Mercurial 3.4 Released

May 04, 2015 at 12:40 PM | categories: Mercurial, Mozilla

Mercurial 3.4 was released on May 1 (following Mercurial's time-based schedule of releasing a new version every 3 months).

3.4 is a significant release for a few reasons.

First, the next version of the wire protocol (bundle2) has been marked as non-experimental on servers. This version of the protocol paves over a number of deficiencies in the classic protocol. I won't go into low-level details. But I will say that the protocol enables some rich end-user experiences, such as having the server hand out URLs for pre-generated bundles (e.g. offload clones to S3), atomic push operations, and advanced workflows, such as having the server rebase automatically on push. Of course, you'll need a server running 3.4 to realize the benefits of the new protocol. won't be updated until at least June 1.

Second, Mercurial 3.4 contains improvements to the tags cache to make performance concerns a thing of the past. Due to the structure of the Firefox repositories, the previous implementation of the tags cache could result in pauses of dozens of seconds during certain workflows. The problem should go away with Mercurial 3.4. Please note that on first use of Mercurial 3.4, your repository may perform a one-time upgrade of the tags cache. This will spin a full CPU core and will take up to a few minutes to complete on Firefox repos. Let it run to completion and performance should not be an issue again. I wrote the patches to change the tags cache (with lots of help from Pierre-Yves David, a Mercurial core contributor). So if you find anything wrong, I'm the one to complain to.

Third, the HTTP interface to Mercurial (hgweb) now has JSON output for nearly every endpoint. The implementation isn't yet complete, but it is better than nothing. But, it should be good enough for services to start consuming it. Again, this won't be available on until the server is upgraded on June 1 at the earliest. This is a feature I added to core Mercurial. If you have feature requests, send them my way.

Fourth, a number of performance regressions introduced in Mercurial 3.3 were addressed. These performance issues frequently manifested during hg blame operations. Many Mozillians noticed them on when looking at blame through the web interface.

For a more comprehensive list of changes, see my post about the 3.4 RC and the official release notes.

3.4 was a significant release. There are compelling reasons to upgrade. That being said, there were a lot of changes in 3.4. If you want to wait until 3.4.1 is released (scheduled for June 1) so you don't run into any regressions, nobody can fault you for that.

If you want to upgrade, I recommend reading the Mercurial for Mozillians Installation Page.