利用者:Nazg-gul/ReleaseTagging

提供: wiki
移動先: 案内検索

In this page i'd want to collect all the notes about how release process happens behind d the scenes, covering all the cases with branching, tagging, AHOYing and so.

Process overview

For the final releases and all the release candidates we're using tags. They are really helpful to grab working tree at a state which fully corresponds to the state of sources used for the actual release or release candidate.

However, it was proved that tagging straight away is the wrong strategy. We are always missing some minor tweaks to the sources after the tag and before doing actual builds. Git is does not like doing re-tagging. It is possible, but if someone fetched wrong tag he'll stick to that one. Git NEVER overrides tags in your local copy because of the security issues.

So we're using a bit different approach. Namely, we create a release branch and building from this branch. This way it's really easy to commit stuff if we forgot it, do re-AHOY and don't worry about those folks who fetched changes from the server.

Branch is only used for blender.git, addons and translations could be tagged right away. So far we never had issues with this.

Process itself

Here i'd like to put all the information about naming conventions and so, so the process keeps all the same with each release.

  • Tag addons and translations
For this the following command is used:
 $ git tag -a vX.XX -m "Tagging blender release(release cnadidate) X.XX"
This command i usually invoke from a separate checkout of addons and translations, but technically if you're accurate you can run this from submodule in blender's working tree.
In order to push tags to server:
 $ git push --tags
  • Update submodule hashes. This we do in the master branch, before branching. For this you first need to be sure submodules fetched all the tags:
 $ git submodule foreach git fetch --tags
After this you nee to checkout submodules to their tags. The easiest way to do this is to run
 $ git checkout vX.XX
from inside submodule directory. Currently we only need to do this for addons and locales. Contrib addons and scons we don't care about.
In order to commit submodules changes:
 ## Stage the changes
 $ git add release/scripts/addons release/datafiles/locale
 ## Actual commit. Need --allow-empty because of side-effect
 ## of ignoring changes to submodules in .gitmodules.
 ## Changes to submodules are not visible, but staged changes to
 ## them would be committed.
 $ git commit -a --allow-empty -m "Updating submodules to vX.XX"
  • When we're ready for the first release candidate, we create "blender-vX.XX-release" branch.
That's correct, it's called "-release" because it'll be used for backporting all the changes (which we really better do gradually, without saving all the backports to be done at the day of release).
  • When the AHOY happened all fine (builds are up and so) we need to tag the release branch:
 $ git tag -a vX.XX[rc1|rc2] -m "Tagging blender release(candidate) X.XX"
  • Once the final release is done, we do tag and delete the branch
 $ git push origin :blender-vX.XX-release

Stuff to be checked

  • Links in readme and slash points to the correct page
  • Subversion is to be bumped
  • Release cycle
  • version char
  • "Release Candidate" label on the splash