OpenOCD
Release Processes

This page provides an introduction to the OpenOCD Release Processes:

Why Produce Releases?

The OpenOCD maintainers produce releases periodically for many reasons. This section provides the key reasons for making releases on a regular basis and why a set of release processes should be used to produce them.

At any time, source archives can be produced by running make dist in the OpenOCD project tree. With the 0.2.0 release, this command will package the tree into several popular archive formats: openocd-<version>.{tar.gz,tar.bz2,zip}. If produced properly, these files are suitable for release to the public.

When properly versioned and released for users, these archives present several important advantages compared to using the source repository (including snapshots downloaded from that repository using gitweb):

  1. They allow others to package and distribute the code using consistent version labels. Users won't normally need to care whose package they use, just the version of OpenOCD.
  2. They contain a working configure script and makefiles, which were produced as part of creating the archive.
  3. Because they have been formally released by the project, users don't need to try a random work-in-process revision. Releasing involves spending some time specifically on quality improvments, including bugfixing source code and documentation.
  4. They provide developers with the flexibility needed to address larger issues, which sometimes involves temporary breakage.

Hopefully, this shows several good reasons to produce regular releases, but the release processes were developed with some additional design goals in mind. Specifically, the releases processes should have the following properties:

  1. Produce successive sets of archives cleanly and consistently.
  2. Implementable as a script that automates the critical steps.
  3. Prevent human operators from producing broken packages, when possible.
  4. Allow scheduling and automation of building and publishing milestones.

The current release processes are documented in the following sections. They attempt to meet these design goals, but improvements may still need to be made.

Version Labels

Users can display the OpenOCD version string in at least two ways. The command line openocd -v invocation displays it; as does the Tcl version command.

Labels for released versions look like 0.3.0, or 0.3.0-rc1 for a preliminary release. Non-released (developer) versions look like 0.3.0-dev, or 0.3.0-rc1-dev. In all cases, additional tags may be appended to those base release version labels.

The tools/release/version.sh script is used to manipulate version IDs found in the source tree.

Release Versions and Tags

The OpenOCD version string is composed of three numeric components separated by two decimal points: x.y.z, where x is the major version number, y is the minor number, and z is the micro. For any bug-fix release, the micro version number will be non-zero (z > 0). For a minor release, the micro version number will be zero (z = 0). For a major releases, the minor version will also be zero (y = 0, z = 0).

After these required numeric components, release version strings may contain tags such as as -rc1 or -rc2. These 'rc' tags indicate "release candidate" versions of the package. Like major/minor/micro numbers, these are updated as part of the release process.

The release process includes version number manipulations to the tree being released, ensuring that all numbers are incremented (or rolled over) at the right time and in the proper locations of the repository. One of those manipulations creates a repository tag matching that release's version label.

Packager Versions

Distributors of patched versions of OpenOCD are encouraged to extend the version string with a unique version tag when producing external releases, as this helps to identify your particular distribution series. Knowing that a release has such patches can be essential to tracking down and fixing bugs.

Packager version tags should always be suffixes to the version code from the OpenOCD project, signifying modifications to the original code base. Each packager release should have a unique version.

For example, the following command will add a 'foo' tag to the configure.ac script of a local copy of the source tree, giving a version label like 0.3.0-foo:

tools/release/version.sh tag add foo

This command will modify the configure.ac script in your working copy only. After running the bootstrap sequence, the tree can be patched and used to produce your own derived versions. You might check that change into a private branch of your git tree, along with the other patches you are providing.

You can also "bump" those tags (so "foo1" becomes "foo2" etc) each time a derived package is released, incrementing the tag's version to facilitate tracking the changes you have distributed.

tools/release/version.sh bump tag foo

Of course, any patches in your branches must be provided to your customers, and be in conformance with the GPL. In most cases you should also work to merge your improvements to the mainline tree.

Development Versions and Tags

Everything except formal releases should have the tag -dev in their version number. This helps developers identify reports created from non-release versions, and it can be detected and manipulated by the release script. Specifically, this tag will be removed and re-added during the release process; it should never be manipulated by developers in submitted patches.

Versions built from developer trees may have additional tags. Trees built from git snapshots have snapshot tags. When built from a "live" git tree, tags specify specific git revisions:

0.3.0-rc1-dev-00015-gf37c9b8-dirty

indicates a development tree based on git revison f37c9b8 (a truncated version of a SHA1 hash) with some non-git patches applied (the dirty tag). This information can be useful when tracking down bugs. (Note that at this writing, the tags do not directly correspond to git describe output. The hash ID can be used with git show, but the relevant repository tag isn't 0.3.0-rc1-dev; this might change in the future.)

Release Manager

OpenOCD archive releases will be produced by an individual filling the role of Release Manager, hereafter abbreviated as RM. This individual determines the schedule and executes the release processes for the community.

RM Authority

Each release requires one individual to fulfill the RM role; however, graceful transitions of this authority may take place at any time. The current RM may transfer their authority to another contributor in a post to the OpenOCD development mailing list. Such delegation of authority must be approved by the individual that will receive it and the community of maintainers. Initial arrangements with the new RM should be made off-list, as not every contributor wants these responsibilities.

RM Responsibilities

In addition to the actual process of producing the releases, the RM is responsible for keeping the community informed of all progress through the release cycle(s) being managed. The RM is responsible for managing the changes to the package version, though the release tools should manage the tasks of adding or removing any required development branch tags and incrementing the version.

These responsibilities matter most towards the end of the release cycle, when the RM creates the first RC and all contributors enter a quality-improvement mode. The RM works with other contributors to make sure everyone knows what kinds of fixes should merge, the status of major issues, and the release timetable.

In particular, the RM has the final decision on whether a given bug should block the release.

Release Schedule

The OpenOCD release process must be carried out on a periodic basis, so the project can realize the benefits presented in answer to the question, Why Produce Releases?.

Starting with the 0.2.0 release, the OpenOCD project expects to produce new releases every few months. Bug fix releases could be provided more frequently. These release schedule goals may be adjusted in the future, after the project maintainers and distributors receive feedback and experience.

More importantly, the statements made in this section do not create an obligation by any member of the OpenOCD community to produce new releases on regular schedule, now or in the future.

Sample Schedule

The RM must pro-actively communicate with the community from the beginning of the development cycle through the delivery of the new release. This section presents guidelines for scheduling key points where the community must be informed of changing conditions.

If Tn is the time of release n, then the following schedule might describe some key T0-to-T1 release cycle milestones.

  • T0 ... End of T0 release cycle. T1 cycle starts, with merge window opening. Developers begin to merge queued work.
  • ... several weeks of merge window ...
  • RC1 ... Close mainline to new work. Produce RC1 release, begin testing phase; developers are in "bugfix mode", all other work is queued; send out planned endgame schedule.
  • RC2 ... Produce RC2 and send schedule update to mailing list, listing priorities for remaining fixes
  • ... more RC milestones, until ready ...
  • T1: End of T1 release cycle. T2 cycle starts, with merge window opening. Developers begin to merge queued work.

Note that until it happens, any date for T1 is just a goal. Critical bugs prevent releases from happening. We are just beginning to use this window-plus-RCs process, so the lengths of the merge windows versus the RC phase is subject to change. Most projects have RC phases of a month or more.

Some additional supplemental communication will be desirable. The above list omits the step-by-step instructions to daily release management. Individuals performing release management need to have the ability to interact proactively with the community as a whole, anticipating when such interaction will be required and giving ample notification.

The next section explains why the OpenOCD project allows significant flexibility in the part of the development that precedes the release process.

Schedule Flexibility

The Release Manager should attempt to follow the guidelines in this document, but the process of scheduling each release milestone should be community driven at the start. Features that don't complete before the merge window closes can be held (perhaps in some branch) until the next merge window opens, rather than delaying the release cycle.

The Release Manager cannot schedule the work that will be done on the project, when it will be submitted, reviewed, and deemed suitable to be committed. That is, the RM cannot act as a priest in a cathedral; OpenOCD uses the bazaar development model. The release schedule must adapt continuously in response to changes in the rate of work. Fewer releases may be required if developers contribute less patches, and more releases may be desirable if the project continues to grow and experience high rates of community contribution. During each cycle, the RM should be tracking the situation and gathering feedback from the community.

Release Process: Step-by-Step

The release process is not final; it may need more iterations to work out bugs. While there are release scripts, key steps require community support; the Release Manager isn't the only participant.

The following steps should be followed to produce each release:

  1. Produce final patches using a local clone of mainline. Nobody except the RM should be committing anything. Everyone with commit privileges needs to know and agree to this in advance! Even the RM only commits a handful of updates as part of the release process itself ... to files which are part of the version identification scheme or release process; and to create the version tag; and then to open the merge window for the next release cycle.
    1. Finalize the NEWS file to describe the changes in the release
      • This file is used to automatically post "blurbs" about the project.
      • This material should have been produced during the development cycle, by adding items for each NEWS-worthy contribution, when committed during the merge window. (One part of closing the merge window, by opening the RC phase of the release, is the commitment to hold all further such contributions until the next merge window opens.)
      • The RM should make sure nothing important was omitted, as part of the RC1 cycle. From then on, no more updates to NEWS content should be needed (except to seed the process for the next release, or maybe if a significant and longstanding bug is fixed late in the RC phase).
    2. Bump library version if our API changed (not yet required)
    3. Update and commit the final package version in configure.ac: (The tools/release/version.sh script might help ensure the versions are named properly.):
      1. Remove -dev tag.
      2. Update any -rc tag:
        • If producing the final release from an -rc series, remove it
        • If producing the first RC in a series, add rc1
        • If producing the next RC in a series, bump the rc number
      3. Commit that version change, with a good descriptive comment.
    4. Create a git tag for the final commit, with a tag name matching the version string in configure.ac (including -rcN where relevant):
      PACKAGE_VERSION="x.y.z"
      PACKAGE_TAG="v${PACKAGE_VERSION}"
      git tag -m "The openocd-${PACKAGE_VERSION} release." "${PACKAGE_TAG}"
      
    5. Do not push those changes to mainline yet; only builds using the source archives you will be creating should ever be labeled as official releases (with no "-dev" suffix). Since mainline is a development tree, these will be pushed later, as part of opening the merge window for the next release cycle (restoring the "-dev" suffix for that next release.) Those version and tag updates are the last ones to be included in the release being made.
  2. Produce the release files, using the local clone of the source tree which holds the release's tag and updated version in configure.ac ... this is used only to produce the release, and all files should already be properly checked out.
    1. Run tools/release.sh package to produce the source archives. This automatically bootstraps and configures the process.
    2. Run tools/release.sh stage to create an archives directory with the release data, including MD5 and SHA1 checksum files.
    3. Sanity check at least one of those archives, by extracting and configuring its contents, using them to build a copy of OpenOCD, and verifying that the result prints the correct release version in its startup banner. (For example, "configure --enable-parport" then "make" and run "src/openocd -v" as a sanity check.)
    4. Run make docs to create the documentation which will be published.
  3. Upload packages and post announcements of their availability:
    1. Release packages into files section of project sites:
      • SF.net:
        1. Under "Project Admin", use the "File Manager"
        2. Create a new folder under "openocd" named "${PACKAGE_VERSION}"
        3. Upload the NEWS file and mark it as the release notes.
        4. Upload the three source archive files, using the Web interface, into that folder. Verify the upload worked OK by checking the MD5 and SHA1 checksums computed by SourceForge against the versions created as part of staging the release.
        5. Also upload doc/openocd.pdf (the User's Guide) so the version matching each release will be easily available.
        6. Select each file in the release, and use the property panel to set its type and select the right release notes.
          • .tar.bz2: Linux, Mac
          • .tar.gz: BSD, Solaris, Others
          • .zip: Windows
          • For openocd.pdf just associate it with the right release notes.
        7. Create an SF.net project news update.
    2. Depending on how paranoid you're feeling today, verify the images by downloading them from the websites and making sure there are no differences between the downloaded copies and your originals.
    3. Publish User's and Developer's Guides to the project web sites:
      1. Use SCP to update the SF.net web site with PDF and HTML for the User's Guide, and HTML for the developer's guide ... you can instantiate a shell.sourceforge.net instance and set up symlinks from your home directory, to simplify this process.
    4. Post announcement e-mail to the openocd-development list.
    5. optionally:
      1. Post an update on the OpenOCD blog.
      2. Announce updates on freshmeat.net and other trackers.
      3. Submit updates to news feeds (e.g. Digg, Reddit, etc.).
  4. Resume normal development on mainline, by opening the merge window for the next major or minor release cycle. (You might want to do this before all the release bits are fully published.)
    • Update the version label in the configure.ac file:
      • Restore -dev version tag.
      • For a new minor release cycle, increment the release's minor number
      • For a new major release cycle, increment the release's major number and zero its minor number
    • Archive NEWS file as "<code>doc/news/NEWS-${PACKAGE_VERSION}</code>".
    • Create a new NEWS file for the next release
    • Commit those changes.
    • Push all the updates to mainline.
      • Last updates for the release, including the release tag (you will need to "git push --tags").
      • Updates opening the merge window
    • At this point, it's OK for commiters to start pushing changes which have been held off until the next release. (Any bugfixes to this release will be against a bug-fix release branch starting from the commit you tagged as this release, not mainline.)
    • Announce to the openocd-development list. Ideally, you will also be able to say who is managing the next release cycle.

To start a bug-fix release branch:

  1. Create a new branch, starting from a major or minor release tag
  2. Restore -dev version tag.
  3. Bump micro version number in configure.ac
  4. Backport bugfix patches from mainline into that branch. (Always be sure mainline has the fix first, so it's hard to just lose a bugfix.)
  5. Commit and push those patches.
  6. When desired, release as above ... except note that the next release of a bugfix branch is never a new major or minor release

Release Script Commands

The release.sh script automates some of the steps involved in making releases, simplifying the Release Manager's work.

The release script can be used for two tasks:

  • Creating releases and starting a new release cycle:
    git checkout master
    tools/release.sh --type=minor --final --start-rc release
  • Creating a development branch from a tagged release:
    git checkout 'v0.2.0'
    tools/release.sh --type=micro branch

Both of these variations make automatic commits and tags in your repository, so you should be sure to run it on a cloned copy before proceding with a live release.

Release Script Options

The release.sh script recognizes some command-line options that affect its behavior:

  • The –start-rc indicates that the new development release cycle should start with -rc0. Without this, the -rc tag will be omitted, leading to non-monotonic versioning of the in-tree version numbers.
  • The –final indicates that the release should drop the -rc tag, to going from x.y.z-rcN-dev to x.y.z.

Release Script Environment

The release.sh script recognizes some environment variables which affect its behavior:

  • CONFIG_OPTS : Passed as options to the configure script.
  • MAKE_OPTS : Passed as options to the 'make' processes.

Release Tutorials

This section should contain a brief tutorial for using the Release Script to perform release tasks, but the new script needs to be used for 0.3.0.

Release Script Shortcomings

Improved automated packaging and distribution of OpenOCD requires more patching of the configure script. The final release script should be able to manage most steps of the processes. The steps requiring user input could be guided by an "assistant" that walks the Release Manager through the process from beginning to end, performing basic sanity checks on their various inputs (e.g. the NEWS blurb).