In version control software, a changeset (also known as commit and revision) is a set of alterations packaged together, along with meta-information about the alterations. A changeset describes the exact differences between two successive versions in the version control system's repository of changes. Changesets are typically treated as an atomic unit, an indivisible set, by version control systems. This is one synchronization model.
In the Git version control system a changeset is called a commit, not to be confused with the commit operation that is used to commit a changeset (or in Git's case technically a snapshot) to a repository.
Other version control systems also use other names to refer to changesets, for example Darcs calls them "patches", while Pijul refers to them as "changes".
Version control systems attach metadata to changesets. Typical metadata includes a description provided by the programmer (a "commit message" in Git lingo), the name of the author, the date of the commit, etc.
Unique identifiers are an important part of the metadata which version control systems attach to changesets. Centralized version control systems, such as Subversion and CVS simply use incrementing numbers as identifiers. Distributed version control systems, such as Git, generate a unique identifier by applying a cryptographic hash function to the changeset.
Because version control systems operate on changesets as atomic units, and because communication within development teams improves performance, there are certain best practices to follow when creating changesets. Only the 2 most significant are mentioned here, changeset content atomicity and changeset descriptions.
Changeset content should involve only one task or fix, and contain only code which works and does not knowingly break existing functionality.
Changeset descriptions should be short, recording why the modification was made, the modification's effect or purpose, and describing non-obvious aspects of how the change works.
Write the smallest amount of code possible to solve a problem. After identifying a problem or enhancement, the best way to try something new and untested is to divide the update into small batches of value that can easily and rapidly be tested with the end user to prove the validity of the proposed solution and to roll back in case it doesn’t work without deprecating the whole new functionality. ... Related to making small changes, atomic commits are a single unit of work, involving only one task or one fix (e.g. upgrade, bug fix, refactor). Atomic commits make code reviews faster and reverts easier, since they can be applied or reverted without any unintended side effects. The goal of atomic commits isn’t to create hundreds of commits but to group commits by context. For example, if a developer needs to refactor code and add a new feature, she would create two separate commits rather than create a monolithic commit which includes changes with different purposes.
Tracking changes ... provides an analysis of previous changes as well as a holistic view of the trajectory of the dataset. The history of the document ... gives on (sic) the purpose of the changes made.