path: root/doc/user-manual/user-manual-metadata.xml

<!DOCTYPE chapter PUBLIC "-//OASIS//DTD DocBook XML V4.2//EN"
    "http://www.oasis-open.org/docbook/xml/4.2/docbookx.dtd">
<chapter id="user-manual-metadata">
  <title>Metadata</title>
  <section>
    <title>Description</title>
    <para>The BitBake task executor together with various types of
	    configuration files form the OpenEmbedded Core.
	    This section provides an overview of the BitBake task executor
	    and the configuration files by describing what they are used for
	    and how they interact.
    </para>
    <para>BitBake handles the parsing and execution of the data files.
	    The data itself is of various types: 
	    <itemizedlist>
		    <listitem>
			    <para>
				    <emphasis>Recipes:</emphasis>
				    Provides details about particular pieces
				    of software.
		    </para>
	    </listitem>
	    
	    <listitem>
		    <para>
			<emphasis>Class Data:</emphasis>
			An abstraction of common build information (e.g. how
			to build a Linux kernel).
		</para>
	</listitem>
      <listitem>
        <para>
		<emphasis>Configuration Data:</emphasis>Defines
		machine-specific settings, policy decisions, etc.
		Configuration data acts as the glue to bind everything
		together.
	</para>
      </listitem>
    </itemizedlist>
    <para>What follows is a large number of examples of BitBake metadata.
	    Any syntax which is not supported in any of the aforementioned areas
	    will be documented as such.
    </para>
    </section>

    <section>
      <title>Basic variable setting</title>
      <para>
        <screen>
        <varname>VARIABLE</varname> = "value"</screen>
      </para>
      <para>In this example, 
      <varname>VARIABLE</varname> is 
      <literal>value</literal>.</para>
    </section>
    
    <section>
      <title>Variable expansion</title>
      <para>BitBake supports variables referencing one another's contents using
	      a syntax which is similar to shell scripting:
      </para>
      <para>
        <screen>
        <varname>A</varname> = "aval" 
        <varname>B</varname> = "pre${A}post"</screen>
      </para>
      <para>This results in 
      <varname>A</varname> containing 
      <literal>aval</literal> and 
      <varname>B</varname> containing 
      <literal>preavalpost</literal>.</para>
    </section>
    
    <section>
      <title>Setting a default value (?=)</title>
      <para>
        <screen>
        <varname>A</varname>?= "aval"</screen>
      </para>
      <para>If 
      <varname>A</varname> is set before the above is called, it
      will retain its previous value. If 
      <varname>A</varname> is unset prior to the above call, 
      <varname>A</varname> will be set to 
      <literal>aval</literal>. Note that this assignment is
      immediate, so if there are multiple ?= assignments to a
      single variable, the first of those will be used.</para>
    </section>
    
    <section>
      <title>Setting a weak default value (??=)</title>
      <para>
        <screen>
        <varname>A</varname>??= "somevalue" 
        <varname>A</varname>??= "someothervalue"</screen>
      </para>
      <para>If 
	      <varname>A</varname> 
	      is set before the above, it will retain that value.
	      If 
	      <varname>A</varname> 
	      is unset prior to the above, 
	      <varname>A</varname> 
	      will be set to <literal>someothervalue</literal>. 
	      This is a lazy/weak assignment in that the assignment does not
	      occur until the end of the parsing process, so that the last,
	      rather than the first, ??= assignment to a given variable will 
	      be used.
	      Any other setting of A using = or ?= will however override the
	      value set with ??=.
      </para>
    </section>
    
    <section>
      <title>Immediate variable expansion (:=)</title>
      <para>:= results in a variable's contents being expanded immediately,
	      rather than when the variable is actually used.
      </para>
      
      <para>
        <screen>
        <varname>T</varname> = "123" 
        <varname>A</varname>:= "${B} ${A} test ${T}" 
        <varname>T</varname> = "456" 
        <varname>B</varname> = "${T} bval" 
        <varname>C</varname> = "cval" 
        <varname>C</varname>:= "${C}append"</screen>
      </para>
      <para>In this example, 
	      <varname>A</varname> 
	      would contain
	      <literal>test 123</literal>,
	      <varname>B</varname> 
	      would contain
	      <literal>456 bval</literal>, and
	      <varname>C</varname> 
	      would be
	      <literal>cvalappend</literal>.
      </para>
    </section>
    
    <section>
      <title>Appending (+=) and prepending (=+)</title>
      <para>
        <screen>
        <varname>B</varname> = "bval" 
        <varname>B</varname>+= "additionaldata" 
        <varname>C</varname> = "cval" 
        <varname>C</varname>=+ "test"</screen>
      </para>
      <para>In this example, 
	      <varname>B</varname> 
	      is now
	      <literal>bval additionaldata</literal> 
	      and
	      <varname>C</varname> 
	      is 
	      <literal>test cval</literal>.
      </para>
    </section>
    <section>
	    <title>Appending (.=) and prepending (=.) without spaces</title>
	    <para>
		    <screen>
			    <varname>B</varname> = "bval" 
			    <varname>B</varname>.= "additionaldata" 
			    <varname>C</varname> = "cval" 
			    <varname>C</varname>=. "test"
		    </screen>
	</para>
	<para>In this example,
		<varname>B</varname> 
		is now
		<literal>bvaladditionaldata</literal> 
		and 
		<varname>C</varname> 
		is 
		<literal>testcval</literal>.
		In contrast to the above appending and prepending operators,
		no additional space will be introduced.
	</para>
    </section>

    <section>
                <title>Appending and Prepending (override style syntax)</title>
                    <para><screen><varname>B</varname> = "bval"
<varname>B_append</varname> = " additional data"
<varname>C</varname> = "cval"
<varname>C_prepend</varname> = "additional data "</screen></para>
                 <para>This example results in <varname>B</varname> becoming <literal>bval additional data</literal>
and <varname>C</varname> becoming <literal>additional data cval</literal>. Note the spaces in the append.
Unlike the += operator, additional space is not automatically added.  You must take steps to add space
yourself.</para>
            </section>
            <section>
                <title>Removing (override style syntax)</title>
                <para><screen><varname>FOO</varname> = "123 456 789 123456 123 456 123 456"
<varname>FOO_remove</varname> = "123"
<varname>FOO_remove</varname> = "456"</screen></para>
                <para>In this example, <varname>FOO</varname> is now <literal>789 123456</literal>.</para>
    </section>

    <section>
	    <title>Conditional metadata set</title>
	    <para>OVERRIDES is a 
		    <quote>:</quote> 
		    separated variable containing each item you want to satisfy
		    conditions.
		    So, if you have a variable that is conditional on
		    <quote>arm</quote>, and
		    <quote>arm</quote> 
		    is in OVERRIDES, then the 
		    <quote>arm</quote> 
		    specific version of the variable is used rather than the
		    non-conditional version.
		    Example:
	    </para>
	    <para>
		    <screen>
			    <varname>OVERRIDES</varname> 
			    = "architecture:os:machine" 
			    <varname>TEST</varname> 
			    = "defaultvalue" 
			    <varname>TEST_os</varname> 
			    = "osspecificvalue" 
			    <varname>TEST_condnotinoverrides</varname> 
			    = "othercondvalue"
		    </screen>
      </para>
      <para>In this example, 
	      <varname>TEST</varname> 
	      would be <literal>osspecificvalue</literal>, 
	      due to the condition
	      <quote>os</quote> being in 
	      <varname>OVERRIDES</varname>.
      </para>
    </section>
    
    <section>
	    <title>Conditional appending</title>
	    <para>BitBake also supports appending and prepending to variables 
		    based on whether something is in OVERRIDES.
		    Example:
	    </para>
	    <para>
		    <screen>
			    <varname>DEPENDS</varname> 
			    = "glibc ncurses" 
			    <varname>OVERRIDES</varname> 
			    = "machine:local" 
			    <varname>DEPENDS_append_machine</varname> 
			    = "libmad"
		    </screen>
	</para>
	<para>In this example, 
		<varname>DEPENDS</varname> '
		is set to
		<literal>glibc ncurses libmad</literal>.
	</para>
	</section>
	
	<section>
	<title>Inclusion</title>
	<para>Next, there is the 
		<literal>include</literal> directive, that causes BitBake to
		parse whatever file you specify, and insert it at that
		location, which is not unlike
		<command>Make</command>. However, if the path specified on the
		<literal>include</literal> 
		line is a relative path, BitBake will locate the first one it
		can find within 
		<envar>BBPATH</envar>.
	</para>
	</section>
	
	<section>
	<title>Requiring inclusion</title>
	<para>In contrast to the 
		<literal>include</literal> 
		directive, 
		<literal>require</literal> 
		will raise a ParseError if the file to be included cannot be
		found.
		Otherwise it will behave just like the 
		<literal>include</literal> 
		directive.
	</para>
	</section>
	
	<section>
	<title>Inline Python variable expansion</title>
	<para>
		<screen>
			<varname>DATE</varname> =
			 "${@time.strftime('%Y%m%d',time.gmtime())}"
		 </screen>
	</para>
	<para>This would result in the
		<varname>DATE</varname> 
		variable containing today's date.
	</para>
	</section>
	
	<section>
	<title>Defining executable metadata</title>
	<para>
		<emphasis>NOTE:</emphasis> 
		This is only supported in .bb and .bbclass files.
	</para>
	<para>
		<screen>do_mytask () { echo "Hello, world!" }
		</screen>
	</para>
	<para>This is essentially identical to setting a variable, except that
		this variable happens to be executable shell code.
	</para>
	<para>
		<screen>python do_printdate () { import time print
			time.strftime('%Y%m%d', time.gmtime()) }
		</screen>
	</para>
	<para>This is the similar to the previous, but flags it as Python so
		that BitBake knows it is Python code.
	</para>
	</section>
	
	<section>
	<title>Defining Python functions into the global Python namespace
	</title>
	<para>
		<emphasis>NOTE:</emphasis> 
		This is only supported in .bb and .bbclass files.
	</para>
	<para>
		<emphasis>NOTE:</emphasis>
		Python functions are in the global namespace so should use
		unique names.
	</para>
	<para>
		<screen>def get_depends(bb, d): if
			d.getVar('SOMECONDITION', True): return
			"dependencywithcond" else: return "dependency" 
			<varname>SOMECONDITION</varname> = "1" 
			<varname>DEPENDS</varname> = "${@get_depends(bb,
			d)}"
		</screen>
	</para>
	<para>This would result in 
		<varname>DEPENDS</varname> 
		containing 
		<literal>dependencywithcond</literal>.
	</para>
	</section>
	
	<section>
	<title>Variable flags</title>
	<para>Variables can have associated flags which provide a way of 
		tagging extra information onto a variable.
		Several flags are used internally by BitBake but they can be 
		used externally too if needed.
		The standard operations mentioned above also work on flags.
	</para>
	<para>
		<screen>
			<varname>VARIABLE</varname>
			[
			<varname>SOMEFLAG</varname>
			] = "value"
		</screen>
	</para>
	<para>In this example,
		<varname>VARIABLE</varname> 
		has a flag, 
		<varname>SOMEFLAG</varname> 
		that is set to 
		<literal>value</literal>.
	</para>
	</section>
	
	<section>
	<title>Inheritance</title>
	<para>
		<emphasis>NOTE:</emphasis> This is only supported in .bb 
		and .bbclass files.
	</para>
	<para>The <literal>inherit</literal> directive is a means of specifying
		what classes of functionality your .bb requires.
		It is a rudimentary form of inheritance.
		For example, you can easily abstract out the tasks involved in
		building a package that uses autoconf and automake, and put 
		that into a bbclass for your packages to make use of. 
		A given bbclass is located by searching for 
		classes/filename.bbclass in 
		<envar>BBPATH</envar>, 
		where filename is what you inherited.
	</para>
	</section>
	
	<section>
	<title>Tasks</title>
	<para>
		<emphasis>NOTE:</emphasis> 
		This is only supported in .bb and .bbclass files.
	</para>
	<para>In BitBake, each step that needs to be run for a given .bb is 
		known as a task. 
		There is a command <literal>addtask</literal> to add new 
		tasks (must be a defined Python executable metadata and must
		start with  <quote>do_</quote>) and describe intertask
		dependencies.
	</para>
	<para>
		<screen>python do_printdate () { import time print
			time.strftime('%Y%m%d', time.gmtime()) } addtask printdate
			before do_build
		</screen>
	</para>
	<para>This defines the necessary Python function and adds it as a task 
		which is now a dependency of do_build, the default task.
		If anyone executes the do_build task, that will result in 
		do_printdate being run first.
	</para>
	</section>
	
	<section>
	<title>Task Flags</title>
	<para>Tasks support a number of flags which control various 
		functionality of the task.
		These are as follows:
	</para>
	<para>'dirs' - directories which should be created before the task 
		runs
	</para>
	<para>'cleandirs' - directories which should created before the task 
		runs but should be empty
	</para>
	<para>'noexec' - marks the tasks as being empty and no execution 
		required.
		These are used as dependency placeholders or used when added
		tasks need to be subsequently disabled.
	</para>
	<para>'nostamp' - don't generate a stamp file for a task.  
		This means the task is always executed.
	</para>
	<para>'fakeroot' - this task needs to be run in a fakeroot
		environment, obtained by adding the variables in FAKEROOTENV 
		to the environment.
	</para>
	<para>'umask' - the umask to run the task under.
	</para>
	<para>For the 'deptask', 'rdeptask', 'depends', 'rdepends'and 
		'recrdeptask' flags please see the dependencies section.
	</para>
	</section>
	
	<section>
	<title>Events</title>
	<para>
		<emphasis>NOTE:</emphasis> 
		This is only supported in .bb and .bbclass files.
	</para>
	<para>BitBake allows installation of event handlers.
		Events are triggered at certain points during operation, such 
		as the beginning of operation against a given .bb, the start of
		a given task, task failure, task success, et cetera.
		The intent is to make it easy to do things like email 
		notification on build failure.
	</para>
	<para>
		<screen>addhandler myclass_eventhandler python
			myclass_eventhandler() { from bb.event import getName from
			bb import data print("The name of the Event is %s" %
			getName(e)) print("The file we run for is %s" %
			data.getVar('FILE', e.data, True)) }
		</screen>
	</para>
	<para>This event handler gets called every time an event is triggered.
		A global variable 
		<varname>e</varname> 
		is defined. 
		<varname>e</varname>.data
		contains an instance of bb.data.
		With the getName(<varname>e</varname>) method one can get the 
		name of the triggered event.
	</para>
	<para>The above event handler prints the name of the event and the 
		content of the 
		<varname>FILE</varname> 
		variable.
	</para>
	</section>
	
	<section>
	<title>Variants</title>
	<para>Two BitBake features exist to facilitate the creation of multiple
		buildable incarnations from a single recipe file.
	</para>
	<para>The first is 
		<varname>BBCLASSEXTEND</varname>. 
		This variable is a space separated list of classes used to 
		"extend" the recipe for each variant.
		As an example, setting
		<screen>BBCLASSEXTEND = "native"</screen> 
		results in a second incarnation of the current recipe being
		available.
		This second incarnation will have the "native" class 
		inherited.
	</para>
	<para>The second feature is
		<varname>BBVERSIONS</varname>. 
		This variable allows a single recipe to build multiple versions
		of a project from a single recipe file, and allows you to 
		specify conditional metadata (using the
		<varname>OVERRIDES</varname> mechanism) for a single version,
		or an optionally named range of versions:
	</para>
	<para>
		<screen>BBVERSIONS = "1.0 2.0 git" SRC_URI_git =
			"git://someurl/somepath.git"
		</screen>
	</para>
	<para>
		<screen>BBVERSIONS = "1.0.[0-6]:1.0.0+ \ 1.0.[7-9]:1.0.7+" 
			SRC_URI_append_1.0.7+ =	"file://some_patch_which_the_new_versions_need.patch;patch=1"
		</screen>
	</para>
	<para>Note that the name of the range will default to the original 
		version of the recipe, so given OE, a recipe file of
		foo_1.0.0+.bb will default the name of its versions to 
		1.0.0+. 
		This is useful, as the range name is not only placed into 
		overrides; it's also made available for the metadata to use in 
		the form of the
		<varname>BPV</varname> 
		variable, for use in file:// search paths 
		(<varname>FILESPATH</varname>).
	</para>
	</section>
	
	<section>
	<title>Variable interaction: Worked Examples</title>
	<para>Despite the documentation of the different forms of variable 
		definition above, it can be hard to work out what happens when 
		variable operators are combined.
		This section documents some common questions people have 
		regarding the way variables interact.
	</para>
	</section>
	
	<section>
	<title>Override and append ordering</title>
	<para>There is often confusion about which order overrides and the 
		various append operators take effect.
	</para>
	<para>
		<screen>
			<varname>OVERRIDES</varname> 
			= "foo" 
			<varname>A_foo_append</varname> 
			= "X"</screen>
	</para>
	<para>In this case, X is unconditionally appended to the variable 
		<varname>A_foo</varname>. 
		Since foo is an override, A_foo would then replace 
		<varname>A</varname>.</para>
	<para>
		<screen>
			<varname>OVERRIDES</varname> 
			= "foo" 
			<varname>A</varname> 
			= "X" 
			<varname>A_append_foo</varname> 
			= "Y"
		</screen>
	</para>
	<para>In this case, only when foo is in OVERRIDES, Y is appended to the
		variable 
		<varname>A</varname> 
		so the value of 
		<varname>A</varname> 
		would become XY (NB: no spaces are appended).
	</para>
	<para>
		<screen>
			<varname>OVERRIDES</varname> 
			= "foo" 
			<varname>A_foo_append</varname> 
			= "X" 
			<varname>A_foo_append</varname>
			+= "Y"
		</screen>
        </para>
	<para>This behaves as per the first case above, but the value of
		<varname>A</varname> 
		would be "X Y" instead of just "X".
	</para>
	<para>
		<screen>
			<varname>A</varname> 
			= "1" 
			<varname>A_append</varname> 
			= "2" 
			<varname>A_append</varname> 
			= "3" 
			<varname>A</varname>
			+= "4" 
			<varname>A</varname>
			.= "5"
		</screen>
        </para>
	<para>Would ultimately result in 
		<varname>A</varname> 
		taking the value "1 4523" since the _append operator executes 
		at the same time as the expansion of other overrides.
	</para>
	</section>
	
	<section>
	<title>Key Expansion</title>
	<para>Key expansion happens at the data store finalisation time just
		before overrides are expanded.
	</para>
	<para>
		<screen>
			<varname>A${B}</varname> 
			= "X"
			<varname>B</varname> 
			= "2" 
			<varname>A2</varname> 
			= "Y"
		</screen>
	</para>
	<para>So in this case 
		<varname>A2</varname> 
		would take the value of "X".
	</para>
	</section>
	
	<section>
	<title>Dependency handling</title>
	<para>BitBake handles dependencies at the task level since to allow for
		efficient operation with multiple processes executing in 
		parallel, a robust method of specifying task dependencies is
		needed.
	</para>
	</section>
	
	<section>
	<title>Dependencies internal to the .bb file</title>
	<para>Where the dependencies are internal to a given .bb file, the
		dependencies are handled by the previously detailed addtask
		directive.
	</para>
	</section>
	
	<section>
	<title>Build Dependencies</title>
	<para>DEPENDS lists build time dependencies.
		The 'deptask' flag for tasks is used to signify the task of 
		each item listed in DEPENDS which must have completed before 
		that task can be executed.
	</para>
	<para>
		<screen>do_configure[deptask] =
			"do_populate_staging"
		</screen>
	</para>

	<para>means the do_populate_staging task of each item in DEPENDS must 
		have completed before do_configure can execute.
	</para>
	</section>
	
	<section>
	<title>Runtime Dependencies</title>
	<para>The PACKAGES variable lists runtime packages and each of these 
		can have RDEPENDS and RRECOMMENDS runtime dependencies.
		The 'rdeptask' flag for tasks is used to signify the task of
		each item runtime dependency which must have completed before
		that task can be executed.
	</para>
	<para>
		<screen>do_package_write[rdeptask] = 
			"do_package"
		</screen>
	</para>
	<para>means the do_package task of each item in RDEPENDS must have 
		completed before do_package_write can execute.
	</para>
	</section>
	
	<section>
	<title>Recursive Dependencies</title>
	<para>These are specified with the 'recrdeptask' flag which is used
		to signify the task(s) of dependencies which must have
		completed before that task can be executed.
		It works by looking though the build and runtime dependencies
		of the current recipe as well as any inter-task dependencies
		the task has, then adding a dependency on the listed task.
		It will then recurse through the dependencies of those tasks
		and so on.
	</para>
	<para>It may be desireable to recurse not just through the dependencies
		of those tasks but through the build and runtime dependencies 
		of dependent tasks too.
		If that is the case, the taskname itself should be referenced 
		in the task list, e.g. do_a[recrdeptask] = "do_a do_b".
	</para>
	</section>
	
	<section>
	<title>Inter task</title>
	<para>The 'depends' flag for tasks is a more generic form which allows 
		an interdependency on specific tasks rather than specifying 
		the data in DEPENDS.
	</para>
	<para>
		<screen>do_patch[depends] = 
			"quilt-native:do_populate_staging"
		</screen>
        </para>
	<para>means the do_populate_staging task of the target 
		quilt-native must have completed before the do_patch can
		execute.
	</para>
	<para>The 'rdepends' flag works in a similar way but takes targets in 
		the runtime namespace instead of the build time dependency 
		namespace.
	</para>
	</section>
	
	<section>
	<title>Parsing</title>
	<section>
		<title>Configuration files</title>
		<para>The first kind of metadata in BitBake is configuration
			metadata.
			This metadata is global, and therefore affects
			<emphasis>all</emphasis> 
			packages and tasks that are executed.
		</para>
		<para>BitBake will first search the current working directory 
			for an optional "conf/bblayers.conf" configuration 
			file.
			This file is expected to contain a BBLAYERS variable 
			that is a space delimited list of 'layer' directories.
			For each directory in this list, a "conf/layer.conf" 
			file will be searched for and parsed with the LAYERDIR 
			variable being set to the directory where the layer was
			found.
			The idea is these files will setup BBPATH and other 
			variables correctly for a given build directory
			automatically for the user.
		</para>
		<para>BitBake will then expect to find 'conf/bitbake.conf'
			somewhere in the user specified 
			<envar>BBPATH</envar>. 
			That configuration file generally has include 
			directives to pull in any other metadata  (generally 
			files specific to architecture, machine,
			<emphasis>local</emphasis>and so on).
		</para>
		<para>Only variable definitions and include directives are
			allowed in .conf files.
		</para>
	</section>
	
	<section>
	<title>Classes</title>
	<para>BitBake classes are our rudimentary inheritance mechanism. 
		As briefly mentioned in the metadata introduction, they're 
		parsed when an
		<literal>inherit</literal> 
		directive is encountered, and they are located in classes/ 
		relative to the directories in
		<envar>BBPATH</envar>.
	</para>
	</section>
	
	<section>
        <title>.bb files</title>
	<para>A BitBake (.bb) file is a logical unit of tasks to be executed.
		Normally this is a package to be built.
		Inter-.bb dependencies are obeyed.
		The files themselves are located via the 
		<varname>BBFILES</varname> 
		variable, which is set to a space separated list of .bb files,
		and does handle wildcards.
	</para>
	</section>
    </section>
    </para>
  
</chapter>