Contents
As a Bazaar developer there are a few things you need to know about configuration:
The first sections in this document summarize the steps needed when adding a new configuration item, the rest of the document gives more internal details on how this is implemented.
Options values are obtained with stack.get(option_name) where stack is one of the daughter classes of config.Stack, see their docstrings for a description of which sections are used from which stores.
The value returned is of the type declared for that Option and if nothing is specifically declared you will get the default for that option.
You add a new Option to the option_registry, either inside bzrlib/config.py or during initialization of your plugin (use register_lazy in this case). New plugins should have systematic hierarchical names so that related values are grouped together:
Homeoption_registry.register(
Option('dirstate.fdatasync', default=True,
from_unicode=bool_from_store,
help="Flush dirstate changes onto physical disk? ...."))
You then need to decide which stack is appropriate to implement the Option policy:
The docstrings for the existing stacks cover most of the known use cases.
Just reading an option is what is needed most of the time, modifying option values or removing options is usually something that is not automated but left to the user (with bzr config).
Nevertheless, if you need to save a modified option value, use .set(option_name, value) and .remove(option_name) to delete the option. Both methods are provided by the Stack object.
But before doing that, you must be sure that the stack you’re using have a writable section (this is true for GlobalStack which uses the DEFAULT section in bazaar.conf and for BranchStack``which uses the no-name section in ``branch.conf).
There is (as of late 2011) some older and some newer configuration code. The old code has specific methods for various checks or uses classes like GlobalConfig. Don’t add to to it; try to remove it.
If you encounter an option using the old code you may want to migrate it. This generally involves:
The new config code provides some help for commonly encountered use cases that can allow further simplifications like:
If you start migrating a given option to the config stacks, don’t stop mid-way, all its uses should be covered (tests included). There are some edge cases where updates via one API will be not be seen by the other API (see http://pad.lv/948339 and http://pad.lv/948344 for details). Roughly, the old API always trigger an IO while the new one cache values to avoid them. This works fine as long as a given option is handled via a single API.
Stacks capture the various places an option can be declared by the user with associated levels of generality and query them in the appropriate order returning the first definition found. For example, the append_revisions_only option may be declared for all branches of a user in bazaar.conf, or for a hierarchy of branches in locations.conf or in a single branch in branch.conf.
Defining a new stack means you need a new way to expose these levels to the user that is not covered by the existing stacks.
This is achieved by declaring:
Mapping different sections to different stacks is a powerful way to organize the options and provide various levels of configuration to the user. This is achieved with Store and SectionMatcher objects.
The following stores are used by bzr in ways that illustrate various uses of sections.
bzr itself defines two sections here:
Plugins can define either additional options in the DEFAULT section or new sections for their own needs (this is not especially encouraged though). The bzr-bookmarks plugin defines a BOOKMARKS section there for example.
The Ubuntu package importer defines a store and two stacks involving pkgimport.conf. A no-name section contains the options common to all packages and sections named after their corresponding package can also be defined.
The ImporterStack uses locations.conf and the no-name section in pkgimport.conf for the importer options.
The PackageStack uses only pkgimport.conf and uses the section name after the package followed by the no-name section.
bzr defines sections corresponding to URLs there and includes the relevant sections in LocationStack and BranchStack. No no-name section is recognized in this file.
This file defines the option for a given branch and uses only the no-name section.
The Option object is used to define its properties:
name: a name: a valid python identifier (even if it’s not used as an identifier in python itself). This is also used to register the option.
from_unicode: a callable accepting a unicode string and returning a suitable value for the option. If the string cannot be coerced it should return None.
override_from_env: a list of environment variables. The first variable set will be used as the option value overriding any other definition of the option.
default: the default value that Stack.get() should return if no value can be found for the option. This can also be a callable as long as it returns a unicode string.
default_from_env: a list of environment variables. The first variable set will provide a default value overriding ‘default’ which remains the default value if no environment variable is set.
help: a doc string describing the option, the first line should be a summary and can be followed by a blank line and a more detailed explanation. This will be displayed to the user with:
bzr help <option name>
invalid: the action to be taken when an invalid value is encountered in a store (during a Stack.get()).
The value of an option is a unicode string or None if it’s not defined. By using from_unicode you can turn this string into a more appropriate representation.
If you need a list value, you should use ListOption instead.
For options that take their values from a Registry object, RegistryOption can be used. This will automatically take care of looking up the specified values in the dictionary and documenting the possible values in help.
Options are grouped into sections which share some properties with the well known dict objects:
MutableSection is needed to set or remove an option, ReadOnlySection should be used otherwise.
Options can be persistent in which case they are saved into Stores.
config.Store defines the abstract interface that all stores should implement.
This object doesn’t provide direct access to the options, it only provides access to Sections. This is deliberate to ensure that sections can be properly shared by reusing the same underlying objects. Accessing options should be done via the Section objects.
A Store can contain one or more sections, each section is uniquely identified by a unicode string.
config.IniFileStore is an implementation that use ConfigObj.
Depending on the object it is associated with (or not) a Store also needs to implement a locking mechanism. LockableIniFileStore implements such a mechanism for IniFileStore based stores.
Classes are provided for the usual Bazaar configuration files and could be used as examples to define new ones if needed. The associated tests provides a basis for new classes which only need to register themselves in the right places to inherit from the existing basic tests and add their own specific ones.
A Store defines how option values are stored, this includes:
For some contexts, only some sections from a given store will apply. The SectionMatcher objects are used to define which sections in a store apply to a given context.
The main constraint here is that a SectionMatcher should delay the loading of the associated store as long as possible. The constructor should collect all data needed for the selection and uses it while processing the sections in get_sections.
Only ReadOnlySection objects are manipulated here but a SectionMatcher can return dedicated Section objects to provide additional context (the LocationSection add an extra_path attribute to implement the section local options for example). If no sections match, an empty list is returned.
Options local to a section can be defined for special purposes and be handled by Section.get(). One such option is relpath which is defined in LocationSection as an alternative to the appendpath policy.
For appendpath, the LocationSection will carry extra_path as the relative path between the section name and the location used. relpath will be available as a Section local option with the same value. basename will carry the location base name and be available as a local option with the same name. Note that such options can only be expanded inside the section that defines them.
Specific section matchers can be implemented by overriding get_sections or just match.
bzrlib provides:
An option can take different values depending on the context it is used. This can involve configuration files, options from the command line, default values in bzrlib and then some.
Such a context is implemented by creating a list of Section stacked upon each other. A Stack can then be asked for an option value and returns the first definition found.
This provides a great flexibility to decide priorities between sections when the stack is defined without to worry about them in the code itself.
A stack also defines a mutable section (which can be None) to handle modifications.
Many sections (or even stores) are aimed at providing default values for an option but these sections shouldn’t be modified lightly as modifying an option used for different contexts will indeed be seen by all these contexts.
Default values in configuration files are defined by users. Developers shouldn’t have to modify them, as such, no mechanism nor heuristics are used to find which section (or sections) should be modified.
A Stack defines a mutable section when there is no ambiguity. If there is one, then the user should be able to decide and in this case a new Stack can be created cheaply.
Different stacks can be created for different purposes, the existing GlobalStack, LocationStack and BranchStack can be used as basis or examples. These classes are the only ones that should be used in code, Stores can be used to build them but shouldn’t be used otherwise, ditto for sections. Again, the associated tests could and should be used against the created stacks.
Also note that MemoryStack can be used without any disk resources which allows for simpler and faster tests. A common pattern is to accept a config parameter related to a given feature and test it with predefined configurations without involving the whole stack. bzrlib.tests.test_commit, bzrlib.tests.test_gpg and bzrlib.tests.test_smtp_connection are good examples.