%poky; ] > This chapter describes the extensible SDK and how to use it. The extensible SDK makes it easy to add new applications and libraries to an image, modify the source for an existing component, test changes on the target hardware, and ease integration into the rest of the OpenEmbedded build system. Information in this chapter covers features that are not part of the standard SDK. In other words, the chapter presents information unique to the extensible SDK only. For information on how to use the standard SDK, see the "Using the Standard SDK" chapter.

Getting set up to use the extensible SDK is identical to getting set up to use the standard SDK. You still need to locate and run the installer and then run the environment setup script. See the "Installing the SDK" and the "Running the SDK Environment Setup Script" sections for general information. The following items highlight the only differences between getting set up to use the extensible SDK as compared to the standard SDK: Default Installation Directory: By default, the extensible SDK installs into the poky_sdk folder of your home directory. As with the standard SDK, you can choose to install the extensible SDK in any location when you run the installer. However, unlike the standard SDK, the location you choose needs to be writable for whichever users need to use the SDK, since files will need to be written under that directory during the normal course of operation. Build Tools and Build System: The extensible SDK installer performs additional tasks as compared to the standard SDK installer. The extensible SDK installer extracts build tools specific to the SDK and the installer also prepares the internal build system within the SDK. Here is example output for running the extensible SDK installer: $ ./poky-glibc-x86_64-core-image-minimal-core2-64-toolchain-ext-2.1+snapshot.sh Poky (Yocto Project Reference Distro) Extensible SDK installer version 2.1+snapshot =================================================================================== Enter target directory for SDK (default: ~/poky_sdk): You are about to install the SDK to "/home/scottrif/poky_sdk". Proceed[Y/n]? Y Extracting SDK......................................................................done Setting it up... Extracting buildtools... Preparing build system... done SDK has been successfully set up and is ready to be used. Each time you wish to use the SDK in a new shell session, you need to source the environment setup script e.g. $ . /home/scottrif/poky_sdk/environment-setup-core2-64-poky-linux After installing the SDK, you need to run the SDK environment setup script. Here is the output: $ source environment-setup-core2-64-poky-linux SDK environment now set up; additionally you may now run devtool to perform development tasks. Run devtool --help for further details. Once you run the environment setup script, you have devtool available.

The cornerstone of the extensible SDK is a command-line tool called devtool. This tool provides a number of features that help you build, test and package software within the extensible SDK, and optionally integrate it into an image built by the OpenEmbedded build system. The devtool command line is organized similarly to Git in that it has a number of sub-commands for each function. You can run devtool --help to see all the commands. Two devtool subcommands that provide entry-points into development are: devtool add: Assists in adding new software to be built. devtool modify: Sets up an environment to enable you to modify the source of an existing component. As with the OpenEmbedded build system, "recipes" represent software packages within devtool. When you use devtool add, a recipe is automatically created. When you use devtool modify, the specified existing recipe is used in order to determine where to get the source code and how to patch it. In both cases, an environment is set up so that when you build the recipe a source tree that is under your control is used in order to allow you to make changes to the source as desired. By default, both new recipes and the source go into a "workspace" directory under the SDK. The remainder of this section presents the devtool add and devtool modify workflows.

The devtool add command generates a new recipe based on existing source code. This command takes advantage of the workspace layer that many devtool commands use. The command is flexible enough to allow you to extract source code into both the workspace or a separate local Git repository and to use existing code that does not need to be extracted. Depending on your particular scenario, the arguments and options you use with devtool add form different combinations. The following diagram shows common development flows you would use with the devtool add command: Generating the New Recipe: The top part of the flow shows three scenarios by which you could use devtool add to generate a recipe based on existing source code. In a shared development environment, it is typical where other developers are responsible for various areas of source code. As a developer, you are probably interested in using that source code as part of your development using the Yocto Project. All you need is access to the code, a recipe, and a controlled area in which to do your work. Within the diagram, three possible scenarios feed into the devtool add workflow: Left: The left scenario represents a common situation where the source code does not exist locally and needs to be extracted. In this situation, you just let it get extracted to the default workspace - you do not want it in some specific location outside of the workspace. Thus, everything you need will be located in the workspace: $ devtool add recipe fetchuri With this command, devtool creates a recipe and an append file in the workspace as well as extracts the upstream source files into a local Git repository also within the sources folder. Middle: The middle scenario also represents a situation where the source code does not exist locally. In this case, the code is again upstream and needs to be extracted to some local area - this time outside of the default workspace. As always, if required devtool creates a Git repository locally during the extraction. Furthermore, the first positional argument srctree in this case identifies where the devtool add command will locate the extracted code outside of the workspace: $ devtool add recipe srctree fetchuri In summary, the source code is pulled from fetchuri and extracted into the location defined by srctree as a local Git repository. Within workspace, devtool creates both the recipe and an append file for the recipe. Right: The right scenario represents a situation where the source tree (srctree) has been previously prepared outside of the devtool workspace. The following command names the recipe and identifies where the existing source tree is located: $ devtool add recipe srctree The command examines the source code and creates a recipe for it placing the recipe into the workspace. Because the extracted source code already exists, devtool does not try to relocate it into the workspace - just the new the recipe is placed in the workspace. Aside from a recipe folder, the command also creates an append folder and places an initial *.bbappend within. Edit the Recipe: At this point, you can use devtool edit-recipe to open up the editor as defined by the $EDITOR environment variable and modify the file: $ devtool edit-recipe recipe From within the editor, you can make modifications to the recipe that take affect when you build it later. Build the Recipe or Rebuild the Image: At this point in the flow, the next step you take depends on what you are going to do with the new code. If you need to take the build output and eventually move it to the target hardware, you would use devtool build: $ devtool build recipe On the other hand, if you want an image to contain the recipe's packages for immediate deployment onto a device (e.g. for testing purposes), you can use the devtool build-image command: $ devtool build-image image Deploy the Build Output: When you use the devtool build command to build out your recipe, you probably want to see if the resulting build output works as expected on target hardware. This step assumes you have a previously built image that is already either running in QEMU or running on actual hardware. Also, it is assumed that for deployment of the image to the target, SSH is installed in the image and if the image is running on real hardware that you have network access to and from your development machine. You can deploy your build output to that target hardware by using the devtool deploy-target command: $ devtool deploy-target recipe target The target is a live target machine running as an SSH server. You can, of course, also deploy the image you build using the devtool build-image command to actual hardware. However, devtool does not provide a specific command that allows you to do this. Optionally Update the Recipe With Patch Files: Once you are satisfied with the recipe, if you have made any changes to the source tree that you want to have applied by the recipe, you need to generate patches from those changes. You do this before moving the recipe to its final layer and cleaning up the workspace area devtool uses. This optional step is especially relevant if you are using or adding third-party software. To convert commits created using Git to patch files, use the devtool update-recipe command. Any changes you want to turn into patches must be committed to the Git repository in the source tree. $ devtool update-recipe recipe Move the Recipe to its Permanent Layer: Before cleaning up the workspace, you need to move the final recipe to its permanent layer. You must do this before using the devtool reset command if you want to retain the recipe. Reset the Recipe: As a final step, you can restore the state such that standard layers and the upstream source is used to build the recipe rather than data in the workspace. To reset the recipe, use the devtool reset command: $ devtool reset recipe

The devtool modify command prepares the way to work on existing code that already has a recipe in place. The command is flexible enough to allow you to extract code, specify the existing recipe, and keep track of and gather any patch files from other developers that are associated with the code. Depending on your particular scenario, the arguments and options you use with devtool modify form different combinations. The following diagram shows common development flows you would use with the devtool modify command: Preparing to Modify the Code: The top part of the flow shows three scenarios by which you could use devtool modify to prepare to work on source files. Each scenario assumes the following: The recipe exists in some layer external to the devtool workspace. The source files exist upstream in an un-extracted state or locally in a previously extracted state. The typical situation is where another developer has created some layer for use with the Yocto Project and their recipe already resides in that layer. Furthermore, their source code is readily available either upstream or locally. Left: The left scenario represents a common situation where the source code does not exist locally and needs to be extracted. In this situation, the source is extracted into the default workspace location. The recipe, in this scenario, is in its own layer outside the workspace (i.e. meta-layername). The following command identifies the recipe and by default extracts the source files: $ devtool modify recipe Once devtoollocates the recipe, it uses the SRC_URI variable to locate the source code and any local patch files from other developers are located. You cannot provide an URL for srctree when using the devtool modify command. With this scenario, however, since no srctree argument exists, the devtool modify command by default extracts the source files to a Git structure. Furthermore, the location for the extracted source is the default area within the workspace. The result is that the command sets up both the source code and an append file within the workspace with the recipe remaining in its original location. Middle: The middle scenario represents a situation where the source code also does not exist locally. In this case, the code is again upstream and needs to be extracted to some local area as a Git repository. The recipe, in this scenario, is again in its own layer outside the workspace. The following command tells devtool what recipe with which to work and, in this case, identifies a local area for the extracted source files that is outside of the default workspace: $ devtool modify recipe srctree As with all extractions, the command uses the recipe's SRC_URI to locate the source files. Once the files are located, the command by default extracts them. Providing the srctree argument instructs devtool where place the extracted source. Within workspace, devtool creates an append file for the recipe. The recipe remains in its original location but the source files are extracted to the location you provided with srctree. Right: The right scenario represents a situation where the source tree (srctree) exists as a previously extracted Git structure outside of the devtool workspace. In this example, the recipe also exists elsewhere in its own layer. The following command tells devtool the recipe with which to work, uses the "-n" option to indicate source does not need to be extracted, and uses srctree to point to the previously extracted source files: $ devtool modify -n recipe srctree Once the command finishes, it creates only an append file for the recipe in the workspace. The recipe and the source code remain in their original locations. Edit the Source: Once you have used the devtool modify command, you are free to make changes to the source files. You can use any editor you like to make and save your source code modifications. Build the Recipe: Once you have updated the source files, you can build the recipe. Deploy the Build Output: When you use the devtool build command to build out your recipe, you probably want to see if the resulting build output works as expected on target hardware. This step assumes you have a previously built image that is already either running in QEMU or running on actual hardware. Also, it is assumed that for deployment of the image to the target, SSH is installed in the image and if the image is running on real hardware that you have network access to and from your development machine. You can deploy your build output to that target hardware by using the devtool deploy-target command: $ devtool deploy-target recipe target The target is a live target machine running as an SSH server. You can, of course, also deploy the image you build using the devtool build-image command to actual hardware. However, devtool does not provide a specific command that allows you to do this. Optionally Create Patch Files for Your Changes: After you have debugged your changes, you can use devtool update-recipe to generate patch files for all the commits you have made. Patch files are generated only for changes you have committed. $ devtool update-recipe recipe By default, the devtool update-recipe command creates the patch files in a folder named the same as the recipe beneath the folder in which the recipe resides, and updates the recipe's SRC_URI statement to point to the generated patch files. You can use the "--append LAYERDIR" option to cause the command to create append files in a specific layer rather than the default recipe layer. Restore the Workspace: The devtool reset restores the state so that standard layers and upstream sources are used to build the recipe rather than what is in the workspace. $ devtool reset recipe

The devtool add command automatically creates a recipe based on the source tree with which you provide it. Currently, the command has support for the following: Autotools (autoconf and automake) CMake Scons qmake Plain Makefile Out-of-tree kernel module Binary package (i.e. "-b" option) Node.js module through npm Python modules that use setuptools or distutils Apart from binary packages, the determination of how a source tree should be treated is automatic based on the files present within that source tree. For example, if a CMakeLists.txt file is found, then the source tree is assumed to be using CMake and is treated accordingly. In most cases, you need to edit the automatically generated recipe in order to make it build properly. Typically, you would go through several edit and build cycles until you can build the recipe. Once the recipe can be built, you could use possible further iterations to test the recipe on the target device. The remainder of this section covers specifics regarding how parts of the recipe are generated.

If you do not specify a name and version on the command line, devtool add attempts to determine the name and version of the software being built from various metadata within the source tree. Furthermore, the command sets the name of the created recipe file accordingly. If the name or version cannot be determined, the devtool add command prints an error and you must re-run the command with both the name and version or just the name or version specified. Sometimes the name or version determined from the source tree might be incorrect. For such a case, you must reset the recipe: $ devtool reset -n recipename After running the devtool reset command, you need to run devtool add again and provide the name or the version.

The devtool add command attempts to detect build-time dependencies and map them to other recipes in the system. During this mapping, the command fills in the names of those recipes in the DEPENDS value within the recipe. If a dependency cannot be mapped, then a comment is placed in the recipe indicating such. The inability to map a dependency might be caused because the naming is not recognized or because the dependency simply is not available. For cases where the dependency is not available, you must use the devtool add command to add an additional recipe to satisfy the dependency and then come back to the first recipe and add its name to DEPENDS. If you need to add runtime dependencies, you can do so by adding the following to your recipe: RDEPENDS_${PN} += "dependency1 dependency2 ..." The devtool add command often cannot distinguish between mandatory and optional dependencies. Consequently, some of the detected dependencies might in fact be optional. When in doubt, consult the documentation or the configure script for the software the recipe is building for further details. In some cases, you might find you can substitute the dependency for an option to disable the associated functionality passed to the configure script.

The devtool add command attempts to determine if the software you are adding is able to be distributed under a common open-source license and sets the LICENSE value accordingly. You should double-check this value against the documentation or source files for the software you are building and update that LICENSE value if necessary. The devtool add command also sets the LIC_FILES_CHKSUM value to point to all files that appear to be license-related. However, license statements often appear in comments at the top of source files or within documentation. Consequently, you might need to amend the LIC_FILES_CHKSUM variable to point to one or more of those comments if present. Setting LIC_FILES_CHKSUM is particularly important for third-party software. The mechanism attempts to ensure correct licensing should you upgrade the recipe to a newer upstream version in future. Any change in licensing is detected and you receive an error prompting you to check the license text again. If the devtool add command cannot determine licensing information, the LICENSE value is set to "CLOSED" and the LIC_FILES_CHKSUM vaule remains unset. This behavior allows you to continue with development but is unlikely to be correct in all cases. Consequently, you should check the documentation or source files for the software you are building to determine the actual license.

The use of make by itself is very common in both proprietary and open source software. Unfortunately, Makefiles are often not written with cross-compilation in mind. Thus, devtool add often cannot do very much to ensure that these Makefiles build correctly. It is very common, for example, to explicitly call gcc instead of using the CC variable. Usually, in a cross-compilation environment, gcc is the compiler for the build host and the cross-compiler is named something similar to arm-poky-linux-gnueabi-gcc and might require some arguments (e.g. to point to the associated sysroot for the target machine). When writing a recipe for Makefile-only software, keep the following in mind: You probably need to patch the Makefile to use variables instead of hardcoding tools within the toolchain such as gcc and g++. The environment in which make runs is set up with various standard variables for compilation (e.g. CC, CXX, and so forth) in a similar manner to the environment set up by the SDK's environment setup script. One easy way to see these variables is to run the devtool build command on the recipe and then look in oe-logs/run.do_compile. Towards the top of this file you will see a list of environment variables that are being set. You can take advantage of these variables within the Makefile. If the Makefile sets a default for a variable using "=", that default overrides the value set in the environment, which is usually not desirable. In this situation, you can either patch the Makefile so it sets the default using the "?=" operator, or you can alternatively force the value on the make command line. To force the value on the command line, add the variable setting to EXTRA_OEMAKE within the recipe as follows: EXTRA_OEMAKE += "'CC=${CC}' 'CXX=${CXX}'" In the above example, single quotes are used around the variable settings as the values are likely to contain spaces because required default options are passed to the compiler. Hardcoding paths inside Makefiles is often problematic in a cross-compilation environment. This is particularly true because those hardcoded paths often point to locations on the build host and thus will either be read-only or will introduce contamination into the cross-compilation by virtue of being specific to the build host rather than the target. Patching the Makefile to use prefix variables or other path variables is usually the way to handle this. Sometimes a Makefile runs target-specific commands such as ldconfig. For such cases, you might be able to simply apply patches that remove these commands from the Makefile.

Often, you need to build additional tools that run on the build host system as opposed to the target. You should indicate this using one of the following methods when you run devtool add: Specify the name of the recipe such that it ends with "-native". Specifying the name like this produces a recipe that only builds for the build host. Specify the "‐‐also-native" option with the devtool add command. Specifying this option creates a recipe file that still builds for the target but also creates a variant with a "-native" suffix that builds for the build host. If you need to add a tool that is shipped as part of a source tree that builds code for the target, you can typically accomplish this by building the native and target parts separately rather than within the same compilation process. Realize though that with the "‐‐also-native" option, you can add the tool using just one recipe file.

You can use the devtool add command in the following form to add Node.js modules: $ devtool add "npm://registry.npmjs.org;name=forever;version=0.15.1" The name and version parameters are mandatory. Lockdown and shrinkwrap files are generated and pointed to by the recipe in order to freeze the version that is fetched for the dependencies according to the first time. This also saves checksums that are verified on future fetches. Together, these behaviors ensure the reproducibility and integrity of the build. You must use quotes around the URL. The devtool add does not require the quotes, but the shell considers ";" as a splitter between multiple commands. Thus, without the quotes, devtool add does not receive the other parts, which results in several "command not found" errors. In order to support adding Node.js modules, a nodejs recipe must be part of your SDK in order to provide Node.js itself.

When building a recipe with devtool build the typical build progression is as follows: Fetch the source Unpack the source Configure the source Compiling the source Install the build output Package the installed output For recipes in the workspace, fetching and unpacking is disabled as the source tree has already been prepared and is persistent. Each of these build steps is defined as a function, usually with a "do_" prefix. These functions are typically shell scripts but can instead be written in Python. If you look at the contents of a recipe, you will see that the recipe does not include complete instructions for building the software. Instead, common functionality is encapsulated in classes inherited with the inherit directive, leaving the recipe to describe just the things that are specific to the software to be built. A base class exists that is implicitly inherited by all recipes and provides the functionality that most typical recipes need. The remainder of this section presents information useful when working with recipes.

When you are debugging a recipe that you previously created using devtool add or whose source you are modifying by using the devtool modify command, after the first run of devtool build, you will find some symbolic links created within the source tree: oe-logs, which points to the directory in which log files and run scripts for each build step are created and oe-workdir, which points to the temporary work area for the recipe. You can use these links to get more information on what is happening at each build step. These locations under oe-workdir are particularly useful: image/: Contains all of the files installed at the do_install stage. Within a recipe, this directory is referred to by the expression ${D}. sysroot-destdir/: Contains a subset of files installed within do_install that have been put into the shared sysroot. For more information, see the "Sharing Files Between Recipes" section. packages-split/: Contains subdirectories for each package produced by the recipe. For more information, see the "Packaging" section.

If the software your recipe is building uses GNU autoconf, then a fixed set of arguments is passed to it to enable cross-compilation plus any extras specified by EXTRA_OECONF set within the recipe. If you wish to pass additional options, add them to EXTRA_OECONF. Other supported build tools have similar variables (e.g. EXTRA_OECMAKE for CMake, EXTRA_OESCONS for Scons, and so forth). If you need to pass anything on the make command line, you can use EXTRA_OEMAKE to do so. You can use the devtool configure-help command to help you set the arguments listed in the previous paragraph. The command determines the exact options being passed, and shows them to you along with any custom arguments specified through EXTRA_OECONF. If applicable, the command also shows you the output of the configure script's "‐‐help" option as a reference.

Recipes often need to use files provided by other recipes on the build host. For example, an application linking to a common library needs access to the library itself and its associated headers. The way this access is accomplished within the extensible SDK is through the sysroot. One sysroot exists per "machine" for which the SDK is being built. In practical terms, this means a sysroot exists for the target machine, and a sysroot exists for the build host. Recipes should never write files directly into the sysroot. Instead, files should be installed into standard locations during the do_install task within the ${D} directory. A subset of these files automatically go into the sysroot. The reason for this limitation is that almost all files that go into the sysroot are cataloged in manifests in order to ensure they can be removed later when a recipe is modified or removed. Thus, the sysroot is able to remain free from stale files.

Packaging is not always particularly relevant within the extensible SDK. However, if you examine how build output gets into the final image on the target device, it is important to understand packaging because the contents of the image are expressed in terms of packages and not recipes. During the do_package task, files installed during the do_install task are split into one main package, which is almost always named the same as the recipe, and several other packages. This separation is done because not all of those installed files are always useful in every image. For example, you probably do not need any of the documentation installed in a production image. Consequently, for each recipe the documentation files are separated into a -doc package. Recipes that package software that has optional modules or plugins might do additional package splitting as well. After building a recipe you can see where files have gone by looking in the oe-workdir/packages-split directory, which contains a subdirectory for each package. Apart from some advanced cases, the PACKAGES and FILES variables controls splitting. The PACKAGES variable lists all of the packages to be produced, while the FILES variable specifies which files to include in each package, using an override to specify the package. For example, FILES_${PN} specifies the files to go into the main package (i.e. the main package is named the same as the recipe and ${PN} evaluates to the recipe name). The order of the PACKAGES value is significant. For each installed file, the first package whose FILES value matches the file is the package into which the file goes. Defaults exist for both the PACKAGES and FILES variables. Consequently, you might find you do not even need to set these variables in your recipe unless the software the recipe is building installs files into non-standard locations.

If you use the devtool deploy-target command to write a recipe's build output to the target, and you are working on an existing component of the system, then you might find yourself in a situation where you need to restore the original files that existed prior to running the devtool deploy-target command. Because the devtool deploy-target command backs up any files it overwrites, you can use the devtool undeploy-target to restore those files and remove any other files the recipe deployed. Consider the following example: $ devtool undeploy-target lighttpd root@192.168.7.2 If you have deployed multiple applications, you can remove them all at once thus restoring the target device back to its original state: $ devtool undeploy-target -a root@192.168.7.2 Information about files deployed to the target as well as any backed up files are stored on the target itself. This storage of course requires some additional space on the target machine. The devtool deploy-target and devtool undeploy-target command do not currently interact with any package management system on the target device (e.g. RPM or OPKG). Consequently, you should not intermingle operations devtool deploy-target and the package manager operations on the target device. Doing so could result in a conflicting set of files.

The extensible SDK typically only comes with a small number of tools and libraries out of the box. If you have a minimal SDK, then it starts mostly empty and is populated on-demand. However, sometimes you will need to explicitly install extra items into the SDK. If you need these extra items, you can first search for the items using the devtool search command. For example, suppose you need to link to libGL but you are not sure which recipe provides it. You can use the following command to find out: $ devtool search libGL mesa A free implementation of the OpenGL API Once you know the recipe (i.e. mesa in this example), you can install it: $ devtool sdk-install mesa By default, the devtool sdk-install assumes the item is available in pre-built form from your SDK provider. If the item is not available and it is acceptable to build the item from source, you can add the "-s" option as follows: $ devtool sdk-install -s mesa It is important to remember that building the item from source takes significantly longer than installing the pre-built artifact. Also, if no recipe exists for the item you want to add to the SDK, you must instead add it using the devtool add command.

If you are working with an extensible SDK that gets occasionally updated (e.g. typically when that SDK has been provided to you by another party), then you will need to manually pull down those updates to your installed SDK. To update your installed SDK, run the following: $ devtool sdk-update The previous command assumes your SDK provider has set the default update URL for you. If that URL has not been set, you need to specify it yourself as follows: $ devtool sdk-update path_to_update_directory The URL needs to point specifically to a published SDK and not an SDK installer that you would download and install.

You might need to produce an SDK that contains your own custom libraries for sending to a third party (e.g., if you are a vendor with customers needing to build their own software for the target platform). If that is the case, then you can produce a derivative SDK based on the currently installed SDK fairly easily. Use these steps: If necessary, install an extensible SDK that you want to use as a base for your derivative SDK. Source the environment script for the SDK. Add the extra libraries or other components you want by using the devtool add command. Run the devtool build-sdk command. The above procedure takes the recipes added to the workspace and constructs a new SDK installer containing those recipes and the resulting binary artifacts. The recipes go into their own separate layer in the constructed derivative SDK, leaving the workspace clean and ready for users to add their own recipes.