options we expect, like
-m <method>.
This patch adds a mkpasswd host package, based on the sources from whois. When
a non-empty root password is set, this package is used as a dependency.
Signed-off-by: Thomas De Schampheleire <thomas.de.schampheleire@gmail.com>
[1] http://lists.busybox.net/pipermail/buildroot/2013-July/075771.html
[2] http://lists.busybox.net/pipermail/buildroot/2013-July/075869.html
[Thomas P: use $(INSTALL) instead of install, put -lcrypt at the end
of build command line to allow gcc to find the crypt() function in
lcrypt.]
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Since a few kernel releases, having 'bc' installed is mandatory to
build the kernel. See commit
https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=70730bca1331fc50c3caacaea00439de1325bd6e
of the kernel.
Since this tool is generally available by default in distributions,
and we're unlikely to see version-specific problems with it, we
just check for it to be installed in
support/dependencies/dependencies.sh.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Peter Korsgaard <jacmet@sunsite.dk>
This reverts commit d66cd067f3.
SSL certificates are no always installed in /etc/ssl/certs. For example, on
CentOS 5.6 the default OpenSSL certificates directory is /etc/pki/tls/certs,
and wget can download using https without any problem.
Moreover, the existence of /etc/ssl/certs does not guarantee the presence of a
CA certificates bundle even on Debian. On my current Debian testing
installation the openssl package itself creates an empty /etc/ssl/certs
directory.
Signed-off-by: Baruch Siach <baruch@tkos.co.il>
Acked-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Peter Korsgaard <jacmet@sunsite.dk>
Some packages are hosted on https:// servers, and wget only works on
these if the SSL certificates are installed. For example, downloading
the kernel sources from kernel.org requires those SSL certificates to
be installed.
[Peter: fix typo]
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Peter Korsgaard <jacmet@sunsite.dk>
The host-autoconf build process requires a full Perl installation, or
at least a Perl installation that has the Data::Dumper module
installed. On a basic Debian system, only 'perl-base' is installed,
but Data::Dumper is in the 'perl' package.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Peter Korsgaard <jacmet@sunsite.dk>
The current dependencies code abort as soon as one program is
missing. It is quite annoying when multiple programs are
missing. Instead, bail out if needed after testing all programs.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Peter Korsgaard <jacmet@sunsite.dk>
awk, bison, flex, makeinfo, gettext should be built as dependencies of
packages when needed. In practice, even the toolchain build doesn't
need any of these, and only a few packages do require them.
It is not needed to list gzip and bzip2 since they are already checked
through ${DL_TOOLS}: whenever a package needs gzip or bzip2 for its
extraction, the dependency is added.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Peter Korsgaard <jacmet@sunsite.dk>
No need to recreate a path we already have.
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Signed-off-by: Peter Korsgaard <jacmet@sunsite.dk>
Many users trying to use external toolchains on x86-64 machines get a
very confusing message:
"Can't execute cross-compiler"
They get this message because they forgot to install the 32 bits
compatibility libraries that are needed to run binaries compiled for
x86 on x86-64 machines.
Since this is the case for both external toolchains and certain
binary-only tools like SAM-BA, we add a new Kconfig option
BR2_HOSTARCH_NEEDS_IA32_LIBS, that packages must select if they need
the 32 bits compatibility libraries. When this option is enabled,
dependencies.sh checks that the 32 bits dynamic library loader is
present on the system, and if not, it stops and shows an error.
The path and name of the 32 bits dynamic loader is hardcoded because
it is very unlikely to change, as it would break the ABI for all
binaries.
Also, it is worth noting that the check will be done even if we're
running on a 32 bits machine. This is harmless, as 32 bits machines
necessarily have the 32 bits dynamic loader installed, so the error
will never show up in this case.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Peter Korsgaard <jacmet@sunsite.dk>
Add an option in the menuconfig to specify a root password.
If set to empty, no root password is created; otherwise, the password is
encrypted using MD5 (MD5 is not the default for crypt(3), DES-56 is, but
MD5 is widely available, not-so-strong, but not-so-weak either).
Add a check for 'mkpasswd' as a new dependency.
[Peter: fix typo/capitilization and simplify logic]
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Arnout Vandecappelle <arnout@mind.be>
Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Peter Korsgaard <jacmet@sunsite.dk>
We should instead simply unset it at runtime, like we do for
PKG_CONFIG_PATH.
This reverts commit 9910eba33a.
Signed-off-by: Peter Korsgaard <jacmet@sunsite.dk>
Having DESTDIR set in the environment before running Buildroot creates
some funky problems in the build process. Prevent users from running
into this kind of troubles.
Cc: Michael Opdenacker <michael.opdenacker@free-electrons.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Peter Korsgaard <jacmet@sunsite.dk>
As suggested by Arnout Vandecappelle, move toolchain/dependencies to
support/dependencies, as it really is not toolchain-specific anymore.
Signed-off-by: Thomas De Schampheleire <thomas.de.schampheleire@gmail.com>
Acked-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Acked-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Peter Korsgaard <jacmet@sunsite.dk>
