%poky; ] > This chapter describes common usage for the Yocto Project. The information is introductory in nature as other manuals in the Yocto Project documentation set provide more details on how to use the Yocto Project.

This section provides a summary of the build process and provides information for less obvious aspects of the build process. For general information on how to build an image using the OpenEmbedded build system, see the "Building Images" section of the Yocto Project Quick Start.

In the development environment you will need to build an image whenever you change hardware support, add or change system libraries, or add or change services that have dependencies. Building an Image The first thing you need to do is set up the OpenEmbedded build environment by sourcing an environment setup script (i.e. &OE_INIT_FILE; or oe-init-build-env-memres). Here is an example: $ source &OE_INIT_FILE; [build_dir] The build_dir argument is optional and specifies the directory the OpenEmbedded build system uses for the build - the Build Directory. If you do not specify a Build Directory, it defaults to a directory named build in your current working directory. A common practice is to use a different Build Directory for different targets. For example, ~/build/x86 for a qemux86 target, and ~/build/arm for a qemuarm target. Once the build environment is set up, you can build a target using: $ bitbake target The target is the name of the recipe you want to build. Common targets are the images in meta/recipes-core/images, meta/recipes-sato/images, etc. all found in the Source Directory. Or, the target can be the name of a recipe for a specific piece of software such as BusyBox. For more details about the images the OpenEmbedded build system supports, see the "Images" chapter. Building an image without GNU General Public License Version 3 (GPLv3), or similarly licensed, components is supported for only minimal and base images. See the "Images" chapter for more information.

When building an image using GPL components, you need to maintain your original settings and not switch back and forth applying different versions of the GNU General Public License. If you rebuild using different versions of GPL, dependency errors might occur due to some components not being rebuilt.

Once an image has been built, it often needs to be installed. The images and kernels built by the OpenEmbedded build system are placed in the Build Directory in tmp/deploy/images. For information on how to run pre-built images such as qemux86 and qemuarm, see the Yocto Project Software Development Kit (SDK) Developer's Guide. For information about how to install these images, see the documentation for your particular board or machine.

The exact method for debugging build failures depends on the nature of the problem and on the system's area from which the bug originates. Standard debugging practices such as comparison against the last known working version with examination of the changes and the re-application of steps to identify the one causing the problem are valid for the Yocto Project just as they are for any other system. Even though it is impossible to detail every possible potential failure, this section provides some general tips to aid in debugging. A useful feature for debugging is the error reporting tool. Configuring the Yocto Project to use this tool causes the OpenEmbedded build system to produce error reporting commands as part of the console output. You can enter the commands after the build completes to log error information into a common database, that can help you figure out what might be going wrong. For information on how to enable and use this feature, see the "Using the Error Reporting Tool" section in the Yocto Project Development Manual. For discussions on debugging, see the "Debugging With the GNU Project Debugger (GDB) Remotely" section in the Yocto Project Developer's Manual and the "Working within Eclipse" section in the Yocto Project Software Development Kit (SDK) Developer's Guide. The remainder of this section presents many examples of the bitbake command. You can learn about BitBake by reading the BitBake User Manual.

The log file for shell tasks is available in ${WORKDIR}/temp/log.do_taskname.pid. For example, the do_compile task for the QEMU minimal image for the x86 machine (qemux86) might be tmp/work/qemux86-poky-linux/core-image-minimal/1.0-r0/temp/log.do_compile.20830. To see what BitBake runs to generate that log, look at the corresponding run.do_taskname.pid file located in the same directory. Presently, the output from Python tasks is sent directly to the console.

Any given package consists of a set of tasks. The standard BitBake behavior in most cases is: do_fetch, do_unpack, do_patch, do_configure, do_compile, do_install, do_package, do_package_write_*, and do_build. The default task is do_build and any tasks on which it depends build first. Some tasks, such as do_devshell, are not part of the default build chain. If you wish to run a task that is not part of the default build chain, you can use the -c option in BitBake. Here is an example: $ bitbake matchbox-desktop -c devshell If you wish to rerun a task, use the -f force option. For example, the following sequence forces recompilation after changing files in the work directory. $ bitbake matchbox-desktop . . make some changes to the source code in the work directory . . $ bitbake matchbox-desktop -c compile -f $ bitbake matchbox-desktop This sequence first builds and then recompiles matchbox-desktop. The last command reruns all tasks (basically the packaging tasks) after the compile. BitBake recognizes that the do_compile task was rerun and therefore understands that the other tasks also need to be run again. You can view a list of tasks in a given package by running the do_listtasks task as follows: $ bitbake matchbox-desktop -c listtasks The results appear as output to the console and are also in the file ${WORKDIR}/temp/log.do_listtasks.

Sometimes it can be hard to see why BitBake wants to build other packages before building a given package you have specified. The bitbake -g targetname command creates the pn-buildlist, pn-depends.dot, package-depends.dot, and task-depends.dot files in the current directory. These files show what will be built and the package and task dependencies, which are useful for debugging problems. You can use the bitbake -g -u depexp targetname command to display the results in a more human-readable form.

You can see debug output from BitBake by using the -D option. The debug output gives more information about what BitBake is doing and the reason behind it. Each -D option you use increases the logging level. The most common usage is -DDD. The output from bitbake -DDD -v targetname can reveal why BitBake chose a certain version of a package or why BitBake picked a certain provider. This command could also help you in a situation where you think BitBake did something unexpected.

Sometimes issues on the host development system can cause your build to fail. Following are known, host-specific problems. Be sure to always consult the Release Notes for a look at all release-related issues. glibc-initial fails to build: If your development host system has the unpatched GNU Make 3.82, the do_install task fails for glibc-initial during the build. Typically, every distribution that ships GNU Make 3.82 as the default already has the patched version. However, some distributions, such as Debian, have GNU Make 3.82 as an option, which is unpatched. You will see this error on these types of distributions. Switch to GNU Make 3.81 or patch your make to solve the problem.

To build a specific recipe (.bb file), you can use the following command form: $ bitbake -b somepath/somerecipe.bb This command form does not check for dependencies. Consequently, you should use it only when you know existing dependencies have been met. You can also specify fragments of the filename. In this case, BitBake checks for a unique match.

You can use the -e BitBake option to display the parsing environment for a configuration. The following displays the general parsing environment: $ bitbake -e This next example shows the parsing environment for a specific recipe: $ bitbake -e recipename

Best practices exist while writing recipes that both log build progress and act on build conditions such as warnings and errors. Both Python and Bash language bindings exist for the logging mechanism: Python: For Python functions, BitBake supports several loglevels: bb.fatal, bb.error, bb.warn, bb.note, bb.plain, and bb.debug. Bash: For Bash functions, the same set of loglevels exist and are accessed with a similar syntax: bbfatal, bberror, bbwarn, bbnote, bbplain, and bbdebug. For guidance on how logging is handled in both Python and Bash recipes, see the logging.bbclass file in the meta/classes folder of the Source Directory.

When creating recipes using Python and inserting code that handles build logs, keep in mind the goal is to have informative logs while keeping the console as "silent" as possible. Also, if you want status messages in the log, use the "debug" loglevel. Following is an example written in Python. The code handles logging for a function that determines the number of tasks needed to be run. See the "do_listtasks" section for additional information: python do_listtasks() { bb.debug(2, "Starting to figure out the task list") if noteworthy_condition: bb.note("There are 47 tasks to run") bb.debug(2, "Got to point xyz") if warning_trigger: bb.warn("Detected warning_trigger, this might be a problem later.") if recoverable_error: bb.error("Hit recoverable_error, you really need to fix this!") if fatal_error: bb.fatal("fatal_error detected, unable to print the task list") bb.plain("The tasks present are abc") bb.debug(2, "Finished figuring out the tasklist") }

When creating recipes using Bash and inserting code that handles build logs, you have the same goals - informative with minimal console output. The syntax you use for recipes written in Bash is similar to that of recipes written in Python described in the previous section. Following is an example written in Bash. The code logs the progress of the do_my_function function. do_my_function() { bbdebug 2 "Running do_my_function" if [ exceptional_condition ]; then bbnote "Hit exceptional_condition" fi bbdebug 2 "Got to point xyz" if [ warning_trigger ]; then bbwarn "Detected warning_trigger, this might cause a problem later." fi if [ recoverable_error ]; then bberror "Hit recoverable_error, correcting" fi if [ fatal_error ]; then bbfatal "fatal_error detected" fi bbdebug 2 "Completed do_my_function" }

Here are some other tips that you might find useful: When adding new packages, it is worth watching for undesirable items making their way into compiler command lines. For example, you do not want references to local system files like /usr/lib/ or /usr/include/. If you want to remove the psplash boot splashscreen, add psplash=false to the kernel command line. Doing so prevents psplash from loading and thus allows you to see the console. It is also possible to switch out of the splashscreen by switching the virtual console (e.g. Fn+Left or Fn+Right on a Zaurus).

Many factors can influence the quality of a build. For example, if you upgrade a recipe to use a new version of an upstream software package or you experiment with some new configuration options, subtle changes can occur that you might not detect until later. Consider the case where your recipe is using a newer version of an upstream package. In this case, a new version of a piece of software might introduce an optional dependency on another library, which is auto-detected. If that library has already been built when the software is building, the software will link to the built library and that library will be pulled into your image along with the new software even if you did not want the library. The buildhistory class exists to help you maintain the quality of your build output. You can use the class to highlight unexpected and possibly unwanted changes in the build output. When you enable build history, it records information about the contents of each package and image and then commits that information to a local Git repository where you can examine the information. The remainder of this section describes the following: How you can enable and disable build history How to understand what the build history contains How to limit the information used for build history How to examine the build history from both a command-line and web interface

Build history is disabled by default. To enable it, add the following INHERIT statement and set the BUILDHISTORY_COMMIT variable to "1" at the end of your conf/local.conf file found in the Build Directory: INHERIT += "buildhistory" BUILDHISTORY_COMMIT = "1" Enabling build history as previously described causes the build process to collect build output information and commit it to a local Git repository. Enabling build history increases your build times slightly, particularly for images, and increases the amount of disk space used during the build. You can disable build history by removing the previous statements from your conf/local.conf file.

Build history information is kept in ${TOPDIR}/buildhistory in the Build Directory as defined by the BUILDHISTORY_DIR variable. The following is an example abbreviated listing: At the top level, there is a metadata-revs file that lists the revisions of the repositories for the layers enabled when the build was produced. The rest of the data splits into separate packages, images and sdk directories, the contents of which are described below.

The history for each package contains a text file that has name-value pairs with information about the package. For example, buildhistory/packages/i586-poky-linux/busybox/busybox/latest contains the following: PV = 1.22.1 PR = r32 RPROVIDES = RDEPENDS = glibc (>= 2.20) update-alternatives-opkg RRECOMMENDS = busybox-syslog busybox-udhcpc update-rc.d PKGSIZE = 540168 FILES = /usr/bin/* /usr/sbin/* /usr/lib/busybox/* /usr/lib/lib*.so.* \ /etc /com /var /bin/* /sbin/* /lib/*.so.* /lib/udev/rules.d \ /usr/lib/udev/rules.d /usr/share/busybox /usr/lib/busybox/* \ /usr/share/pixmaps /usr/share/applications /usr/share/idl \ /usr/share/omf /usr/share/sounds /usr/lib/bonobo/servers FILELIST = /bin/busybox /bin/busybox.nosuid /bin/busybox.suid /bin/sh \ /etc/busybox.links.nosuid /etc/busybox.links.suid Most of these name-value pairs correspond to variables used to produce the package. The exceptions are FILELIST, which is the actual list of files in the package, and PKGSIZE, which is the total size of files in the package in bytes. There is also a file corresponding to the recipe from which the package came (e.g. buildhistory/packages/i586-poky-linux/busybox/latest): PV = 1.22.1 PR = r32 DEPENDS = initscripts kern-tools-native update-rc.d-native \ virtual/i586-poky-linux-compilerlibs virtual/i586-poky-linux-gcc \ virtual/libc virtual/update-alternatives PACKAGES = busybox-ptest busybox-httpd busybox-udhcpd busybox-udhcpc \ busybox-syslog busybox-mdev busybox-hwclock busybox-dbg \ busybox-staticdev busybox-dev busybox-doc busybox-locale busybox Finally, for those recipes fetched from a version control system (e.g., Git), a file exists that lists source revisions that are specified in the recipe and lists the actual revisions used during the build. Listed and actual revisions might differ when SRCREV is set to ${AUTOREV}. Here is an example assuming buildhistory/packages/qemux86-poky-linux/linux-yocto/latest_srcrev): # SRCREV_machine = "38cd560d5022ed2dbd1ab0dca9642e47c98a0aa1" SRCREV_machine = "38cd560d5022ed2dbd1ab0dca9642e47c98a0aa1" # SRCREV_meta = "a227f20eff056e511d504b2e490f3774ab260d6f" SRCREV_meta = "a227f20eff056e511d504b2e490f3774ab260d6f" You can use the buildhistory-collect-srcrevs command with the -a option to collect the stored SRCREV values from build history and report them in a format suitable for use in global configuration (e.g., local.conf or a distro include file) to override floating AUTOREV values to a fixed set of revisions. Here is some example output from this command: $ buildhistory-collect-srcrevs -a # i586-poky-linux SRCREV_pn-glibc = "b8079dd0d360648e4e8de48656c5c38972621072" SRCREV_pn-glibc-initial = "b8079dd0d360648e4e8de48656c5c38972621072" SRCREV_pn-opkg-utils = "53274f087565fd45d8452c5367997ba6a682a37a" SRCREV_pn-kmod = "fd56638aed3fe147015bfa10ed4a5f7491303cb4" # x86_64-linux SRCREV_pn-gtk-doc-stub-native = "1dea266593edb766d6d898c79451ef193eb17cfa" SRCREV_pn-dtc-native = "65cc4d2748a2c2e6f27f1cf39e07a5dbabd80ebf" SRCREV_pn-update-rc.d-native = "eca680ddf28d024954895f59a241a622dd575c11" SRCREV_glibc_pn-cross-localedef-native = "b8079dd0d360648e4e8de48656c5c38972621072" SRCREV_localedef_pn-cross-localedef-native = "c833367348d39dad7ba018990bfdaffaec8e9ed3" SRCREV_pn-prelink-native = "faa069deec99bf61418d0bab831c83d7c1b797ca" SRCREV_pn-opkg-utils-native = "53274f087565fd45d8452c5367997ba6a682a37a" SRCREV_pn-kern-tools-native = "23345b8846fe4bd167efdf1bd8a1224b2ba9a5ff" SRCREV_pn-kmod-native = "fd56638aed3fe147015bfa10ed4a5f7491303cb4" # qemux86-poky-linux SRCREV_machine_pn-linux-yocto = "38cd560d5022ed2dbd1ab0dca9642e47c98a0aa1" SRCREV_meta_pn-linux-yocto = "a227f20eff056e511d504b2e490f3774ab260d6f" # all-poky-linux SRCREV_pn-update-rc.d = "eca680ddf28d024954895f59a241a622dd575c11" Here are some notes on using the buildhistory-collect-srcrevs command: By default, only values where the SRCREV was not hardcoded (usually when AUTOREV was used) are reported. Use the -a option to see all SRCREV values. The output statements might not have any effect if overrides are applied elsewhere in the build system configuration. Use the -f option to add the forcevariable override to each output line if you need to work around this restriction. The script does apply special handling when building for multiple machines. However, the script does place a comment before each set of values that specifies which triplet to which they belong as shown above (e.g., i586-poky-linux).

The files produced for each image are as follows: image-files: A directory containing selected files from the root filesystem. The files are defined by BUILDHISTORY_IMAGE_FILES. build-id.txt: Human-readable information about the build configuration and metadata source revisions. This file contains the full build header as printed by BitBake. *.dot: Dependency graphs for the image that are compatible with graphviz. files-in-image.txt: A list of files in the image with permissions, owner, group, size, and symlink information. image-info.txt: A text file containing name-value pairs with information about the image. See the following listing example for more information. installed-package-names.txt: A list of installed packages by name only. installed-package-sizes.txt: A list of installed packages ordered by size. installed-packages.txt: A list of installed packages with full package filenames. Installed package information is able to be gathered and produced even if package management is disabled for the final image. Here is an example of image-info.txt: DISTRO = poky DISTRO_VERSION = 1.7 USER_CLASSES = buildstats image-mklibs image-prelink IMAGE_CLASSES = image_types IMAGE_FEATURES = debug-tweaks IMAGE_LINGUAS = IMAGE_INSTALL = packagegroup-core-boot run-postinsts BAD_RECOMMENDATIONS = NO_RECOMMENDATIONS = PACKAGE_EXCLUDE = ROOTFS_POSTPROCESS_COMMAND = write_package_manifest; license_create_manifest; \ write_image_manifest ; buildhistory_list_installed_image ; \ buildhistory_get_image_installed ; ssh_allow_empty_password; \ postinst_enable_logging; rootfs_update_timestamp ; ssh_disable_dns_lookup ; IMAGE_POSTPROCESS_COMMAND = buildhistory_get_imageinfo ; IMAGESIZE = 6900 Other than IMAGESIZE, which is the total size of the files in the image in Kbytes, the name-value pairs are variables that may have influenced the content of the image. This information is often useful when you are trying to determine why a change in the package or file listings has occurred.

As you can see, build history produces image information, including dependency graphs, so you can see why something was pulled into the image. If you are just interested in this information and not interested in collecting specific package or SDK information, you can enable writing only image information without any history by adding the following to your conf/local.conf file found in the Build Directory: INHERIT += "buildhistory" BUILDHISTORY_COMMIT = "0" BUILDHISTORY_FEATURES = "image" Here, you set the BUILDHISTORY_FEATURES variable to use the image feature only.

Build history collects similar information on the contents of SDKs (e.g. bitbake -c populate_sdk imagename) as compared to information it collects for images. Furthermore, this information differs depending on whether an extensible or standard SDK is being produced. The following list shows the files produced for SDKs: files-in-sdk.txt: A list of files in the SDK with permissions, owner, group, size, and symlink information. This list includes both the host and target parts of the SDK. sdk-info.txt: A text file containing name-value pairs with information about the SDK. See the following listing example for more information. sstate-task-sizes.txt: A text file containing name-value pairs with information about task group sizes (e.g. do_populate_sysroot tasks have a total size). The sstate-task-sizes.txt file exists only when an extensible SDK is created. sstate-package-sizes.txt: A text file containing name-value pairs with information for the shared-state packages and sizes in the SDK. The sstate-package-sizes.txt file exists only when an extensible SDK is created. sdk-files: A folder that contains copies of the files mentioned in BUILDHISTORY_SDK_FILES if the files are present in the output. Additionally, the default value of BUILDHISTORY_SDK_FILES is specific to the extensible SDK although you can set it differently if you would like to pull in specific files from the standard SDK. The default files are conf/local.conf, conf/bblayers.conf, conf/auto.conf, conf/locked-sigs.inc, and conf/devtool.conf. Thus, for an extensible SDK, these files get copied into the sdk-files directory. The following information appears under each of the host and target directories for the portions of the SDK that run on the host and on the target, respectively: The following files for the most part are empty when producing an extensible SDK because this type of SDK is not constructed from packages as is the standard SDK. depends.dot: Dependency graph for the SDK that is compatible with graphviz. installed-package-names.txt: A list of installed packages by name only. installed-package-sizes.txt: A list of installed packages ordered by size. installed-packages.txt: A list of installed packages with full package filenames. Here is an example of sdk-info.txt: DISTRO = poky DISTRO_VERSION = 1.3+snapshot-20130327 SDK_NAME = poky-glibc-i686-arm SDK_VERSION = 1.3+snapshot SDKMACHINE = SDKIMAGE_FEATURES = dev-pkgs dbg-pkgs BAD_RECOMMENDATIONS = SDKSIZE = 352712 Other than SDKSIZE, which is the total size of the files in the SDK in Kbytes, the name-value pairs are variables that might have influenced the content of the SDK. This information is often useful when you are trying to determine why a change in the package or file listings has occurred.

You can examine build history output from the command line or from a web interface. To see any changes that have occurred (assuming you have BUILDHISTORY_COMMIT = "1"), you can simply use any Git command that allows you to view the history of a repository. Here is one method: $ git log -p You need to realize, however, that this method does show changes that are not significant (e.g. a package's size changing by a few bytes). A command-line tool called buildhistory-diff does exist, though, that queries the Git repository and prints just the differences that might be significant in human-readable form. Here is an example: $ ~/poky/poky/scripts/buildhistory-diff . HEAD^ Changes to images/qemux86_64/glibc/core-image-minimal (files-in-image.txt): /etc/anotherpkg.conf was added /sbin/anotherpkg was added * (installed-package-names.txt): * anotherpkg was added Changes to images/qemux86_64/glibc/core-image-minimal (installed-package-names.txt): anotherpkg was added packages/qemux86_64-poky-linux/v86d: PACKAGES: added "v86d-extras" * PR changed from "r0" to "r1" * PV changed from "0.1.10" to "0.1.12" packages/qemux86_64-poky-linux/v86d/v86d: PKGSIZE changed from 110579 to 144381 (+30%) * PR changed from "r0" to "r1" * PV changed from "0.1.10" to "0.1.12" The buildhistory-diff tool requires the GitPython package. Be sure to install it using Pip3 as follows: $ pip3 install GitPython --user Alternatively, you can install python3-git using the appropriate distribution package manager (e.g. apt-get, dnf, or zipper). To see changes to the build history using a web interface, follow the instruction in the README file here. . Here is a sample screenshot of the interface:

Build time can be an issue. By default, the build system uses simple controls to try and maximize build efficiency. In general, the default settings for all the following variables result in the most efficient build times when dealing with single socket systems (i.e. a single CPU). If you have multiple CPUs, you might try increasing the default values to gain more speed. See the descriptions in the glossary for each variable for more information: BB_NUMBER_THREADS: The maximum number of threads BitBake simultaneously executes. BB_NUMBER_PARSE_THREADS: The number of threads BitBake uses during parsing. PARALLEL_MAKE: Extra options passed to the make command during the do_compile task in order to specify parallel compilation on the local build host. PARALLEL_MAKEINST: Extra options passed to the make command during the do_install task in order to specify parallel installation on the local build host. As mentioned, these variables all scale to the number of processor cores available on the build system. For single socket systems, this auto-scaling ensures that the build system fundamentally takes advantage of potential parallel operations during the build based on the build machine's capabilities. Following are additional factors that can affect build speed: File system type: The file system type that the build is being performed on can also influence performance. Using ext4 is recommended as compared to ext2 and ext3 due to ext4 improved features such as extents. Disabling the updating of access time using noatime: The noatime mount option prevents the build system from updating file and directory access times. Setting a longer commit: Using the "commit=" mount option increases the interval in seconds between disk cache writes. Changing this interval from the five second default to something longer increases the risk of data loss but decreases the need to write to the disk, thus increasing the build performance. Choosing the packaging backend: Of the available packaging backends, IPK is the fastest. Additionally, selecting a singular packaging backend also helps. Using tmpfs for TMPDIR as a temporary file system: While this can help speed up the build, the benefits are limited due to the compiler using -pipe. The build system goes to some lengths to avoid sync() calls into the file system on the principle that if there was a significant failure, the Build Directory contents could easily be rebuilt. Inheriting the rm_work class: Inheriting this class has shown to speed up builds due to significantly lower amounts of data stored in the data cache as well as on disk. Inheriting this class also makes cleanup of TMPDIR faster, at the expense of being easily able to dive into the source code. File system maintainers have recommended that the fastest way to clean up large numbers of files is to reformat partitions rather than delete files due to the linear nature of partitions. This, of course, assumes you structure the disk partitions and file systems in a way that this is practical. Aside from the previous list, you should keep some trade offs in mind that can help you speed up the build: Remove items from DISTRO_FEATURES that you might not need. Exclude debug symbols and other debug information: If you do not need these symbols and other debug information, disabling the *-dbg package generation can speed up the build. You can disable this generation by setting the INHIBIT_PACKAGE_DEBUG_SPLIT variable to "1". Disable static library generation for recipes derived from autoconf or libtool: Following is an example showing how to disable static libraries and still provide an override to handle exceptions: STATICLIBCONF = "--disable-static" STATICLIBCONF_sqlite3-native = "" EXTRA_OECONF += "${STATICLIBCONF}" Some recipes need static libraries in order to work correctly (e.g. pseudo-native needs sqlite3-native). Overrides, as in the previous example, account for these kinds of exceptions. Some packages have packaging code that assumes the presence of the static libraries. If so, you might need to exclude them as well.