I'm using Yocto to install clBLAS library (https://github.com/clMathLibraries/clBLAS) using the recipe https://github.com/CogentEmbedded/meta-opencl/blob/master/meta-ocl-common/recipes-graphics/clblas/clblas_git.bb
But I'm getting the below warning everytime & .so file is not present in the built image.
WARNING: QA Issue: clblas: Files/directories were installed but not shipped in any package:
/usr/lib
/usr/lib/libclBLAS.so.2.12.0
/usr/lib/libclBLAS.so.2
/usr/lib/libclBLAS.so
/usr/lib/.debug
/usr/lib/pkgconfig
/usr/lib/cmake
/usr/lib/.debug/libclBLAS.so.2.12.0
/usr/lib/pkgconfig/clBLAS.pc
/usr/lib/cmake/clBLAS
/usr/lib/cmake/clBLAS/clBLASTargets-debug.cmake
/usr/lib/cmake/clBLAS/clBLASConfigVersion.cmake
/usr/lib/cmake/clBLAS/clBLASTargets.cmake
/usr/lib/cmake/clBLAS/clBLASConfig.cmake
Please set FILES such that these items are packaged. Alternatively if they are unneeded, avoid installing them or delete them within do_install.
clblas: 14 installed and not shipped files. [installed-vs-shipped]
How to overcome this warning & make the .so file present in the target's /usr/lib folder?
Add below lines to your clblas_git.bb
FILES_${PN} += "${libdir}/*"
FILES_${PN}-dev = "${libdir}/* ${includedir}"
For good explanation you will get it here
If you are using new version of yocto this will help:
FILES:${PN} ="name of the dirs is not shipping";
In old versions it is FILES_${PN}.
The problem is that multilib is not considered correctly during build, looking at cmake files in clBLAS, its using CMake variable for constructing multilib path SUFFIX_LIB and yocto recipe is setting it to be empty here however its not encoding the yocto logic for multilib paths. A potential fix would be in recipe as below
--- clblas_git.bb.org 2019-12-07 12:41:56.784649031 -0800
+++ clblas_git.bb 2019-12-07 12:42:25.317982206 -0800
## -16,7 +16,7 ## S = "${WORKDIR}/git/src"
inherit cmake pythonnative
-EXTRA_OECMAKE += "-DSUFFIX_LIB= -DUSE_SYSTEM_GTEST=ON -DBUILD_TEST=OFF -DPREBUILT_CLT_PATH=${WORKDIR}/clt"
+EXTRA_OECMAKE += "-DSUFFIX_LIB=${#d.getVar('baselib', True).replace('lib', '')} -DUSE_SYSTEM_GTEST=ON -DBUILD_TEST=OFF -DPREBUILT_CLT_PATH=${WORKDIR}/clt"
DEPENDS += "virtual/opencl"
Related
I new to Yocto so there are probably some mistakes and misunderstanding that I've had, I appreciate if you can help.
So, I want to add a new package (tar file) to my custom image.
I have followed steps and steps in manual and some online instructions. While running: "bitbake mylayer", the layer is built fine but I got this error while building the image, here is the log file:
DEBUG: Executing python function rootfs_deb_bad_recommendations
DEBUG: Python function rootfs_deb_bad_recommendations finished
DEBUG: Executing python function extend_recipe_sysroot
NOTE: Installed into sysroot: []
NOTE: Skipping as already exists in sysroot: ['depmodwrapper-cross', 'apt-native', 'dpkg-native', 'pseudo-native', 'update-rc.d-native', 'prelink-native', 'makedevs-native', 'ldconfig-native', 'opkg-util$
DEBUG: Python function extend_recipe_sysroot finished
DEBUG: Executing python function do_rootfs
NOTE: ###### Generate rootfs #######
NOTE: Installing the following packages: apt busybox copy-uefiimg-to-sda coreutils dpkg e2fsprogs-resize2fs libfontconfig1 libfreetype6 libglib-2.0-0 gptfdisk libjemalloc2 kernel-module-axi-dma-sensor ku$
ERROR: Unable to install packages.
Reading package lists...
Building dependency tree...
Reading state information...
Package mypackage is not available, but is referred to by another package.
This may mean that the package is missing, has been obsoleted, or
is only available from another source
E: Package 'mylayer' has no installation candidate
DEBUG: Python function do_rootfs finished
ERROR: Function failed: do_rootfs
And here is mylayer.bb:
SUMMARY = ""
LICENSE = "CLOSE"
FILESEXTRAPATHS_prepend := "${THISDIR}/files:"
SRC_URI += "file://mypackage.tar"
Also, I have included the package in conf/local.conf:
IMAGE_INSTALL_append += " mylayer"
So beside trying to figure out how to solve this problem, I also have some questions:
I have read some example of .bb, and they mentioned about LIC_FILES_CHKSUM. The mypackage.tar.gz is a package to install a platform for the device and I don't know much about the source code, so I don't know if it is necessary to include the license? Or how to know that this package need license to install?
In some answer I found online, there is one saying that I need to include PACKAGE_CLASSES ?= "package_deb" (they want to install the .deb file), so probably in my case I will need PACKAGE_CLASSES ?= "package.tar" right? I have tried to change variable, but it still not successful.
The mypackage.tar includes some deb files. If I could not install mypackage.tar, can I instead install these .deb files? Can I put it all in mylayer.bb?
Thank you in advanced, I have tried to study much documents as I could but I get so confused and there is huge amount of information to digest.
First, before answering your questions
Let me mention some best practices advice for you:
Rename the recipe to some significant name related to you compressed package.
Naming the recipe to mylayer confuses Yocto users, because there is the term layer also.
Regarding you recipe:
There is no need for FILESEXTRAPATHS because the recipe path is added automatically to Yocto paths.
FILESEXTRAPATHS it is required for .bbappend files.
You need to override the do_install task function, it does nothing by default.
do_install is the first essential task to make sure that your sources are included in the final image.
Beside that, when specifying a compressed source file into SRC_URI, yocto automatically decompresses it.
This is mentioned here.
So, here what your recipe should look like:
SUMMARY = ""
LICENSE = "CLOSED"
# Prevent Yocto from decompressing the file
SRC_URI = "file://mypackage.tar;unpack=0"
do_install(){
# Create the opt folder into the final image, ${D} is ${WORKDIR}/image
install -d ${D}/opt
# Copy the compressed file there; You can change permissions as you want
install -m 0755 ${WORKDIR}/mypackage.tar ${D}/opt
}
# Very important to specify what you installed in (do_install)
FILES_${PN} = "/opt/*"
Now, when you run IMAGE_INSTALL_append += " mylayer" your file will be installed.
Regarding your questions:
You mentioned that your compressed file contains .deb files, I assume that no license checksum is needed. Also, I understand that you may wanted to point to SRC_URI[md5sum] or other checksums for the full package. That is also not needed for local files, it is used to check for the integrity of online sources.
PACKAGE_CLASSES as mentioned here, is used by the system to know in what type the data should be packaged. By the data I mean the data that you installed with do_install. That data get packaged for according to your PACKAGE_CLASSES variable, for example, to deb file. And that is used, along side with all other recipes packages, to build the final rootfs.
Yes, if you are installing the tar file into the image and then unpack it to install all deb files, for example, with dpkg. You can use the bin_package class to do that, now the recipe must be changed for that reason:
Decompress the tar file and provide the deb files in the local files folder.
Add all deb files to SRC_URI
Inherit the bin_package class
Specify the files to be packaged.
Your recipe should look like this:
SUMMARY = ""
LICENSE = "CLOSED"
SRC_URI = "file://deb_file1.deb \
file://deb_file2.deb"
# No need to `do_install` , it is invoked by the (bin_package) class
FILES_${PN} = ""
Important:
About FILES_${PN}, you need to add all what the deb installed into the image folder
Example, if your deb file installs this:
/usr/bin/hello
/etc/hello.cfg
Specify them:
FILES_${PN} = "/usr/bin/*"
FILES_${PN} += "/etc/*"
Use * so if other deb files install files into the same folder as others it will include all.
I am trying to build a custom recipe with Yocto (Rocko) for my Linux i.MX6 based embedded system.
The main recipe had dependency on a other custom recipe(as the main recipe is using header-files from this recipe) which is also creating some binaries which needs to be included in the final image.
I have added other_recipe(nbdkit) in the "DEPENDS" of main_recipe.bb
DEPENDS += "nbdkit"
the main recipe is creating a .so with the help of its own source file which is including header-files from the 'kit' recipe. I am able to install the binaries & .so generated using this main_recipe by adding it in do_install().
Now in the other_recipe(nbdkit http://cgit.openembedded.org/meta-openembedded/tree/meta-networking/recipes-support/nbdkit/nbdkit_git.bb?h=master), When I add do_install()to include the binaries generated from that recipe the main_recipe build gets failed with PKG_CONFIG error as follow,
| Package nbdkit was not found in the pkg-config search path.
| Perhaps you should add the directory containing `nbdkit.pc'
| to the PKG_CONFIG_PATH environment variable
| No package 'nbdkit' found
Other build errors says that the header-files of kit included in the main_recipe is not found.
app-nbdkit-plugin.c:2:10: fatal error: nbdkit-plugin.h: No such file or directory
Where app-nbdkit-plugin.c is source file of main_recipe & kit-plugin.h is header file of other_recipe.
The strange thing is, when I remove do_install() from other_recipe(nbdkit) the main_recipe is getting built successfully.
Now I doubt, Is it possible to set a recipe as DEPENDS of other recipe and at the same time it provides output file as do_install()?
Will sharing header-files from other_recipe to main_recipe resolves the issue? If yes, how to do that?
Thanks.
[EDIT] Added nbdkit recipe link.
I am trying to install a set of CMake utilty functions under /usr/share/cmake-3.4/Modules/MYMODULES/useful.cmake within a Yocto build.
Here is my current (sanitized) recipe (call it my-useful-modules.bb)
SECTION = "devel"
LICENSE = "CLOSED"
inherit cmake
EXTERNALSRC := "path/to/source/code"
do_compile() {
:
}
FILES_${PN} += "${datadir}/cmake-3.4/Modules/MYMODULES/*"
BBCLASSEXTEND = "nativesdk"
The configure and install tasks work fine, and if I look under image in tmp/work/..., I see the full tree (including all my host directories as expected).
But I keep getting the following error
ERROR: nativesdk-my-useful-modules-1.0-r0 do_package: QA Issue:
nativesdk-my-useful-modules: Files/directories were installed but not
shipped in any package:
Followed by a long list of files which basically includes everything under image.
These modules need to be available both in the native sysroot during the build, and in the standard SDK built with populate_sdk.
Which package should I be specifying with FILES_${PN} to get them packaged?
I'd also appreciate knowing how to either avoid specifying the cmake version in the FILES statement, or get it from the build system.
Updating FILES_${PN} will resolve your error and also there is no need to provide cmake version in this case.
FILES_${PN} += "${datadir}/*"
I am trying to create an eclipse project from a cmake project .
I used the following command
cmake -G "Eclipse CDT4 - Unix Makefiles" ./`
it gives the following error
CMake Error at CMakeLists.txt:119 (find_package):
By not providing "FindGlib.cmake" in CMAKE_MODULE_PATH this project has
asked CMake to find a package configuration file provided by "Glib", but
CMake did not find one.
Could not find a package configuration file provided by "Glib" (requested
version 2.28) with any of the following names:
GlibConfig.cmake
glib-config.cmake
Add the installation prefix of "Glib" to CMAKE_PREFIX_PATH or set
"Glib_DIR" to a directory containing one of the above files. If "Glib"
provides a separate development package or SDK, be sure it has been
installed.
-- Configuring incomplete, errors occurred!
I have glib installed . actually it couldn't resolve the path i guess. wherever find is there in cmake file , it is giving the smiler errors. please i suggest a way out, i badly need to load this project in cmake. Thanks.
Here is line 119 where error message is pointing
find_package(Glib 2.28 REQUIRED)
include_directories(${Glib_INCLUDE_DIRS})
list(APPEND LIBS ${Glib_LIBRARIES})
add_definitions(${Glib_DEFINITIONS})
When you call find_package(MyPackage) in a CMake file, it tries to find a FindMyPackage.cmake configuration in its system path (/usr/share/cmake-2.8/Modules on my Ubuntu box), or in the directory you did specify as CMAKE_MODULE_PATH).
The solution to your problem is to create a directory for modules in your source tree (e.g. CMakeModules), put in it a FindGlib.cmake file that you can find using Google, and add
set(CMAKE_MODULE_PATH ${PROJECT_SOURCE_DIR}/CMakeModules)
in your CMakeLists.txt before the actual call to find_package.
(your problem is not related to the Eclipse generator, you could remove that from the title of the question).
I use Android NDK with cygwin with Eclipse on Windows.
In my makefile I want to set path of prebuild library using environment variable in eclipse.
So I do the following:
And in makefile:
LOCAL_SRC_FILES = $(QCAR_SDK_ROOT)build/lib/$(TARGET_ARCH_ABI)/libQCAR.so
LOCAL_EXPORT_C_INCLUDES := $(QCAR_SDK_ROOT)build/include
But I get error:
Android NDK: ERROR:jni/Android.mk:QCAR-prebuilt: LOCAL_SRC_FILES points to a missing file
/cygdrive/d/Development/Android/android-ndk-r7/build/core/prebuilt-library.mk:43: *** Android NDK: Aborting . Stop.
make: *** [all] Error 2
Android NDK: Check that jni//cygdrive/D/Development/Android/qcar-android-1-5-4-beta1/build/lib/armeabi/libQCAR.so exists or that its path is correct
Tried defining variable in makefile directly, got the same result.
So, obviously, that path is not what I wanted. How do I set the proper path?
Why not just specify the paths in the make file? (N.B Just noticed you tried that.)
Using Cygwin with Android NDK
This site seems to indicate that the make file won't run correctly within Eclipse and you should run in via Windows Explorer. Are you running this within Eclipse? Try this and see if you still get the issues.
That's a problem of the previous NDK builds that Google fixed with NDK-9. "Updated ndk-build to support absolute paths in LOCAL_SRC_FILES."
See the release notes here:http://developer.android.com/tools/sdk/ndk/index.html
Try to play with LOCAL_PATH variable. As documentation (docs/ANDROID-MK.html in Android NDK package, or here) states:
LOCAL_SRC_FILES
This is a list of source files that will be built for your module.
Only list the files that will be passed to a compiler, since the
build system automatically computes dependencies for you.
Note that source files names are all relative to LOCAL_PATH and
you can use path components
Also, NDK hints you to Check that jni//cygdrive/D/Development/Android/qcar-android-1-5-4-beta1/build/lib/armeabi/libQCAR.so exists or that its path is correct.
Thus, I would try the following:
LOCAL_PATH := /
...or to reset it at all:
LOCAL_PATH :=
You can edit eclipse.ini file and add it there.
e.g. -DLOCAL_SRC_FILES=/home/user/.../
Or declare a path variable. It is a convenient way of sharing a common location among multiple projects within a workspace.
Hope that help you!
Recent NDK releases on Windows do not need cygwin. Worse, they do not recognize the cygdrive notation. You can simply use
QCAR_SDK_ROOT = D:/Development/Android/qcar-android-1-5-4-beta1
correction absolute paths for LOCAL_SRC_FILES do not work for ndk.r7, and even for r9 the ANDROID-MK.doc does not encourage using absolute paths there.