the Chromium logo

The Chromium Projects

New & Upgrade Package Process

Purpose

This page documents the process of upgrading a package in our source to the latest upstream version, or for pulling in a new package that doesn't yet exist (since that's the same thing as upgrading from version <none>). These instructions are intended for ChromiumOS developers.

The upgrade process documented here uses a script, called cros_portage_upgrade, developed to help automate/simplify this process. It is not required that you use this script, and if you are confident that you can update packages in another way you are free to do so. Some of the FAQ on this page may still be of use to you, in any case.

Note that this document mostly applies to packages in the portage-stable overlay (more on this later). The workflow it describes can be partially applied to start an uprev for packages in chromiumos-overlay but not at all to cros-workon-able packages.

Quickstart

A quick reminder of the steps to perform in your chroot set up per ChromiumOS developer guide:

# Enter chroot, using full path if necessary:
cros_sdk

# Set up boards to test on:
setup_board --board=<YOUR-BOARD>

# Start a new git branch in portage-stable repo:
cd /mnt/host/source/src/third_party/portage-stable
repo start <YOUR-NEW-BRANCH>

# Upgrade package locally:
cros_portage_upgrade --upgrade --board=<board>:<board>... some_package1 some_package2 ...
  1. Repeat previous step as needed, fixing any upgrade errors that appear.
  2. Test. Yes. Really. You.
  3. Proceed to follow the contributor's guide.

Assumptions

The following assumptions are made with these instructions:

  1. You know which package or packages you want to upgrade.
  2. You know how you intend to specifically test the package(s) after the upgrade (beyond the basic smoke tests).
  3. You are familiar with the build environment from a developer perspective.
  4. You know how to start branches, amend commits, upload commits, end branches, etc. See the ChromiumOS developer guide.

Brief Background

Any packages, upstream included, can be marked as stable or unstable for any architecture. You don't have to know anything more about this detail unless you need to upgrade a package to an unstable version. However, this happens somewhat frequently. Perhaps a feature you need is only available in the absolute latest version of a package, but that version has not been marked stable for x86, yet. If you do need to upgrade to an unstable version, check the appendix entry below.

The ChromeOS source has projects that are Gentoo "overlays", all under src/third_party. Three of them are of interest here.

  1. The src/third_party/portage-stable overlay has copies of all upstream packages that we use unaltered. When we "upgrade" a package, what we are really doing is putting a copy of the current upstream package into this overlay.
  2. The src/third_party/chromiumos-overlay is probably familiar to you. For package upgrade purposes, it will only be used in the following scenario: If a package is locally patched before the upgrade, then you will probably have to port the local patch to the newer version as well. That should be done in this overlay.

The Process

Plan your upgrade

Plan around packages

Figure out which package or packages you want to upgrade, using their full "category/package" name (e.g. sys-apps/dbus). Decide whether you want to upgrade them to the latest stable or unstable versions. Generally, your first choice should be to upgrade packages to the latest stable upstream version.

Plan around patches and overlays

Note whether any of the packages you need to upgrade are locally patched. This means that somebody, possibly you, has made a patch to the package that we are using right now. It is important to determine what should happen to that patch after the upgrade. Inspect the git history for the patch, talk to the committer, inspect the patch, etc.

Plan around boards

Decide which boards you want to try the upgrade for. You should include one board from each supported arch type (e.g. amd64, arm, arm64), unless the package is only used on one arch for some reason. Typically, there is no reason to specify more than one board per architecture type, unless you know of board-specific testing that must be done. If your package or packages are used on the host, then you want to upgrade for the host (which can be done at the same time as the boards). Many packages are used on a ChromeOS board as well as the host (the "host" is the chroot environment). Choose boards that you can test the upgrades on, or boards that you know the packages need to be tested on.

Prepare Environment

You probably want to repo sync your entire source tree.

cd ~/chromiumos && repo sync -j 4

Prepare a new branch in the src/third_party/portage-stable project. This is where the upgrades will be staged/committed.

cd ~/chromiumos/src/third_party/portage-stable && repo start pkg_upgrade .

Now, you need to setup all boards that you will need. Note that this is not required if you're upgrading a host-only package.

cros_sdk --enter
setup_board --board=amd64-generic
setup_board --board=cherry

Upgrade Locally

For these instructions, let's say you want to upgrade media-libs/alsa-lib, media-sound/alsa-headers, and media-sound/alsa-utils on boards amd64-generic and cherry. You plan to try the stable upstream versions first. These commands must be run from within your chroot. If you also need to upgrade on the host (chroot/amd64), use the --host option, which can be used in combination with the --board option or without it.

cros_portage_upgrade --upgrade --board=amd64-generic:cherry media-libs/alsa-lib \
    media-sound/alsa-headers media-sound/alsa-utils

If cros_portage_upgrade succeeded the first time, then you have no missing dependencies or other build-related obstacles to foul the upgrade. But you still need to test the upgrade results! The upgrade should be prepared as a commit in src/third_party/portage-stable. You should inspect/edit the commit message before uploading it, though.

If cros_portage_upgrade didn't succeed in one go, then it should give you the means to determine why. Typically, this means that one or more of the upgraded packages could not be built (using emerge) after the upgrade. This may be due to a package dependency that must also be upgraded or a package mask that must be edited. In any case, you are given the output of the failing emerge command so that you can determine for yourself what is needed. Make the necessary change (perhaps adding another package to upgrade at the command line), then run again.

To upgrade to unstable versions, simply add the --unstable-ok option and follow the instructions. Or see the guidelines for upgrading to unstable versions below.

Run cros_portage_upgrade --help to see addition options and usage. If you are having difficulty with this step, please do not hesitate to contact the ChromeOS development mailing list for assistance.

Clean up Commit

You may also need to clean up the files in your commit. Our convention is to only check in files that are used, and some upgrades come with extra files (such as unused patch files, for example). You can usually tell by inspecting the list of files and the ebuild whether any files are unused. You can remove these files from your commit with a command like the following within portage-stable:

git reset HEAD~ <filepath>
rm <filepath>
git commit --amend

If you already built your upgraded package, make sure to do it again after cleaning up extraneous files to verify that you haven't removed something important!

Testing

Use common sense, and your specific expertise, to test the upgraded packages. The changes are active in your source right now, so you can build your target boards to test. If your package is added to the host (chroot), you'll need to emerge it onto your chroot and test it there as well.

You probably want to run at least the suite:smoke tests for each board, which you can do by following the tips here. In particular, you can use trybot to determine what effect your upgrade will have on the greenness of the waterfall.

One common cause of failure is that upstream has introduced new default USE flags. For example, if the upstream ebuild was set to -foo before (which defaults foo to off) and now is set to +foo, a new dependency or a new runtime behavior might have been introduced by the change. You would need to evaluate the change and determine whether the new USE flag default is one that we should adapt to. If, on the other hand, the flag adds a dependency on something that we don't use in ChromeOS, you might overlay the USE flag default in the chromiumos-overlay repo's profiles/targets/chromeos/package.use in a separate commit, first, before submitting the portage-stable upgrade. See below for an explanation of how chromiumos-overlay functions but note that you do not need to copy the upstream package into chromiumos-overlay to override the default USE flags.

Beyond the suite:smoke tests, test whatever you think makes sense given the packages you just upgraded.

Uploading

When you are satisfied that the upgrade is safe to push, you still need to edit the commit message that was created for you by cros_portage_upgrade. Add a bug number, test details, and any other details you want.

cd ~/chromiumos/src/third_party/portage-stable
git commit --amend

Upload the usual way:

cd ~/chromiumos/src/third_party/portage-stable
repo upload .

Then go through the usual code review process on Gerrit per the contributor's guide.

Clean up Environment

After your changelist passes review and submitted to the tree, you will want to get off the branch you created in src/third_party/portage-stable. You know the drill.

cd ~/chromiumos/src/third_party/portage-stable
repo abandon pkg_upgrade .

Appendices

Re-applying a patch after upgrade

It is not uncommon for a package to be locally patched in the source now (typically any upstream package in the src/third_party/chromiumos-overlay overlay). To upgrade that package, the patch will (probably) need to be applied again to the upgraded version. Let's say you upgraded the foo/bar package to upstream version 1.2.3, but you need to re-apply a patch to it that was previously applied to version 1.1.1 (perhaps 1.1.1-r1 is the active version now with the patch).

Perform pristine upgrade

Allow the cros_portage_upgrade script to complete the pristine upgrade in src/third_party/portage-stable for the package first. Then copy the entire directory over to the analogous location under src/third_party/chromiumos-overlay, creating the directory if necessary.

Mask your pristine upgrade

You will commit the unaltered package as it is in the first changelist, but first you must mark it as unstable so that it will not be used. To do so, add a new file like the following to the src/third_party/chromiumos-overlay/profiles/default/linux/package.mask/ directory:

# Copyright <copyright_year> The ChromiumOS Authors
# Use of this source code is governed by a BSD-style license that can be
# found in the LICENSE file

# TODO(username): Masked for testing!
=foo/bar-1.2.3

The above masks the specific package version foo/bar-1.2.3 for all targets. (If your intent is to patch the ebuild only for certain platforms, please speak to the build team as this is generally unnecessary and strongly discouraged.) Now you are ready to submit the pristine copy of the upstream package to chromiumos-overlay.

Modify your pristine upgrade

Make a copy of the ebuild file with a revision suffix. For example:

cp bar-1.2.3.ebuild bar-1.2.3-r1.ebuild

Then re-apply the patch to the new revision ebuild, by copying the epatch line or lines in the previously patched ebuild (bar-1.1.1-r1.ebuild). Be sure to keep whatever patch files under the files directory are required for the patch, too. You might have to rename them (e.g. cp files/bar-1.1.1-something.patch files/bar-1.2.3-something.patch). Be careful to apply the patch responsibly, because the source file(s) the patch alters may have changed since the patch was created. If so, you can create a new patch or consult the original patch author (if not you).

To inspect the source files involved, both before and after patching, use the ebuild utility. For example, to see the source files for foo/bar-1.2.3 before any patching run this inside chroot, you might run one of the following:

ebuild `equery which foo/bar-1.2.3` unpack
ebuild-amd64-generic `equery-amd64-generic which foo/bar-1.2.3` unpack
ebuild-cherry `equery-cherry which foo/bar-1.2.3` unpack

The above will unpack source files and tell you where they are. You can inspect them to verify whether the source file(s) the patch touches have changed between versions. To see what the source files look like after patching, use the same commands as above but replace unpack with prepare. Also, make sure your newly patched version is the version being picked up by the build system by running equery which inside chroot:

equery which foo/bar
equery-amd64-generic which foo/bar
equery-cherry which foo/bar

Make sure your new ebuild is the one being picked up. Your upgrade with patch is now ready for testing. Return to the testing step of the process above.

Remember to delete the new entry you made to the package.mask file.

Don't forget to discard the upgrade in portage-stable that cros_portage_upgrade created for you.

Handling eclasses

Some packages make use of eclasses, which are essentially ebuild code factored out into library files shared across multiple packages. If an ebuild file includes an inherit statement, then it uses an eclass. You don't have to check for this before you upgrade. The upgrade script will attempt to detect when an eclass must be upgraded along with a package upgrade, but if it is unable to, then you must upgrade the eclass yourself.

Upgrading to unstable version

If you need to upgrade a package to an unstable version for a compelling reason, then use the --unstable-ok option to the cros_portage_upgrade script and it will try to do the right thing. What it does is edit the KEYWORDS line (or lines) in the ebuild to mark all architectures as stable. This is the only exception to the rule that packages in portage-stable should be unedited. You should inspect the resulting ebuild file or files to confirm that the edit worked correctly, and file a bug on cros_portage_upgrade if it did not.

FAQ

I ran the upgrade script multiple times, and the runtime dropped drastically after the first time. What gives?

The cros_portage_upgrade script needs to have a local checkout of the upstream origin/gentoo source to use as a reference when it runs. By default, it creates a clone in a temporary directory and then re-uses it in later runs (updating it each time). This explains the runtime symptoms. Messages from the script should say something to this effect, as well.

Alternatively, you can clone your own copy of upstream origin/gentoo to point cros_portage_upgrade to, using the --upstream option. This will have the same runtime benefits if you run more than once. Do this inside your chroot:

cd
git clone https://chromium.googlesource.com/chromiumos/overlays/portage gentoo-portage
cd gentoo-portage/
git checkout origin/gentoo

Then add --upstream=$HOME/gentoo-portage . to your invocation of cros_portage_upgrade.

Why doesn't the upgrade script just run on all boards/architectures that matter at once? Why do I have to specify them?

The boards that "matter" vary. The same is true, to a lesser extent, with architectures. You know how you intend to test the package, and you may have plans to test it on specific boards for specific reasons. Plus, cros_portage_upgrade requires that setup_board be completed for all specified boards because it evaluates packages and their dependencies within the context of that board, making use of emerge-<board> and equery-<board> utilities heavily. If you already have a reasonable subset of boards setup, or available to test on, then running on those boards makes sense.

Why doesn't the upgrade script run setup_board for me as needed?

This script is intended to live within the build environment we have today. It lives between the setup_board and build_packages stages, which all developers are familiar with. The script is not intended to be a master script with knowledge of build process order.

Why doesn't the upgrade script create the branch in src/third_party/portage-stable for me?

This was considered, but discarded in favor of transparency. You need to know that a branch was created because you must amend the commit message, upload it, and especially abandon the branch in the end. If you created the branch in the first place you are more likely to be aware that you need to abandon the branch to return to your previous state.

Why can't the upgrade script create a complete commit message so no amend is needed, then do the upload and abandon steps as well?

  1. Only you know what tests you will run to verify the upgrade and those should be mentioned in the commit message.
  2. There are many scenarios where the upgrade result is not final and requires intervention/repetition by you.
  3. Only you know what bug to associate with this upgrade work.

How do you install a new package that is not currently installed? I get the "The following packages were not found in current overlays (but they do exist upstream): ..." error.

Make sure you pass --upgrade to the script since installing a new package is the same thing as upgrading it from <none>.

How do you manually upgrade an eclass?

Assuming that you have already run cros_portage_upgrade and let it fail, repeat the following procedure for each eclass in need of upgrade:

  1. In your chroot, look under /tmp/portage/eclass/ for the upstream copy.
  2. Copy said upstream version out of /tmp/portage/eclass/ into portage-stable/eclass/ in your source tree.
  3. Create a CL and pass it through the CQ.

If this procedure appears incomplete, then please reach out to the build team for additional guidance.

How do I deal with incompatible features in upstream ebuilds?

Upstream Gentoo sometimes moves faster than ChromiumOS. For example, they may bump to new Python versions, EAPI versions, etc., that we don't yet support. This may cause cros_portage_upgrade to skip those ebuilds, and eventually bail with an error, like:

Unable to find "foo/bar" upstream on <arch>.

You may be able to fix these issues locally, and try again. For example:

# Modify the local portage cache, downgrading from EAPI 8 to 7.
sed -i '/EAPI=/s:8:7:' /tmp/portage/foo/bar/bar*.ebuild
# Try again without upgrading the portage cache.
cros_portage_upgrade --local-only [args]