blob: ebe8e2636a8574aedb3c39dd86dcb34704054f69 [file] [log] [blame]
.. _format:
The BuildStream Format
======================
At the core of BuildStream is a data model of :mod:`Elements <buildstream.element>` which
are parsed from ``.bst`` files in a project directory and configured from a few different
sources.
This page should tell you everything you need to know about the base YAML format
which BuildStream uses.
This will not cover the configurations needed for various plugins, plugin configurations
are documented in the plugins themselves.
The Project Directory
---------------------
A BuildStream project is a directory consisting of:
* A project configuration file
* BuildStream element files
* User defined Plugins
A typical project structure may look like this::
myproject/project.conf
myproject/elements/element1.bst
myproject/elements/element2.bst
myproject/elements/...
myproject/plugins/customelement.py
myproject/plugins/customelement.yaml
myproject/plugins/...
Except for the project configuration file, the user is allowed to structure
their project directory in any way. For documentation on the format of the project
configuration file, refer to the :ref:`projectconf` documentation.
Simpler projects may choose to place all element definition files at the
root of the project directory while more complex projects may decide to
put stacks in one directory and other floating elements into other directories,
perhaps placing deployment elements in another directory, this is all fine.
The important part to remember is that when you declare dependency relationships,
a project relative path to the element one depends on must be provided.
Element Composition
-------------------
Below are the various sources of configuration which go into an element in the order
in which they are applied. Configurations which are applied later have a higher priority
and override configurations which precede them.
1. Builtin Defaults
~~~~~~~~~~~~~~~~~~~
The :ref:`projectconf` provides a set of default values for *variables*
and the *environment* which are all documented with your copy of BuildStream.
2. Project Configuration
~~~~~~~~~~~~~~~~~~~~~~~~
The project wide defaults are now applied on top of builtin defaults. If you specify
anything in the *variables* or *environment* sections in your ``project.conf`` then it
will override the builtin defaults.
3. Element Defaults
~~~~~~~~~~~~~~~~~~~
Elements are all implemented as plugins. Each plugin installs a ``.yaml`` file along side
their plugin to define the default *variables*, *environment* and *config*. The *config*
is element specific and as such this is the first place where defaults can be set on the
*config* section.
The *variables* and *environment* specified in the declaring plugin's defaults here override
the project configuration defaults for the given element ``kind``.
4. Project Configuration
~~~~~~~~~~~~~~~~~~~~~~~~
The ``project.conf`` now gives you another opportunity to override *variables*, *environment*
and *config* sections on a per element basis.
Configurations specified in the *elements* section of the ``project.conf`` will override
the given element's default.
5. Element Declarations
~~~~~~~~~~~~~~~~~~~~~~~
Finally, after having resolved any `Architecture Conditionals`_ or `Variant Conditionals`_
in the parsing phase of loading element declarations; the configurations specified in a
``.bst`` file have the last word on any configuration in the data model.
Element Basics
--------------
Here is a rather complete example using the autotools element kind and git source kind:
.. code:: yaml
# Specify the kind of element this is
kind: autotools
# Specify some dependencies
depends:
- elements/element1.bst
- elements/element2.bst
# Specify the source which should be built
sources:
- kind: git
url: upstream:modulename.git
track: master
ref: d0b38561afb8122a3fc6bafc5a733ec502fcaed6
# Override some variables
variables:
sysconfdir: %{prefix}/etc
# Tweak the sandbox shell environment
environment:
LD_LIBRARY_PATH: /some/custom/path
# Specify the configuration of the element
config:
# Override autotools element default configure-commands
configure-commands:
- "%{configure} --enable-fancy-feature"
# Specify public domain data, visible to other elements.
public:
bst:
integration-commands:
- /usr/bin/update-fancy-feature-cache
For most use cases you would not need to specify this much detail, we've provided
details here in order to have a more complete initial example.
Let's break down the above and give a brief explanation of what these attributes mean.
Kind
~~~~
.. code:: yaml
# Specify the kind of element this is
kind: autotools
The ``kind`` attribute specifies which plugin will be operating on the element's input to
produce its output. Plugins define element types and each of them can be referred to by
name with the ``kind`` attribute.
Depends
~~~~~~~
.. code:: yaml
# Specify some dependencies
depends:
- elements/element1.bst
- elements/element2.bst
Relationships between elements are specified with the ``depends`` attribute. Element
definitions may depend on other elements by specifying the project relative path
to the elements on which they depend here. See `Dependencies`_ for more information
on the dependency model.
Sources
~~~~~~~
.. code:: yaml
# Specify the source which should be built
sources:
- kind: git
url: upstream:modulename.git
track: master
ref: d0b38561afb8122a3fc6bafc5a733ec502fcaed6
Here we specify some input for the element, any number of sources may be specified.
By default the sources will be staged in the root of the element's build directory
in the build sandbox, but sources may specify a ``directory`` attribute to control
where the sources will be staged. The ``directory`` attribute may specify a build
sandbox relative subdirectory.
For example, one might encounter a component which requires a separate data package
in order to build itself, in this case the sources might be listed as:
.. code:: yaml
sources:
# Specify the source which should be built
- kind: git
url: upstream:modulename.git
track: master
ref: d0b38561afb8122a3fc6bafc5a733ec502fcaed6
# Specify the data package we need for build frobnication,
# we need it to be unpacked in a src/frobdir
- kind: tarball
directory: src/frobdir
url: data:frobs.tgz
sha256sum: 9d4b1147f8cf244b0002ba74bfb0b8dfb3...
Like Elements, Source types are plugins which are indicated by the ``kind`` attribute.
Asides from the common ``kind`` and ``directory`` attributes which may be applied to all
Sources, refer to the Source specific documentation for meaningful attributes for the
particular Source.
Variables
~~~~~~~~~
.. code:: yaml
# Override some variables
variables:
sysconfdir: "%{prefix}/etc"
Variables can be declared or overridden from an element. Variables can also be
declared and overridden in the :ref:`projectconf`
See `Using Variables`_ below for a more in depth discussion on variables in BuildStream.
Environment
~~~~~~~~~~~
.. code:: yaml
# Tweak the sandbox shell environment
environment:
LD_LIBRARY_PATH: /some/custom/path
Environment variables can be set to literal values here, these environment
variables will be effective in the :mod:`Sandbox <buildstream.sandbox>` where
build instructions are run for this element.
Environment variables can also be declared and overridden in the :ref:`projectconf`
Config
~~~~~~
.. code:: yaml
# Specify the configuration of the element
config:
# Override autotools element default configure-commands
configure-commands:
- "%{configure} --enable-fancy-feature"
Here we configure the element itself. The autotools element provides sane defaults for
building sources which use autotools. Element default configurations can be overridden
in the ``project.conf`` file and additionally overridden in the declaration of an element.
For meaningful documentation on what can be specified in the ``config`` section for a given
element ``kind``, refer to the element specific documentation.
.. _format_public:
Public
~~~~~~
.. code:: yaml
# Specify public domain data, visible to other elements.
public:
bst:
integration-commands:
- /usr/bin/update-fancy-feature-cache
Metadata declared in the ``public`` section of an element is visible to
any other element which depends on the declaring element in a given pipeline.
BuildStream itself consumes public data from the ``bst`` domain. The ``integration-commands``
demonstrated above for example, describe commands which should be run in an
environment where the given element is installed but before anything should be run.
An element is allowed to read domain data from any element it depends on, and users
may specify additional domains to be understood and processed by their own element
plugins.
.. _format_dependencies:
Dependencies
------------
The dependency model in BuildStream is simplified by treating software distribution
and software building as separate problem spaces. This is to say that one element
can only ever depend on another element but never on a subset of the product which
another element produces.
In this section we'll quickly go over the few features BuildStream offers in its
dependency model.
Expressing Dependencies
~~~~~~~~~~~~~~~~~~~~~~~
Dependencies in BuildStream are parameterizable objects, however as demonstrated
in the above example, they can also be expressed as strings as a convenience
shorthand whenever the default dependency attributes are suitable.
Shorthand:
.. code:: yaml
# Shorthand Dependencies
depends:
- elements/foo.bst
- elements/bar.bst
Dependency dictionary:
.. code:: yaml
# Fully specified dependency
depends:
- filename: elements/foo.bst
variant: bar
type: build
The ``variant`` attribute is explained below in `Variant Conditionals`_, and
the ``type`` attribute can be used to express the dependency type.
Dependency Types
~~~~~~~~~~~~~~~~
The dependency ``type`` attribute defines what the dependency is required for
and is essential to how BuildStream plots a build plan.
There are two types which one can specify for a dependency, ``build`` and ``runtime``.
A ``build`` dependency type states that the given element's product must
be staged in order to build the depending element. Depending on an element
which has ``build`` dependencies will not implicitly depend on that element's
``build`` dependencies.
A ``runtime`` dependency type states that the given element's product
must be present for the depending element to function. An element's
``runtime`` dependencies need not be staged in order to build the element.
If ``type`` is not specified, then it is assumed that the dependency is
required both at build time and runtime.
.. note::
It is assumed that a dependency which is required for building an
element must run while building the depending element. This means that
``build`` depending on a given element implies that that element's
``runtime`` dependencies will also be staged for the purpose of building.
.. _format_variables:
Using Variables
---------------
Variables in BuildStream are a way to make your build instructions and
element configurations more dynamic.
Referring to Variables
~~~~~~~~~~~~~~~~~~~~~~
Variables are expressed as ``%{...}``, where ``...`` must contain only
alphanumeric characters and the separators ``_`` and ``-``. Further, the
first letter of ``...`` must be an alphabetic character.
.. code:: yaml
This is release version %{version}
Declaring and Overriding Variables
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
To declare or override a variable, one need only specify a value
in the relevant *variables* section:
.. code:: yaml
variables:
hello: Hello World
You can refer to another variable while declaring a variable:
.. code:: yaml
variables:
release-text: This is release version %{version}
The order in which you declare variables is arbitrary, so long as there is no cyclic
dependency and that all referenced variables are declared, the following is fine:
.. code:: yaml
variables:
release-text: This is release version %{version}
version: 5.5
.. note::
It should be noted that variable resolution only happens after all `Element Composition`_
has already taken place.
This is to say that overriding ``%{version}`` at a higher priority will effect
the final result of ``%{release-text}``.
**Example:**
.. code:: yaml
kind: autotools
# Declare variable, expect %{version} was already declared
variables:
release-text: This is release version %{version}
config:
# Customize the installation
install-commands:
- |
%{make-install} RELEASE_TEXT="%{release-text}"
Architecture Conditionals
-------------------------
To BuildStream, an architecture is simply an arbitrary name that is associated with
the target architecture and compiler tuning. Conditional YAML segments can be applied
for a given target architecture, like so:
.. code:: yaml
kind: autotools
config:
something: 5
arches:
x86_64:
config:
something: 6
x86_32:
config:
something: 7
The ``arches`` attribute, if provided, overrides the element for a given architecture
name. It is not considered an error if the element does not provide an architecture
clause for the specific architecture BuildStream was launched to build for.
There is also a ``host-arches`` attribute, which operates in the same manner but
follows the *host* architecture rather than the *target* architecture.
In the above example we demonstrate that a given ``config`` attribute can be overridden
by an architecture conditional, this can however be done for any segment of the
element such as ``depends``, ``sources`` and ``public`` as well. It is however illegal
to override the element ``kind`` in any conditional.
Further, it should be noted that when applying elements to a list in the element YAML,
the conditional segments are *appended* to the parent list and do not replace the list
entirely.
Consider for example:
.. code:: yaml
kind: autotools
depends:
- elements/foo.bst
arches:
x86_64:
depends:
- elements/bar.bst
When targetting the ``x86_64`` architecture name, the above element YAML will
expand to the following YAML:
.. code:: yaml
kind: autotools
depends:
- elements/foo.bst
- elements/bar.bst
Variant Conditionals
--------------------
Variants are a way for a single element to provide multiple features. In contrast
with the architecture conditionals described above, which are resolved once for
the entirety of a pipeline; variant conditionals are selected by way of dependency.
Declaring Variants
~~~~~~~~~~~~~~~~~~
If an element declares any variants, it must declare at least two variants.
One of the variant declarations may be left empty so that they do not override
or effect the base element declaration, but at least two variant names must be
declared.
The first declared variant is the default. It may have whatever name you decide
to give it, but the default variant is what will be selected if all dependencies
on the given element are *ambivalent* of the variant.
Here is an example of how an element declares multiple variants:
.. code:: yaml
# Unconditionally depend on foo.bst
kind: autotools
depends:
- elements/foo.bst
variants:
# The default variant needs to disable flying ponies, or else
# our configure script bails out if the ponies are not found
- variant: default
config:
configure-commands:
- "%{configure} --without-flying-ponies"
# For the flying-ponies variant, we want to pull in the extra
# ponies so they will be available for flying
- variant: flying-ponies
depends:
- elements/ponies.bst
Depending on Variants
~~~~~~~~~~~~~~~~~~~~~
To depend on a specific variant of a given element, one must simply use
the ``variant`` attribute in a dependency that is expressed as a dictionary:
.. code:: yaml
# Depend on the flying-ponies variant of the foo element
depends:
- filename: elements/foo.bst
variant: flying-ponies
When depending on an element which advertizes variants without specifying
any particular variant, the dependency is said to be *ambivalent*.
Variant Resolution
~~~~~~~~~~~~~~~~~~
Variants of an element may augment the given element's dependencies, as
such there may be many possible ways in which a pipeline can be constructed.
As a rule, every variant of a given element should be buildable without
presenting any conflict when building the element as your pipeline *target*.
When resolving variants in a complex pipeline however, it is possible that
sibling elements depend on specific variants of common dependencies. BuildStream
will resolve which variants to build deterministically by traversing an
element's variants in the order of declaration, always choosing the first
buildable variant for any *ambivalent* dependency.
If there is no suitable build plan found for the selected variant of the
pipeline *target*, then it is considered a variant disagreement error and
the build will be aborted during the parse phase.