MMS • RSS
Article originally posted on InfoQ. Visit InfoQ
Support for the Git wire Protocol version 2 is now available in the latest official version of the Git client, Git 2.18, along with other new features aimed at improving performance and UI.
Git protocol version 2 (v2) was merged a couple of weeks ago to Git’s master branch, shortly before core team member Brandon Williams announced it publicly. Git’s wire protocol defines how Git client and server communicate to carry through clone, fetch, and push operations. This new version aims to improve performance and is better suited for future improvements, according to Williams.
The main motivation for the new protocol was to enable server side filtering of references (branches and tags).
This means the Git server does not need to send the client a list of all the references in the repository and have the client do the filtering. In large repositories, there can be a huge number of references, amounting to several MBs of data being transferred irrespective of the operation the client wants to carry through. When using v2, Git servers will filter references based on the required operation before sending the list back to the client. For example, says Williams, if you are updating a branch that is only a few commits behind its remote counterpart, or checking if your branch is up-to-date, transferring the whole list of references is just a waste of time and bandwidth. Indeed, based on Google’s internal use of v2, Williams states that accessing a large repository, such as Chrome’s, which contains over 500k branches and tags, may be up to three times faster compared to version 1. Additionally, the new protocol makes it easier to implement new features like ref-in-want and fetching and pushing symrefs.
Git clients using v2 will be still able to talk to older servers that do not support it. This happens thanks to a conscious design decision to use a side-channel to send the additional information required by v2. Older servers will simply ignore the side channel and respond with the full list of references.
To allow developers to select the protocol version they would like to use, Git now includes a new
-c command line option, as in the following example:
git -c protocol.version=2 ls-remote
To make v2 the default, you can change Git configuration:
git config --global protocol.version=2
Another new feature in Git 2.18 aimed to improve performance is the serialized commit-graph. Basically, now Git can store the commit graph structure in a file along with some extra metadata to speed up graph walks. This is especially effective when listing and filtering commit history or computing merge bases. According to Derrick Stole, member of the team at Microsoft that implemented the feature, enabling it will provide a 75–99% speedup on such operations when run against large repos such as the Linux kernel’s or Git’s itself. Git commit graph is still experimental since a few Git features do not play well with the commit graph, such as shallow clones, replace-objects, and commit grafts. If you do not need them, you can enable the commit graph by running
git config core.commitGraph true.
Read the full list of features in Git 2.18 in the official release notes.