Skip to content

Branches and releases

Rico Wind edited this page Aug 13, 2015 · 23 revisions

We have a rather simple branch setup for the dart project. Normally, everything is developed on the master branch. Features that are not yet ready for prime time are hidden under a flag, and enabled when sufficient stability has been proven. We occasionally create feature branches for landing big and disruptive changes. In addition to that, we have our dev and stable branches used for releasing the sdk.

Release cycle

Our normal release cycle is roughly 2 months long, but we don't make any guarantees, i.e., we may ship early if we feel that the stability is good or we may ship late if it is not. We don't normally follow a feature driven release cycle, but in some cases for larger changes to the language we may postpone a release to get all tools in sync.

On the normal 2 month cycle we use the first ~6 weeks doing full merges of master to dev, basically releasing a green build from master on dev roughly once a week. In case of bugs found quickly we do a patch release or another full push.

We then spend the last ~3 weeks stabilizing the dev channel, only cherry picking critical fixes. People continue working on the master with normal changes. In this 3 week period we do multiple batches of cherry picks a week, and release those on dev as we have them build. We call this cherry-pick season.

Once we have something that looks good we merge it to stable and release it there. During the 2 months we do patch security, crash and critical bug releases on stable based on the last stable version.

Getting your changes to dev channel during cherry pick season

See the cherry pick to dev page for all the details on how to get a change cherry picked.

Pushing to dev

Find a suitable green build on master, by looking at the bots, we call this #MASTER_HASH_TO_BASE_RELEASE_OFF. We call the version we are pushing $THE_VERSION_BEING_PUSHED, this is based on the actual values in tools/VERSION, but looks like this "1.11.0-dev.3.0".

git new-branch --upstream origin/dev release
git merge --no-commit #MASTER_HASH_TO_BASE_RELEASE_OFF

Update tools/VERSION (for these full pushes reset PRERELEASE_PATCH to 0, increase PRERELEASE by 1, the updated version is your $THE_VERSION_BEING_PUSHED)

git commit -a

Update the commit message, it should read like:

Version $THE_VERSION_BEING_PUSHED

Merge commit '#MASTER_HASH_TO_BASE_RELEASE_OFF' into dev

Note the s/release/dev in the last line

Sanity check:

git diff #MASTER_HASH_TO_BASE_RELEASE_OFF

should only give differences in the tools/VERSION file.

Push the changes:

git push

Tag the new release:

git tag -a $THE_VERSION_BEING_PUSHED -m $THE_VERSION_BEING_PUSHED
git push --tags

Cherry picking to dev

Please don't just cherry pick your changes to dev, we normally batch these up. Instead file a merge request (see above).

We do this using two branches to get a nice looking history on the dev branch, and to not trigger builds with non working versions on the bots (the bot git poller will do git log --first-parent when figuring out commits.

$THE_VERSION_BEING_PUSHED has the same meaning as above. The #HASH_1 ... #HASH_N are the cls we want to cherry pick

git new-branch --upstream origin/dev cherry
git cherry-pick #HASH_1
.
.
.
git cherry-pick #HASH_N

Now, we have all the cherry picks on the cherry branch, create the branch we will use for pushing, merge the cherry picks in (to get a merge node similar to what we have for normal full pushes). We use --no-ff to get the merge node instead of fast forward commits.

git new-branch --upstream origin/dev dev
git merge --no-ff --no-commit cherry

Edit tools/VERSION, increase PRERELEASE_PATCH by 1

git commit -a

Commit message should read:

Version $THE_VERSION_BEING_PUSHED

Cherry-pick #HASH_1 to dev
.
.
.
Cherry-pick #HASH_N to dev

Sanity check the log, make sure all the cherry picks are there and that the Version commit has the right structure.

Push the changes:

git push

Tag the new release:

git tag -a $THE_VERSION_BEING_PUSHED -m $THE_VERSION_BEING_PUSHED
git push --tags

Stable branch

The way we handle stable releases is more or less equal to the way we handle dev. When we have a good dev channel build after 3 weeks of cherry-picking, we merge that over to the stable branch:

Doing a new full stable build

Assume that the dev channel build we want to base this of is #DEV_HASH_TO_BASE_RELEASE_OFF

git new-branch --upstream origin/stable release
git merge --no-commit #MASTER_HASH_TO_BASE_RELEASE_OFF

There may very well be merge conflicts if we did patch releases on stable in the prior release. Always just solve these by taking the dev version. Because of this, always do a sanity check:

git diff #DEV_HASH_TO_BASE_RELEASE_OFF

This should only give a difference in the version file. If not, checkout the file(s) from #DEV_HASH_TO_BASE_RELEASE_OFF.

Update tools/VERSION (for these full pushes reset PRERELEASE_PATCH to 0, reset PRERELEASE to 0, reset PATCH to 0, set MINOR to the version from DEV (i.e., if we are doing this to release 1.12.0 set it to 12). As before, we call the version $THE_VERSION_BEING_PUSHED, on stable this is more simple, e.g., 1.12.0.

git push

Tag the new release:

git tag -a $THE_VERSION_BEING_PUSHED -m $THE_VERSION_BEING_PUSHED
git push --tags

Cherry picks to stable

First of, before doing any patch releases on stable the commits should already have been on dev for a number of days without reported issues (preferably more than a week). Also, before releasing it, we need more manual validation and sanity checking. Please send out a mail internally requesting for feedback and ask the people responsible for the patches going in to validate functionality.

We normally simply cherry pick the changes from master, and there is no difference in how we do this compared to dev channel, except, we don't increase PRERELEASE_PATCH, we increase PATCH in the tools/VERSION file.

Clone this wiki locally