This is a list of tips that should help you be more productive when working on Fuchsia.
Install the Gerrit Monitor Chrome extension to include the list of Gerrit changes that require your attention in your Chrome toolbar.
Review Gerrit settings and tweak them to your liking. For instance you may want to enable “Publish comments on push”, which will automatically send draft comments when a new patch set is published, rather than you having to do this manually from the web UI.
By default Git uses two-way diffs when presenting conflicts. It does not display what the original text was before the conflict, which makes it hard to solve some conflicts.
You can configure Git to show the original text by enabling three-way diffs:
git config --global merge.conflictstyle diff3
$FUCHSIA_DIR/scripts/git to your PATH to be able to use fuchsia-specific git commands such as
git fuchsia-review [<commit ref>], which opens the current or given commit in gerrit.
You are encouraged to add your own questions (and answers) here!
There are a wide variety of workflows used by the Fuchsia team. A daily workflow to get you started is as follows:
$ jiri update -gc # Start a new feature on a `myfeature` branch from the current stable commit $ git checkout -b myfeature JIRI_HEAD # Do work, making changes, etc. $ git commit # Upload your work to Gerrit for review $ jiri upload # OR $ git push origin HEAD:refs/for/main
Congratulations, you made your first Gerrit change!
Suppose you want to start new work on an
otherfeature branch while you wait for review of the work located on your
# Start a new independent line of work on the `otherfeature` branch # while waiting for review of `myfeature`: $ git checkout -b otherfeature JIRI_HEAD # OR # Start a derivative line of work while waiting for review: $ git checkout -b otherfeature
When you want to update your
myfeature branch but you've been working on an “independent” line of work on the
# Commit any present dirty work, then, switch to "myfeature": $ git checkout myfeature # Make any relevant edits to the code, then: $ git commit --amend # Now upload the new patchset to Gerrit: $ jiri upload # OR $ git push origin HEAD:refs/for/main
When you want to update your
myfeature branch because you got some review comments, and you are using a “derivative” line of work:
# Now you get a review comment that needs a change in "myfeature" # Commit your present work, if you aren't finished, maybe use a work-in-progress change: $ git commit -a -m "work in progress" # Start a rebase operation, so you can edit your first change: $ git rebase -i JIRI_HEAD # Replace "pick" with "edit" on the change you need to update and save and close the file # Make the relevant code changes, then: $ git add . && git rebase --continue # You may need to make additional "rebase" steps if your edits need integration # with later commits For each case, look at "git status" to see what files are # in conflict, and make the relevant adjustments. The rebase is complete when # git reports "Successfully rebased and updated ...." If you made a "work in # progress" change and want to unwind that commit: $ git reset HEAD # Now you can upload your modified changes to Gerrit: $ jiri upload # OR $ git push origin HEAD:refs/for/main
When you see “merge conflict” in Gerrit because your change can't cleanly be integrated with the
# Checkout the branch for the change you need to update (e.g. "myfeature"): $ git checkout myfeature # Update your git repository: $ git fetch # Update your branch: $ git rebase origin/main # Fixup and continue the rebase as necessary, until you see "Successfully rebased ..." # Then upload your newly updated code: $ jiri upload # OR $ git push origin HEAD:refs/for/main
When you've been working for more than a day, and you need to “sync your code” with upstream (you generally want to do this at least once per day):
# Commit any in-progress work, then # Checkout the stable upstream you last sync'd $ git checkout JIRI_HEAD # Update your local repository (this will include updates for prebuilts, third # party repositories, and so on): $ jiri update -gc # Now to switch back to, and update your working branch (e.g. "myfeature"): $ git checkout myfeature # Updating "myfeature" with the latest stable code: $ git rebase JIRI_HEAD # Perform fixups and "git rebase --continue" until you get to "Successfully rebased ..."
You can find more information on parts of workflows below. You can find more information on general git workflows in gitworkflows(7). You can find more information on git in general at git-scm.com/doc.
Update all projects simultaneously, and rebase your work branch on
$ jiri update -gc -rebase-untracked $ git checkout <my_branch> $ git rebase JIRI_HEAD
git rebase to
JIRI_HEAD should be done in each repo where you have ongoing work. It‘s not needed for repos you haven’t touched.
You'll need to upload a patch set to Gerrit to have it reviewed by others. We do this with
Gerrit uses an auto-generated metadata marker in the change's description to figure out which Gerrit review thread to upload a patch to, such as:
This is different from a git commit's SHA hash, and can be considered stable during review, as you make edits to your changes and commits. Use the same Change-Id for a given review (in case you are squashing multiple commits).
If you've made changes and want to upload a new patch set, then (assuming that this is the latest change in your branch; use
git log to find out) you can do something like:
$ git commit -a --amend # -a for all uncommitted files, --amend to amend latest commit $ jiri upload
Attempt a rebase:
$ git fetch origin && git rebase origin/main # Resolve conflicts as needed... $ jiri upload
But read below about how a
git rebase can negatively interact with
You can save all uncommitted changes aside, and re-apply them at a later time. This is often useful when you're starting out with Git.
$ git stash # uncommitted changes will go away # do stuff $ git stash pop # uncommitted changes will come back
A: They are not related.
jiri is a wrapper around git that provides support for managing more than one git repository in sync (the Fuchsia code base is composed of many git repositories), as well as synchronizing a set of prebuilt artifacts, such as those found in
fx is a convenience wrapper around many tools built in the Fuchsia tree, and helps with many daily workflow tasks, such as building, running tests, consuming logs, connecting to shells on devices, and many other operations.
A: Yes, unless jiri is configured to sync the rebased repository/petal to HEAD instead of the globally integrated version. This is not the case if you use the current/new default bootstrap setup, which tracks global integration for all repos, but may be the case if you set up your checkout in the past or used
fx set-petal X.
When working at petal X (accomplished with
fx set-petal X),
jiri update will rebase the local branches in repo X onto HEAD of origin/main. But other petals' repos will be synced to specific revisions that may be behind HEAD of their origin/main.
Fuchsia‘s continuous integration system (specifically rollers) makes a new revision of a petal available to other petals only after testing that the new revision doesn’t break other petals.
jiri update will always leave other petals synced to these successfully-tested revisions. But a git rebase to origin/main for a petal may advance that repo beyond the tested revision, which has the potential to introduce breaking changes. The result may be that you can build for a certain petal, but not for other petals (e.g., correctly build garnet, but not be able to build topaz).
If you have a particular commit that you want jiri to honor, download its
jiri.update file and feed it to
A: Can‘t, sorry. Try to arrange your changes so that they don’t break each petal during a transition (i.e., do a soft transition). But sometimes you will necessarily break things; aim to minimize the duration of breakage (i.e., a hard transition).
Example scenario: I have an interface defined in stem, and it is implemented in another petal. If I change the interface, am I doomed to break other petals?
Yes. But you can “babysit” the rollers so that the breakage range is minimized. The caveat with babysitting is that others may also be babysitting a breakage, and you may end up babysitting for longer than you had intended.
Alternatively, you could do something as follows:
lowerthat is a copy of the original interface.
lower-rollerto roll into
upper, or roll yourself by updating the file
upperto use the new clone interface that maintains the old contract.
lowersuch that the original interface’s contract is modified to the new, desired form.
lower-roller, or roll yourself.
upperto use the original interface name, now with its new contract. Make any changes required.
Note: this answer is subject to change/breakage shortly after authorship.
Let's assume you want to produce four builds:
First, one must build Zircon, as the Zircon build directory is shared across Fuchsia build targets. It doesn't matter at this stage which product/board combination you pick, we just need to start building Zircon.
# We start with bringup, because it's small, but it doesn't matter which you start with: $ fx --dir out.bringup.x64 set bringup.x64 $ fx --dir out/bringup.x64 build
Now you have Zircon built, you can start building several other builds concurrently:
$ fx --dir out/minimal.x64 set minimal.x64 $ fx --dir out/minimal.x64 build > minimal.x64.build.log & $ fx --dir out/core.vim3 set core.arm64 $ fx --dir out/core.vim3 build > core.vim3.build.log & $ fx --dir out/minimal.vim3 set minimal.arm64 $ fx --dir out/minimal.vim3 build > minimal.vim3.build.log &
You can reference each of these builds while running
fx tools by passing
--dir to your fx command, e.g. to run
fx serve using the vim3 minimal product, you would use:
$ fx --dir out/minimal.vim3 serve
You can also change which build directory is your current default by using
$ fx use out/core.vim3
A: You'll need to
jiri update against a jiri snapshot file, an XML file that captures the state of each repo tracked by jiri.
fx sync-from-stem will do this. It uses
jiri under the hood. However, instead of syncing fuchsia.git and dependencies to match the current integration repo, it instead finds the integration commit that matches currently checked out fuchsia.git, and syncs integration and dependencies to match that fuchsia.git commit.
Put another way, fuchsia.git will be untouched, and everything else is synced to match. This can be useful to bisect within fuchsia.git.
A: You'll want to run
fx setup-macos, which registers all the relevant Fuchsia tools with the MacOS Application Firewall.
See this section about hard and soft transitions.
A: The preferred method for updating a FIDL protocol is to use a soft transition. In order for a soft transition to work, you need to create an intermediate state that supports both the old and new versions of the protocol.
Use the following steps to execute a soft transition:
Modify the FIDL definition in the Stem repository to support both the old and new protocol elements. Before landing the change, trigger the global integration tryjobs to validate that step 2 will succeed.
Publish the Stem repository, either by waiting for the daily automatic publication or by manually publishing the repository.
Update all the clients to use the new protocol elements.
Publish all the clients.
Remove the old protocol elements from the FIDL definition in the Stem repository.
Publish the Stem repository, typically by waiting for the daily automatic publication.
A: Coordinating an atomic change across multiple Petals (or between the Stem repository and one or more Petals) requires performing a hard transition.
Use the following steps to execute a hard transition:
Prepare changes to all affected repositories. If all of these repositories are part of the Platform Source Tree:
Notify the team stating your intention to execute a hard transition.
Land all the changes in the affected repositories. This step will break local integration in these repositories but will not break global integration because the changes have not been published yet.
Land a change in the global integration repository that references the new versions of the affected repositories. This change will publish the new version of all the affected repositories and should not break global integration. This change should unbreak local integration in the affected repositories.
A: To bisect history, perform the following steps:
Bisect the history in the configuration repository, which contains the revision history of global integration, before and after the observable change. The result of this bisect will be a single change to configuration repository, presumably that includes the publication of one or more repositories or prebuilt packages.
If the change to the configuration repository is a publication of a single repository, bisect the history of that repository before and after the publication of global integration. The result of this bisect should be the revision at which the behavior changed.
If the change to the configuration repository is a publication of prebuilt packages, switch to the source tree from which the prebuilt packages were created. Consult the documentation for that repository regarding how to bisect changes in that repository.
If the change to the configuration repository is a publication of multiple repositories, bisecting history becomes complicated because the two repositories have likely been changed in concert and you will need to traverse their history in concert. Consider studying the history of the repositories to understand why they were published together.