Wiki Tools

  • Find Page
  • Recent Changes
  • Page History
  • Attachments

Revision 10 as of 2007-10-11 17:04:03

Clear message

A brief review of the Bazaar SCM model

High Level Objects


A location storing committed data. This can be shared between multiple branches.


A branch is a pointer into the revision graph and associated metadata.

Gives a pointer into the DAG the repository holds. Also a location for storing branch-specific information. (Associated branches, user configuration, etc.) These are uniquely identified by URL (which may just be a local path). They always have an associated Repository, which is either in the same directory, or in a parent directory (Branches are always in a subdir of their Repository ?)

Working Tree

Files in a location that can be edited by the user. Always associated with a Branch, but may not be in the same location.

Repository Objects

This section describes the data model for how data is stored.


A revision is a marker for of a Tree and its associated metadata. Each revision is a record with the following fields:

  • Revision_id - Unique identifier for this revision.
  • Parents - A list of Revision_id's.
  • Properties - A dictionary of metadata. Generally, the meta data is at least: * Committer - Who committed the revision. * Date - When the revision was committed * (other metadata)

The DAG (directed acyclic graph) implied by the set of revisions is the revision graph of the project. (add here)

A Bazaar revision is equivalent to a git commit object.


An inventory is literally a snapshot of the Tree for a particular revision. There is a 1-1 mapping Revision to Inventory. The snapshot is represented as a list of (file_id, revision) tuples, defining the state of the text (i.e. contents of the file) for a particular revision. In detail, an Inventory is a series of records each with the following fields:

  • File_id - Unique identifier for the file. Same for all revisions of the file.
  • Revision - The Revision_id in which this variation of this file was introduced.
  • Parent ID - File_id of the parent directory, or ROOT_ID.
  • Name - The filename relative to the parent directory.
  • Executable - Indicates that this file should be executable on systems that support it.
  • Text SHA1 - SHA-1 hash of the full text of the file (not the delta).
  • Text size - Size in bytes of the text of the file.
  • The reason there is a revision ID for each record in the inventory is for storing the state of files which did not change in this revision. The revision may point to an earlier Revision_id than the revision associated with this Inventory.
  • The tree implied by the Parent ID fields is a directory tree, and is not related to the revision tree.
  • This is similar to a Hg manifest.


Is more of an API, which is used as an interface managing the Inventory, and access to the actual texts of the files. (A WorkingTree or a RevisionTree.)

Alternatively, we could define Tree, and ignore the specifics of Inventory.


An arbitrary unqiue string which identifies a specific snaphot.


Rather than being defined soley by their path, files are uniquely identified by a file_id (arbitrary string). As files are moved and/or renamed, their file_id stays associated with the same logical file.

(Allows comparing 2 trees with arbitrary large history, and an arbitrary large number of renames, etc.)