Introduction

Welcome to the OpenJDK Developers’ Guide!

The OpenJDK Community is the place to collaborate on open-source implementations of the Java Platform, Standard Edition, and related projects. It was created in November 2006, when initial portions of the JDK source code were published under the GPLv2 license.

In order to work together efficiently, clear directions are sometimes needed to avoid misconceptions and to align developers’ views of terminology and process. The OpenJDK Community is a fairly pragmatic place. “Do the right thing” is most often the right course of action. Still, if people do things in the same right way then everyone’s work becomes more transparent and easier for others to follow. For this reason most parts of the development process have standard flows that are the recommended ways to do things.

The goal of this guide is to answer questions that developers of the JDK might have around development process, tooling, standards, and so forth. The formal rules and processes are described in other documents, such as JEP 1 for the JDK Enhancement-Proposal & Roadmap Process, and JEP 3 for the JDK Release Process. This guide is meant to be a complement to such documents, with tutorials and examples for how to follow these rules and how to work together with the rest of the OpenJDK Community.

There are many common use cases that aren’t detailed in the formal process. This guide suggests how to work in such cases.

OpenJDK

OpenJDK consists of a number of Groups. Members of a group collaborate on an area of mutual interest. The right hand side bar on the OpenJDK website has a list of all groups in OpenJDK. If you’re interested in a specific area, this is where you would start your OpenJDK experience. Look at the group’s information and wiki pages, and see what projects they sponsor on the Census page. The Census shows the structure of the OpenJDK Community.

Projects are where the coding and much of the other work is done in OpenJDK. There are many different projects, some produce shippable artifacts, like the JDK Project, some produce tools to be used by developers of these artifacts, like the Code Tools Project or Project Skara, and some produce documentation, like the Developers’ Guide Project. Many projects designs and develops new features for the Java language or the JVM, but there are also less code centric projects like the Duke Project which collects images of the Java mascot, Duke.

Author, Committer, Reviewer

OpenJDK has a few different roles that determine who has the right to do what in the different projects. These roles are defined in the OpenJDK Bylaws. The roles are earned based on experience and knowledge within each project.

A Contributor can have different roles in different projects. When you’re new to a project you don’t yet have a formal role in that specific project, even though you might have earned roles in other OpenJDK projects or have been recognized as a Contributor or a Member of OpenJDK. By contributing high-quality content you’ll soon be eligible for OpenJDK roles in the project. First Author, then Committer, and finally Reviewer if you stay active and earn the trust of the community. Trust is an important part of earning these roles. There’s a rough guideline saying that to become a Committer you should have contributed 8 significant changes, and to become a Reviewer you should have contributed 32 significant changes. In reality it’s not as easy as “just” contributing code. You need to build a track record of good decisions and sound judgment and show that you know what differentiates a good change from a not so good one. It’s not only correctness of the code that matters, it’s also the appropriateness. In the end the trust you’ve earned is put to the test through a vote.

Note that when a new project is created an initial set of members can be brought in at different levels without a vote.

Becoming an Author

Becoming an Author is the first step. To achieve this you need to contribute two changes to the project in which you wish to become an Author. Once your changes are pushed into the code base and has been vetted enough to determine that the changes were indeed good changes you can go ahead and send an email to the project lead of that particular project and ask to be added as an Author. Note that “the project” is not OpenJDK, but rather the specific development project where you did your contributions (e.g. “JDK”, “JDK Updates”, “Amber”, etc). The OpenJDK Project description has a template for such an email. In short the email should contain your name, the project name, your email address, and GitHub links to your changes. In response to your email you will get a time-limited invite which you should fill out.

To see who the project lead is for your project, see the OpenJDK Census. The Census unfortunately doesn’t provide email addresses for people, but assuming you have been active on the project mailing list (since you are applying for Author after all), you should be able to find the lead’s email address in your local email archive, or ask your sponsor.

As an Author you will get your OpenJDK user name. Once you have gotten the user name, this should be associated with your GitHub account in order for the bots to be able to identify you on GitHub. See the Skara documentation for more details on that. Once that’s done you can create PRs and get them reviewed, but you will still need a sponsor to approve your change before it can be integrated. You’ll also get write access to JBS and the OpenJDK wiki related to the project in question, and to cr.openjdk.org via an SSH key provided at the time you accept your invitation.

The rules of any particular project may have different guidelines regarding requirements for Authorship at the discretion of the Lead.

Becoming a Committer

To become a Committer you should show that you intend to actively contribute to the project and that you can produce non-trivial changes that are accepted for inclusion into the project code base. The number eight has been seen as a formal lower limit on the number of changes, but since the changes must be non-trivial, or “significant” as the OpenJDK Project description says, and the definition of significant is subjective, the general recommendation is to wait with a Committer nomination until there’s at least 10-12 changes pushed to have some margin for different interpretations of “significant”. In practice though, we have seen several examples where the number of significant changes hasn’t been the dominating factor in a Committer vote. A contributor’s work in another OpenJDK project may also be relevant for the vote. What the vote should ultimately test is the committer’s commitment to the OpenJDK project for which the vote applies - is it believed that the person is dedicated and willing to spend time and effort on the project? Is the person believed to be a good citizen of the project? It’s always a good idea to seek the advice of a sponsor who can guide you through the process to becoming a Committer - you will need one to run the Committer vote anyway. They will probably also have a better idea of what constitutes a “significant” change.

Once you have the required changes, a Committer in the project can start a vote by sending an email proposing that you should become a Committer. The email should follow the template found in the OpenJDK Project description.

A Committer is allowed to push changes without the aid of a sponsor. A Committer is also allowed to nominate other non-Committers to become Committers in the project.

Becoming a Reviewer

To become a Reviewer you must show a track record of sound and correct judgment calls as mentioned above. Being a good Committer doesn’t necessarily make you a good Reviewer. As a Reviewer you have the power to approve changes for inclusion into the project source code. This means that a Reviewer needs to be able to judge the quality and appropriateness of any proposed change, not just the mechanics of the code.

The assumption is that after having produced 32 significant changes one should have become familiar with the process around reviews and the requirements around getting a change approved. This should really be seen as a minimum requirement though. A more practical consideration would be to look at whether the non-trivial commits of a potential Reviewer are accepted largely intact or whether they are always being refined by the review process. There may be cases where it will take significantly more than 32 changes for a Committer to be ready to become a Reviewer.

Once you are deemed ready, a Reviewer in the project can start a vote by sending an email proposing that you should become a Reviewer. The email should follow the template found in the OpenJDK Project description.

Non-trivial/Significant changes

One key definition when advancing through the OpenJDK roles is the significant change. What exactly does it take for a change to be significant?

Instead of describing the significant change (because that’s quite difficult to define) provided here is a few examples of changes that wouldn’t be considered significant or for other reasons wouldn’t count as significant contributions.

• Purely aesthetic changes like renaming or fixing indentation
• Repeated follow-up bugfixes from earlier changes
• Larger changes where only a non-significant portion of the work was done by the Contributor under vote
• Trivial backports of someone else’s changes

Contributing to an OpenJDK Project

Contributing to OpenJDK can take many forms. Writing code and providing patches is just one of them. A big part of developing a feature or a bugfix is testing and code review. Anything you can do to help out in these areas will be recognized as a contribution. Join the mailing lists to engage in design discussions and reviews, and download the latest EA builds or project repositories to try out new features and give feedback. If you see some misbehavior, or if you see somebody mention some misbehavior on some internet forum, try to track it down. Good bug reports with reproducible test cases are extremely valuable and make excellent contributions.

Anything you can do to spread the word about Java, new features, and your experiences using the JDK will be helpful for the community and to the OpenJDK developers. Trying out a new feature and reporting your experiences is also a contribution. Whether you find that the new feature improves your application, or if you find some area that needs to be improved, your feedback is valuable to the developers of that feature.

If you have a success story where Java solved your problem, or if you successfully upgraded to a more recent version of the JDK and noticed some improvements, spreading this story through a blog, news article, or some other channel is also a contribution.

If you’re in a position to choose what programming language to use in a project, in a tutorial, or in a class, you have the power to enlarge the Java community in a very direct way, and your colleagues or students will get an opportunity to learn one of the most used programming languages in the world.

Things to consider before proposing changes to OpenJDK code

Every change to JDK code carries a risk of changes in behavior which may adversely affect applications. Generally we’re looking to improve the functionality and capability and sometimes performance of the platform without that negative consequence. So we need to ask ourselves whether each change is worthwhile - and some may not be no matter how well intentioned.

One question to ask yourself is: What problem are you trying to solve?

The important thing here is to understand the problem itself, independent of any solution (and independently of the solution that currently happens to be in the PR). A number of other questions and lines of thought emanate from thinking about the problem. For example: is this the right problem to solve? Does solving this problem create worse problems elsewhere; that is, is there a net benefit? Does this problem need to be solved in the JDK, or can and should it be solved elsewhere (e.g., in tooling)?

The next question you need to answer before making any change is: What is the main intention of your change?

Depending on your answer to that question you will need to consider one or more of the following paragraphs.

• Correctness If your change improves program correctness, that’s important. And to broaden this, fixing of all kinds of bugs that really make things better for applications in ways they can detect is important.

• Robustness Updating code to use a newer platform API can be a good change. Moving away from APIs that are being deprecated or that are no longer maintained is likely desired. Do note though that supposedly equivalent APIs may not be the drop in replacement you think they are. You’ll need to prove that the code has the same behavior after the change through extensive testing.

• Security If you intend to increase the overall security of the platform, changes following secure coding practices and using appropriate platform APIs for that are usually welcome. The exception might be if it’s a potentially destabilizing change in code where there’s only a theoretical problem. Please note: If you think you found a real security bug that might compromise the platform you should follow the process here.

• Refactoring / Cleanup Making code easier to understand or reducing code size may be a good change in areas that are under active development. Stable code however isn’t a good candidate for refactoring regardless of what the code looks like. The OpenJDK has evolved over many years and some parts have been stable for decades. If there’s no immediate need to work on the code for other reasons, then what will a cleanup actually achieve? One specific area where refactoring and cleanups are explicitly discouraged is in third-party code.

• Performance Can you demonstrate a user perceptible change? If you can’t measure the change, or a user can’t notice the change, or the change only improves code that is used infrequently, then maybe it isn’t worth it. Do you have benchmarks to back up your claims? Do you understand the results? Performance testing is complex and often there are many factors unrelated to your change that affects the numbers. What’s the tradeoff? The performance improvements that just make everything better do exist, but they are extremely rare. Most often the code gets more complex, or you speed up one case but slow down another. Are you making the right tradeoff?

• Modernizing Changing code purely for the sake of using a new language feature isn’t usually a good change. Be a subject matter expert, not just a language expert. Writing code in “a better way” is not guaranteed to be safe. A change of behavior is always possible and unless you understand the code you are changing at more than the core language/API level, and have looked into appropriate tests and can speak to the risks, then you should first find a subject matter expert to discuss it with. Keep in mind that the OpenJDK code is developed by a large community. If a new language feature is introduced, all developers working in that code must learn this new feature and understand the implications of using it.

I have a patch, what do I do?

In many GitHub projects the standard way to propose a change is to create a pull request (PR) and discuss the patch in the PR. For OpenJDK projects the situation is somewhat different. The JDK is used for mission critical applications and by millions of developers, the bar to contributing changes is high. Please follow the steps outlined below to make sure your change passes above the bar before creating a PR.

1. Sign the OCA

Oracle is the steward of OpenJDK. In order to make your patch available for review you must first sign the Oracle Contributor Agreement (OCA). This agreement gives Oracle and you as a contributor joint copyright interests in the code. You will retain your copyright while also granting those rights to Oracle.

When you sign the OCA, please make sure that you specify your GitHub user name in the Username field of the OCA. If you try to create a PR before you have signed the OCA, or if you didn’t specify your GitHub user name, you’ll get instructions telling you to do so, and the PR won’t be published until this is done. OCA registration is a manual process. Please allow for up to several days to have your OCA application processed, even though it’s normally processed swiftly. An alphabetical list of all of the assigned OpenJDK usernames may be found on the OpenJDK people list.

2. Socialize your change

Once the OCA is signed, please restrain your urge to create a PR just a little while longer. In order to prepare the community for your patch, please socialize your idea on the relevant mailing lists. Almost all changes, and in particular any API changes, must go this route and have a broad agreement in place before there is any point in presenting code. To understand the criteria by which your patch is going to be judged, please read Why is My Change Rejected? below. In short, hidden constraints and assumptions, stability and quality, maintainability, compatibility, and conformance to specifications must be considered before your PR is ready to be submitted. If you don’t understand the constraints for acceptance, you might be surprised when your PR is rejected.

3. Find a sponsor

Socializing your change on the mailing lists also prevents the surprise that would otherwise make the community choke on their morning coffee when they see a huge patch in a new, unknown PR. As a new developer in the community you’ll need to make a few friends that agree with your change. There are many good reasons to make friends, but the one relevant here is that for your first changes you’ll need a sponsor to facilitate the integration of your work. The sponsor will perform any number of administrative tasks like JBS updates, additional testing, etc. It’s usual for a sponsor to also be a reviewer of a change and thus familiar with it, but it’s not a requirement.

4. Create a tracking issue in JBS

Many OpenJDK projects require a tracking issue to be filed in the JDK Bug System (JBS) before a change can be pushed. This is the case for instance for the JDK and the JDK Updates projects. In order to obtain write access to JBS you need to be an Author in an OpenJDK project (see Becoming an Author). For your first changes, ask your sponsor to help you create the issue or file the bug through the Bug Report Tool.

5. Get acquainted with local process

Even though we strive to unify how things are done within OpenJDK, different areas and projects in OpenJDK may have slight variations in how they work. Some of these differences are highlighted throughout this guide, some aren’t. If you’re new to an area, make sure you understand local differences before you proceed. Ask your sponsor who should be your main point of contact through your first developer experience in OpenJDK.

Why is my change rejected?

Java and the JDK are very popular products, and just about every Java developer out there has an idea or two for how to enhance something. And (obviously not referring to you) believe it or not, not every idea is a good idea. Even though many ideas are indeed good, we must be quite restrictive on what we actually include into the JDK. There are many reasons for this.

• Hidden constraints and assumptions. Many sections of code have constraints and assumptions that aren’t necessarily visible at first glance. This might preclude certain changes, even those that might seem obvious.

• Stability and quality. The JDK is used by millions of developers and as a widely deployed commercial product, it’s held to a high standard of quality. Changes should include tests where practical, and core tests should pass at all times. The value of the change should outweigh the risk of introducing a bug or performance regression.

• Maintainability. Any new feature or code change will need to be maintained in the JDK essentially forever, thus imposing a maintenance burden on future maintainers. The code might still be in use long after you and the people who reviewed it have moved on. New maintainers must be able to understand how to fix bugs in this code.

• Complexity. Each new feature interacts with all the existing features, which can result in geometric growth of the interactions among features if features are added unchecked. Sometimes we avoid adding a new feature, even if it seems like an obvious thing to add, if that feature would make it difficult to add a more important feature in the future.

• Adherence to specifications. Much of the JDK is governed by a series of specifications, in particular the Java Language Specification, the Java Virtual Machine Specification, and the Java API Specification (“javadocs”). All changes must be checked and tested carefully to ensure that they don’t violate these specifications.

• Javadoc comments are specifications. The Java API Specification is authored in the form of javadoc comments, so even apparently innocuous changes to comments can be quite significant. It’s not always easy to tell what comments are part of the specification and what parts are merely code comments. Briefly, documentation comments on public packages, classes, and class members of exported modules are specifications.

• Specification changes. It’s possible to change the API specifications, and this is done regularly. However, these changes require even more scrutiny than code changes. This extra review is handled by the CSR Process. Specifications are written in stylized, somewhat formal language, and they don’t simply describe what the code does. Writing specifications is a separate skill from coding.

• Compatibility. Changes should also adhere to high standards of binary, source, and behavioral compatibility. The compatibility impact of apparently innocuous changes is sometimes startling.

For reasons like these it’s quite possible that your change, even though it adds value to you, isn’t deemed to add enough value to the larger community.

Mailing Lists

The mailing lists are the key communications mechanism for all OpenJDK work. All participation in an OpenJDK project starts with joining the relevant mailing list. A subscriber to an OpenJDK mailing list is referred to as a Participant in the Bylaws. As a general recommendation we suggest to subscribe to announce, discuss, and the -dev lists covering your explicit area of interest. All OpenJDK mailing lists are found here:

mail.openjdk.org

The OpenJDK Community is a friendly place. To keep it that way it’s important to keep a professional tone in emails and be aware that the community is global. Many different people with different backgrounds collaborate in these lists. Even though English is the required language for all lists, many Participants speak other languages as their native language. A high tolerance for non-perfect English is expected from anyone joining these lists. You’re also strongly encouraged to use your real name on the mailing lists. This adds to the professional tone of your email. Postings from anonymized mailboxes risk being seen as spam. If you do work in OpenJDK on behalf of your employer, please also list this affiliation. If your GitHub username differs from your real name it’s also a good idea to include that to identify yourself and your actions on GitHub.

You must be a member of a list to be able to post to that list. Some lists are moderated to keep the content on topic. Each list has its own archive where you can browse older conversations on the list.

There are a few different types of lists. The list name has two parts to explain what the list is intended for, <name>-<suffix>. The name often refers to the project that owns the list or a specific area of interest that the list focuses on. The suffix is explained below. Not all projects or areas have all types of lists described here.

-dev

Technical discussions around the implementation of the project artifacts. This is also where code reviews happen.

-use

Technical discussions around the usage of the project artifacts.

-discuss

General discussions around the project. The special case discuss@openjdk.org is used for general discussions around OpenJDK. Discussions around new project proposals usually happen here.

-changes

Changeset notifications from the source code repositories maintained by the project.

-announce

General project announcements. These lists are tightly moderated and are expected to be low traffic. The special case announce@openjdk.org is used for announcements for OpenJDK.

-experts

Expert group discussions. The list is restricted; only members of the expert group can subscribe.

-observers

Open for anyone to subscribe to see what the experts are discussing and potentially to have some dialog with other non-experts. There is no guarantee that an expert is subscribed to the -observers list or will see any responses on that list.

-comments

Used by observers to directly provide feedback/comments to the experts (typically a lead will process the comments list and forward things on to the experts list).

Changing your email address

If you need to change your registered email address, or if you have any other problems with the mailing lists, please contact mailman@openjdk.org.

Code Conventions

JBS - JDK Bug System

JBS is a public issue tracker used by many OpenJDK projects and is open for anyone to read and search. To get write access you need to be registered in the OpenJDK Census by becoming, for instance, an Author in an OpenJDK Project.

Filing an issue

When a new failure is found, or an improvement identified, an issue should be filed to describe and track its resolution. Depending on your role in OpenJDK you can either use the Bug Report Tool or, if you are registered in the OpenJDK Census, you can report the issue directly in JBS.

When filing an issue, try to make the report as complete as possible in order to make it easier to triage, investigate and resolve the issue.

Types of issues

The most common issue types are:

A few things to keep in mind when filing an issue:

• Before filing, verify that there isn’t an open issue already filed.
  • Search JBS for things like the name of the failing test, assert messages, the name of the source code file where a crash occurred etc.
  • If you find a similar issue that was closed as Cannot Reproduce then it may be appropriate to re-open that one - if you don’t have direct access to JBS you can file a bug using the Bug Report Tool requesting that it be reopened.
• Make a reasonable attempt to narrow down which build or release the failure first appeared in.
• Set Affects Version/s to the earliest JDK version(s) where the failure was seen.
  • If the failure is found in an update train of the JDK (e.g. 11.0.x), if possible please see if it’s also present in mainline.
  • For enhancements the Affects Version should be left empty, unless it is only relevant to a specific release family.
• Add relevant Labels like intermittent, regression, noreg-self, tier1 etc.
  • For more information see the JBS Label Dictionary.
• Set the priority.
  • It’s not the reporter’s responsibility to set a correct priority, this will be done later in triage, but a qualified guess is always better than the default - in JBS the range of priorities go from P1 (High/Important) to P5 (Very Low/Not Important), with the default being P4.
  • If you have a sense that the issue is critical, or not critical at all, then adjusting the priority higher or lower makes sense, otherwise you can leave it as the default.
  • When setting the priority, consider things like how the bug impacts the product and our testing, how likely is it that the bug is triggered, how difficult is it to work around, and whether it’s a regression, since that may break existing applications. Regressions are often higher priority than long-standing bugs and may block a release if not fixed. An example of a P1 would be an issue that is blocking a build or a release, whereas a P5 would be a minor typo in a code comment.
• In the Description, always include (if possible):
  • error messages
  • assert messages
  • stack trace
  • command line information
  • relevant information from the logs
  • full name of any failing tests
• Avoid including in the description or comments:
  • personal information
  • passwords, logins, machine names
  • logs which may include sensitive data
  • large amount of text or data - large logfiles are better provided as attachments
• If the failure isn’t reproducible with an existing OpenJDK test, attach a reproducer if possible, having a test case will decrease the time required to resolve the issue.
• In general the CPU and/or OS fields should not be set. ONLY use them if you know that the issue is only relevant to a particular platform or set of platforms.
• Provide the output of java -version whenever possible - this version information is particularly critical for hangs, JVM bugs, and network issues.
• Always file separate bugs for different issues.
  • If two crashes look related, but not similar enough to be sure they are the same, it’s easier to later close a bug as a duplicate than it is to separate out one bug into two.

Things to keep in mind when requesting an improvement:

• Unlike reporting a problem, when it comes to improvements, what constitutes a reasonable request can take discussion, and in general it’s encouraged that the mailing list for the area is used to suggest an improvement before filing.
• Enhancements to The Java Language Specification and The JVM Specification are managed through the Java Community Process.

To find out which component to use for different bugs, consult the directory to area mapping.

Implementing a large change

When managing the work for a large change, especially when the work involves multiple engineers, it’s recommended that the work is distributed across one or more “implementation” issues which should be linked to the main enhancement with a “blocks” link along with any relevant CSRs. The enhancement shouldn’t be considered done until all the blocking elements are completed. The use of sub-tasks for enhancements is not recommended unless all the sub-tasks are relevant to the fix, if it were to be backported, for example JDK-8231641 or JDK-8171407.

Implementing a JEP

It’s recommended for JEPs that the implementation is spread across one or more Enhancements as described above.

Issue states

JBS only has a few states in which a bug or enhancement can be in:

JBS Issue Flow

Triaging an issue

For most JDK areas, triage is performed on a regular basis (at least weekly) by triage teams. Each triage team consists of contributors who are area experts for a specific area or feature. If you haven’t been selected to be part of a triage team for a specific area you shouldn’t be triaging bugs in that area.

When triaging an issue, first give it a general review.

1. If the issue is a duplicate, close it as such.
2. If the issue belongs to a different area (it was filed in libraries, but it’s an HotSpot issue), transfer it to the correct component/subcomponent making sure that the state remains New.
3. If the issue is incomplete, add a comment noting what is needed and close the bug as Resolved - Incomplete. This is the JBS way of saying “need more information”. If no more information is obtained within reasonable time, the issue should be closed (Closed - Incomplete).

Now that the issue is in the right component and has the basic information, the analysis continues to get a more detailed understanding of the issue, and what should be done:

1. Ensure the priority is correct.
   • An approach that has been used for getting a consistent view of priority is to consider three aspects of the issue: Impact of the issue; Likelihood of it occurring; and, whether there is a Workaround. The higher the impact and likelihood the higher the priority; then, having a workaround reduces that priority - but mostly where the impact and likelihood aren’t that severe.
2. Ensure the Affects Version/s field is correct (within reason).
   • This may involve reproducing the bug, if doing so is fast and easy.
   • In addition to the version where the bug was found, take special care to also investigate if the bug affects mainline.
   • The Affects Version/s isn’t meant to be an exhaustive list of releases the issue is relevant to - it should initially be set to the release the issue was reproduced or identified on, and by implication it will be relevant on all releases past that point (see the (Rel)-na label). If it’s later found to be applicable to an earlier release family then adding that earlier release is encouraged if the issue needs to be fixed in that release.
   • Do not add additional release values to Affects Version/s for the same release family. E.g. If there is the value 11.0.2, don’t add 11.0.5, 11.0.7 etc. Adding an additional value for a separate release family where it’s still reproducible, e.g. 12, isn’t necessary but ok, especially if there’s been a few releases since the latest version noted. E.g. Affects Version/s is 8 but its still relevant to the latest mainline release.
   • For enhancements the Affects Version should be empty unless you feel that it is only relevant to a particular release family, and should not go into a future mainline release.
3. Set the Fix Version/s.
   • A bug should be fixed first in the most recent version where it exists. If you don’t know what version the fix will go into set the Fix Version/s to tbd.
   • If the bug also exists in older versions it may require backporting.
     • The decision to backport to a release should be made inline with the guidelines of the lead for that release.
     • There are two options for creating backport issues to track the backport: one is to create it manually once it’s agreed that the bug should be backported; the second, is to let the bots create the backport issue once you push the fix to the repo (see Working with backports in JBS). In most cases, letting the bots create the backport issue is preferred.
   • For project internal changes intended to be pushed to a project repository rather than the JDK or JDK Updates repositories, the fix version should be set to internal, or if the project is large enough to have its own repo-* fix version, use that.
   • Only one fixversion should ever be set, if the issue is to be fixed in additional releases then a separate backport must be created (see Working with backports in JBS). There are exceptions to this rule for CSRs and Release Notes.
4. Make sure the bug has all the required labels – see JBS Label Dictionary.
   • Bugs where behavior has incorrectly changed from a previous build or release: regression
   • Changes that don’t affect product code, but are only against the regression test, or problem-listing: noreg-self
   • Changes that don’t affect product code, but are only against documentation: noreg-doc
   • Issues that seem to be trivial to fix: starter
   • Enhancements that are pure cleanups: cleanup
   • Project specific issues usually have their own labels as well

At this point move the issue into the Open state.

Sensitive information (e.g. hs_err.log)

It may sound obvious, but avoid placing sensitive information in bug reports. Things like usernames, IP addresses, host names, and exact version information of third party libraries etc. may be crucial to be able to debug in some cases, but could also help an attacker gain information about your system. JBS is a public database that anyone can search, so be mindful of what you place there. In particular when attaching log files like the hs_err.log you should make sure that you are comfortable sharing the details exposed in it. Sometimes it may be better to leave a comment saying that these details can be obtained on request.

If you file a bug through the Bug Report Tool there’s a specific field that should be used to place sensitive information like this. Information placed there will not be part of the public bug report.

Updating an issue while fixing

Once you are made, or you make yourself, the assignee of an issue you take on the responsibility of moving the issue through to resolution - providing the current status, and ultimately leaving a record for others in the future to understand what happened. There are no set rules for how you manage the bug while you are assigned to it, as it depends on the type and importance of an issue. A simple update to the doc needs little to be done, fix the problem and close the issue; an intricate timing issue or crash should be handled differently - documenting your progress in identifying the problem (e.g. JDK-8212207, JDK-6984895, JDK-8287982), this is especially helpful if you ultimately move the issue to a different area as you have found that the problem lies elsewhere, or is closed as Won’t Fix. Your updates then provide a resource to others to better understand what has been done or the code itself. See The Importance of Writing Stuff Down for a good explanation as to why it’s important.

Some additional fields should be filled out or updated as you get a better understanding of the issue:

• For a regression, if you identify the fix that caused it, add a relates-to link to that issue (and add a regression_(ID) label) and set the Introduced in Build or Introduced in Version field.
• The Description usually explains what went wrong and how the failure was found, then there’s some investigation and eventually the root cause is found. At this point the Summary should be updated to correctly describe the bug. The Description however should remain a description of how the failure was found.
• The Affects Version/s should be updated if you in your investigation finds that the issue is older than what is indicated by the current Affects Version/s.

Resolving or Closing an issue

Once the work on an issue has been completed the issues Status should be in a “completed” state. There are two “completed” states: Resolved and Closed. These are accompanied by a Resolution. Which combination of Status and Resolution you should use depends on how the issue is completed.

Closing issues as duplicates

If the same issue is described in another JBS issue then close one against the other as Closed/Duplicate. In general the newer issue is closed as a Duplicate of the older one, but where the newer issue has a clearer description, or more useful, up-to-date comments then doing it the other way round is ok as long as none of them has been Fixed already. If one of the issues has been Fixed the other one should be closed as a Duplicate of the Fixed issue. There may be other reasons to choose to close one or the other issue as the Duplicate. As always - use your best judgement to make the end result as good as possible.

Note: Any issue closed as a Duplicate must have a “Duplicates” link to the duplicating issue.

Closing incomplete issues

As mentioned above, issues that lack the information needed to investigate the problem are placed in status Resolved - Incomplete. Triage teams should monitor incomplete issues in their area and if needed ping the relevant person. If the required information hasn’t been obtained within reasonable time (3-4 weeks) the bug should be closed as Incomplete.

Verifying an issue

Once a bug is marked as fixed there is now the option of someone, other than the person that fixed it, of marking it as Verified to confirm that the issue is fixed after testing; marking it as Fix Failed if it didn’t solve the issue; or, Not Verified to indicate that it wasn’t explicitly tested. Note that the JBS UI doesn’t highlight when Fix Failed has been set, you need to look for the Verification field at the bottom of the left-hand column in the Details section.

Removing an issue

Removing a JBS issue is a rare extreme case that shouldn’t be part of the normal workflow. For this reason, removing issues is restricted to admins only. If you for some reason need to remove an issue, send an email to ops@openjdk.org. You need to provide the bug id and a well considered reason the issue should be removed.

Note that JBS issues are not removed just because something was a bad idea, or a reported issue turned out to be an embarrassing user mistake. Such issues are simply closed.

JBS labels

JBS labels are used to tag and group related issues. JBS labels are an open namespace, which means that anyone can create new labels at any time. In order to avoid confusion, however, it’s best to reuse existing labels where possible. Most areas have their commonly used labels to identify issues in their respective area. Make an effort to find and use these labels. This can be done by editing the Labels field of a bug and entering the first few characters of the label you want to add. JIRA will pop up an autocomplete window with existing labels that match that prefix. Then choose one of the existing labels. Using the autocomplete window is preferable to typing the whole label name (even if you’re a good typist) because it’s easy for minor spelling errors to creep in, which can inadvertently introduce multiple labels with spurious spelling variations.

JBS labels should not be used to write documentation - don’t try to write sentences using labels. Adding a number of random labels is unlikely to be helpful to anyone.

JBS label dictionary

This table contains some frequently used JBS labels and their meaning. Please help keeping this dictionary up to date by adding your favorite labels. This table doesn’t dictate how to use labels, but rather document how they are used. That said, obviously it will help everyone if we try to follow a common standard and use similar labels in the same way across all entities that use JBS.

Fixing a Bug

This is the list of steps which should be performed when fixing a small bug. Small bugs include typos in code or specification, algorithm improvements for correctness or performance, and code changes required to correctly implement the specification.

For the purposes of brevity this document will use the term “bug” to refer to both bugs and enhancements unless otherwise noted. Hence “fix for a bug” could also imply “implementation for an enhancement”.

1. Discuss the intended change

   Send an e-mail to the appropriate development mailing list for the Project that maintains the code. The e-mail should have a subject line of the form:

   6543210: My favorite bug

   where 6543210 is replaced with the actual bug id number or “[NEW BUG]” if the bug id isn’t known and My favorite bug is replaced with the bug’s summary. The message should describe the intended change, which components may be affected, and any other risks or concerns.

2. Does a bug id exist for the work?

   Yes

   Continue

   No

   Create a bug in an appropriate component/sub-component.

3. Set the bug status to “Open”

   This communicates intent to fix the bug to other members of the Project. It also sets the expectation for downstream teams such as SQE and JCK that the bug will be fixed in an upcoming integration.

4. Does the fix for the bug require a specification change, directly affect an external interface, or otherwise have a compatibility impact?

   Yes

   Submission of a CSR request is required. The CSR must be [approved] before the bug fix is pushed to a feature release or update release repository. The work may begin concurrently with the CSR review, but may need to be modified in response to CSR feedback.

   No

   Continue

5. Fix the bug

   Assuming that the development team approves of the intended approach, begin working on the code using the latest source available from the appropriate OpenJDK Project repository.

6. [Is it possible to write a test to detect the bug?]

   Yes

   For bugs, provide a jtreg regression test as part of the changeset. For enhancements that directly affect one or more exported interfaces, add an adequate set of jtreg unit tests as part of the changeset. By convention all regression and unit tests should contain a @bug tag referencing at least one bugid.

   An entirely new test (or tests) may not be required. For example, if the bug is an existing regression test failure, then when fixing the bug you should just add the new bug ID to the list of space-delimited bugs in the @bug tag of the failing regression test, even if the test didn’t need to be updated.

   No

   An explanation for why a test is unnecessary or infeasible is required. Such an explanation is recorded by adding a label, and possibly a comment, to the bug report. The label has the prefix noreg for regression tests and nounit for unit tests. The suffix of the label is described in detail in the JBS Label Dictionary

7. Is modification of shared Java code needed?

   Yes

   It’s often sufficient to build and test on a single platform, but sometimes it’s not. Use discretion.

   No

   Continue

8. Is modification of shared C code needed?

   Yes

   Build and test on at least one instance of all three of the supported operating systems (Solaris, Linux, and Windows).

   No

   Continue

9. Is modification of C or Java platform-specific code needed?

   Yes

   Build and test on all relevant platforms. Code under src/solaris builds on Solaris, Linux, and MacOS X despite its name.

   No

   Continue

10. Run relevant regression and unit tests on all relevant platforms

    These include tests for external interfaces as well as other kinds of tests, e.g., HotSpot tests that use internal verification mechanisms.

11. Run relevant JCK tests on all relevant platforms

    Running JCK tests is particularly important if the change may have unexpected side-effects.

12. Request a review of the changes by sending an e-mail to the development alias

    A patch can be submitted as described in Contributing. Alternatively, a “webrev” may be generated and uploaded to the community code review server. The complete webrev generation and upload procedure is described at https://cr.openjdk.java.net.

    Changeset pushes before the Feature Complete require at least one Reviewer; pushes after the Feature Complete require at least two Reviewers. In either case, the more the merrier. Some teams may require more Reviewers. Check with members of the Project.

    Reviewers should examine not only the code being added or changed but also the relevant unit or regression tests.

    A change may require multiple Reviewers because it affects multiple areas. Reviewers should be aware that they take full responsibility for the appropriateness and correctness of any changes in their area of expertise. If something goes wrong (e.g., the build breaks) and the change's author is unavailable, they may be asked to deal with the problem. Potential Reviewers are encouraged to refuse to review code for which they aren’t qualified.

13. Create a changeset

    Follow the instructions in Working With Pull Requests.

14. Update the bug content

    Bug descriptions and comments should be written in a professional manner.

15. Push the changeset into the Project’s forest

    Follow the instructions in Working With Pull Requests. If working with a Sponsor, send the changeset to the development mailing list so that they can handle the final push.

    The push will trigger a update to the bug which will make the following changes:

    • Set the bug’s “Status” to “Resolved” and “Resolution to”Fixed".
    • Set the bug’s “Fix Version/s” to an appropriate release.
    • Set the bug’s “Resolved in Build” to “team”.
    • Add a Comment containing a reference to the changeset.

Congratulations! Your changeset will now make its way towards a promoted build. When the changeset becomes part of a promoted build, the bug’s “Resolved in Build” will have a value of "b[1-9][0-9]*" to indicate the build number.

Cloning the JDK

The complete source code for the JDK is hosted at GitHub. You can browse the code directly in the openjdk/jdk repository, or download the code for offline browsing, editing, and building using git clone.

$ git clone https://github.com/openjdk/jdk.git

openjdk/jdk is the mainline JDK development repository where the next major release of the JDK is being developed. Other projects have their own repositories on GitHub.

If you intend to contribute patches, you should first fork the repository on GitHub and clone your own personal fork as shown below. To fork a project on GitHub, go to the GitHub project page and click the ‘Fork’ button in the upper right corner, then follow the on screen instructions.

This is the typical development model:

Diagram of upstream repos and user's clone

Pushes to your personal fork can be made either using HTTPS or SSH. These examples assume you have an SSH key installed on GitHub. If this is the first time you clone your personal fork of an OpenJDK repository you may want to create an SSH key to use with it. See Generating an SSH key below. Once you have your personal fork and an SSH key to go with it, go ahead and clone.

$ git clone git@github.com:OpenDuke/jdk.git
$ cd jdk
$ git remote add upstream https://github.com/openjdk/jdk.git

In the example above Duke cloned his personal fork of the JDK mainline repository using SSH. You should of course use your own GitHub username instead. Then, by adding a new remote named ‘upstream’, the clone is associated with openjdk/jdk. Doing this will allow the tooling to automatically create a PR on openjdk/jdk whenever a change is pushed to the personal fork. The way that works is that once the change has been pushed to the personal fork, and you navigate to the openjdk/jdk repository on GitHub, there will be a message saying that you just pushed a change and asking if you want to create a PR.

It is strongly recommended to always create a new branch for any change you intend to implement. If your PR gets accepted, it will be squashed and pushed by the OpenJDK bots. This means that if you make changes to your master branch, it will diverge from the upstream master branch. This in turn means that your repo will forever be out of sync with the upstream repo, which will cause merge conflicts every single time you want to update your repo from upstream. Having a separate branch for each change also means that you can easily work on many different changes in parallel in the same code repository. Unless you know what you are doing, the recommendation is also to always base your new branch on the master branch.

$ git switch -c JDK-8272373 master

Here we create a new branch called JDK-8272373 based on the master branch and set the repository up to work in that new branch.

git switch was introduced in Git version 2.23. For earlier versions of Git git checkout can be used instead. However it is always recommended to use the latest versions of all your tools when possible.

Generating an SSH key

For security reasons you should always create new keys and use different keys with each repository you clone. The ssh-keygen command generates an SSH key. The -t option determines which type of key to create. ed25519 is recommended. -C is used to add a comment in the key file, to help you remember which key it is. While it’s possible to use SSH without a passphrase, this is strongly discouraged. Empty or insecure passphrases may be reset using ssh-keygen -p; this doesn’t change the keys.

$ ssh-keygen -t ed25519 -C openjdk-jdk -f ~/.ssh/openjdk-jdk
Generating public/private ed25519 key pair.
Enter passphrase (empty for no passphrase):
Enter same passphrase again:
Your identification has been saved in /Users/duke/.ssh/openjdk-jdk.
Your public key has been saved in /Users/duke/.ssh/openjdk-jdk.pub.
The key fingerprint is:
SHA256:WS4jCQMtat75ZEue+so+Lgj7V/sdMtj1FTNkfNsCfHA openjdk-jdk
The key's randomart image is:
+--[ED25519 256]--+
|  ..       ..oE  |
|  ...       o+o .|
| . .o     .  o+.o|
|..   o . +    .=.|
|o . . o S o   .. |
|.. o +.+ + . .   |
|o.  *.+.+ . .    |
|o....=.  + .     |
| .=B=. .. .      |
+----[SHA256]-----+

~/.ssh/openjdk-jdk is a text file containing your private ssh key. There’s a corresponding public key in ~/.ssh/openjdk-jdk.pub (as detailed in the example above). You should never share your private key. The public key on the other hand should be uploaded to GitHub. Follow the steps below to do that.

• Go to the GitHub settings for your account by choosing “Settings” in the menu by your avatar in the upper right corner
• Go to “SSH and GPG keys”
• Click “New SSH key”
• Title “OpenJDK” (or something else appropriate)
• Paste the content of ~/.ssh/openjdk-jdk.pub into the text field
  • To get the content of the file you can for instance use cat ~/.ssh/openjdk-jdk.pub
  • It will look something like this: ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIO8+egiIgWV+tE7LVVJmlR7WS2Lr3Fj7dXVo9HiasD6T openjdk-jdk
• Click “Add SSH key”

Now you are ready to clone your openjdk/jdk fork using SSH.

Building the JDK

The JDK build system is a fairly complex machine that has the ability to build anything from a single module to a complete shippable JDK bundle with various levels of debug capabilities, run tests, install your newly built JDK on your system, or cross-compile for some other system. The build uses make and a few other tools that you will have to install on your system before starting.

The JDK supports incremental builds. This means that if you have a complete build and make changes in just a single part of the JDK (e.g. a module or part of the JVM), only that particular part needs to be rebuilt. So subsequent builds will be faster and you can always use a make target that results in a complete JDK image without having to worry about actually building the entire JDK every time. Please note that the incremental build do have limits in its understanding of what you change. For instance, if you change behaviors or conventions in one module there may be other parts of the JDK that implicitly depends on these without make’s knowledge. For this reason you may have to rebuild several modules, or do a clean build if you change things that may have a wider impact.

The examples below show the steps taken to build the JDK source code. Please see Cloning the JDK for information on how to download it. These examples were written in the JDK 17 development time frame which is why the boot JDK used here is JDK 16. Note that the download links used here point to JDK 16 bundles. To build JDK N, use JDK N-1 as the boot JDK.

The configure script will tell you what additional packages you need. In this first example several packages were needed since this build was performed on a clean Ubuntu installation. The configure script was run several times to get all the dependencies, but only the commands actually needed to get the JDK built are included in the log. This is just an example log, don’t copy the apt-get install line. Instead run sh ./configure to see what packages you actually need on your system.

$ wget https://download.java.net/java/GA/jdk16/7863447f0ab643c585b9bdebf67c69db/36/GPL/openjdk-16_linux-x64_bin.tar.gz
$ tar xzf openjdk-16_linux-x64_bin.tar.gz
$ sudo apt-get install autoconf zip make gcc g++ libx11-dev libxext-dev libxrender-dev libxrandr-dev libxtst-dev libxt-dev libcups2-dev libfontconfig1-dev libasound2-dev
$ cd jdk
$ sh ./configure --with-boot-jdk=$HOME/jdk-16/
$ make images

The built JDK can be found in build/linux-x86_64-server-release/jdk. The exact path depends on your build platform and selected configuration.

The second example is from a clean (newly installed) Mac running MacOS Big Sur. Please note that in this case there are some steps taken outside of the terminal. First XCode and the XCode command line tools must be installed. It could be that the most recent version of XCode that you get from App Store is too new to have been properly tested with the JDK build. See the JDK build instructions for supported versions and more details in case you need to install an older version of XCode. In this example Mac Ports is used to install autoconf. autoconf can also be installed using Homebrew and surely through other sources as well.

$ curl https://download.java.net/java/GA/jdk16.0.1/7147401fd7354114ac51ef3e1328291f/9/GPL/openjdk-16.0.1_osx-x64_bin.tar.gz --output openjdk-16.0.1_osx-x64_bin.tar.gz
$ tar xzf openjdk-16.0.1_osx-x64_bin.tar.gz
$ sudo port install autoconf
$ sh ./configure --with-boot-jdk=$HOME/jdk-16.0.1.jdk/Contents/Home
$ make images

In this case the built JDK can be found in build/macosx-x86_64-server-release/jdk.

Configuration options

The JDK build is extremely configurable. This list only contains the most basic configure options needed to get you started. Use configure --help to see a complete list of options.

Working with multiple configurations

Through the configure flags you will select what configuration of the JDK to build. The name of the output directory for the build depends on this configuration. In the example above the JDK ended up in linux-x86_64-server-release. This means that we made a release build of a 64 bit linux x86 version of the server JDK. If we change some of these options the output directory will be affected accordingly.

--with-debug-level is one example of a configure option that will change the output directory name. Sometimes it makes sense to have several different configurations in parallel. For example while debugging some code you might want to have both a debug build and a release build to be able to test it properly. The directory naming scheme makes this very easy. Simply configure and build the JDKs you need and they will end up next to each other in the build directory.

In the example above we built a release image. To build a debug image as well we can configure with --with-debug-level=slowdebug. This will give us a JDK where for instance asserts in the JDK source code are enabled. To select which JDK to work with in later calls to make add CONF=<configuration>.

$ sh ./configure --with-boot-jdk=$HOME/jdk-16/ --with-debug-level=slowdebug
$ make CONF=slowdebug images
$ ls build/
linux-x86_64-server-release
linux-x86_64-server-slowdebug

Make targets

make images, as used in the example above, will build a JDK image which is very close to what you’d get from any JDK provider. There are several other make targets you can use depending on what you’re looking for. The table below contains some commonly used make targets.

There are many other targets available as well. Use make help to find out more.

Testing the JDK

In addition to your own Java applications, OpenJDK has support for two test frameworks to test the JDK, jtreg and GTest. jtreg is a Java regression test framework that is used for most of the tests that are included in the OpenJDK source repository. The Google Test (GTest) framework is intended for unit testing of the C++ native code. Currently only JVM testing is supported by the GTest framework. Other areas use jtreg for unit testing of C++ code.

This section provides a brief summary of how to get started with testing in OpenJDK. For more information on configuration and how to use the OpenJDK test framework, a.k.a. “run-test framework”, see doc/testing.md.

Please note that what’s mentioned later in this section, like GHA and tier 1 testing, will only run a set of smoke-tests to ensure your change compiles and runs on a variety of platforms. They won’t do any targeted testing on the particular code you have changed. You must always make sure your change works as expected before pushing, using targeted testing. In general all changes should come with a regression test, so if you’re writing product code you should also be writing test code. Including the new tests (in the right places) in your change will ensure your tests will be run as part of your testing on all platforms and in the future.

A few key items to think about when writing a regression test:

• A regression test should execute fast - a few seconds at most
• The test should only test the desired functionality - if you have several features to test, write more tests
• The test should pass reliably on all supported platforms - watch out for platform-specific differences such as path separators
• Binary files shouldn’t be checked in, if your test needs to use one, the test should create it in some fashion
• Avoid shell scripts and relying on external commands as much as possible

The jtreg documentation has a section on how to write good jtreg tests.

There are a few examples where it doesn’t make sense to write an explicit regression test. These should be tagged in JBS with one of the noreg-labels.

jtreg

In-depth documentation about the jtreg framework is found here: jtreg harness. jtreg itself is available in the Code Tools Project.

Below is a small example of a jtreg test. It’s a clean Java class with a main method that is called from the test harness. If the test fails we throw a RuntimeException. This is picked up by the harness and is reported as a test failure. Try to always write a meaningful message in the exception. One that actually helps with understanding what went wrong once the test fails.

/*
 * @test
 * @summary Make sure feature X handles Y correctly
 * @run main TestXY
 */
public class TestXY {
    public static void main(String[] args) throws Exception {
        var result = X.y();
        if (result != expected_result) {
            throw new RuntimeException("X.y() gave " + result + ", expected " + expected_result);
        }
    }
}

This example only utilizes three jtreg specific tags, @test, @summary, and @run. @test simply tells jtreg that this class is a test, and @summary provides a description of the test. @run tells jtreg how to execute the test. In this case we simply tell jtreg to execute the main method of the class TestXY. @run isn’t strictly necessary for jtreg to execute the test, an implicit @run tag will be added if none exists. However, for clarity and in order to avoid bugs it’s recommended to always explicitly use the @run tag.

There are several other tags that can be used in jtreg tests. You can for instance associate the test with a specific bug that this test is a regression test for. The bug id is written without the JDK- prefix.

@bug 8272373

You can add several bug ids in the same @bug tag, separated by a single space. These bug ids refer to product bugs for which a fix is verified by this test. JBS issues that track changes to the test itself are not listed here.

You can also specify a number of requirements that must be fulfilled for jtreg to execute the test.

@requires docker.support
@requires os.family != "windows"
@requires os.maxMemory > 3G
@requires os.arch=="x86_64" | os.arch=="amd64"

And you can specify if the test requires specific modules, or command line flags to run the test in several different ways.

@modules java.base/jdk.internal.misc
@run main/othervm -Xmx128m TestXY

Note that you can have several @run tags in the same test with different command line options.

jtreg also have support for labeling tests with keys using the @key tag. These keywords can then be used to filter the test selection. For instance if you have a UI test which needs to display a window you’ll want to make sure the test harness doesn’t try to run this test on a system which doesn’t support headful tests. You do this by specifying

@key headful

Another example is @key randomness that should be used to indicate that a test is using randomness - i.e. is intentionally non-deterministic.

There are many other keywords in use and their usage may differ between areas in the JDK. Make sure you understand the conventions for the particular area you are testing since these are just examples.

The jtreg documentation provides information on many more tags like these.

The compiler group has a section in their wiki with Guidelines for “langtools” tests.

Running OpenJDK jtreg tests

When configuring the OpenJDK build you can tell it where your jtreg installation is located. When providing this information you can later run make run-test to execute jtreg tests.

sh ./configure --with-jtreg=/path/to/jtreg
make run-test TEST=tier1

In the OpenJDK source tree you can find a directory called test. There are a large number of tests in this directory that are written to be used with jtreg.

make run-test TEST=test/jdk/java/lang/String/

You can also run jtreg without invoking make. In this case you’ll need to tell jtreg which JDK to test.

jtreg -jdk:/path/to/jdk /path/to/test

GTest

As mentioned the Google test framework is mainly used for C++ unit tests. There are several of these in the test/hotspot directory. Currently, only the C++ code in the JVM area is supported by the OpenJDK GTest framework. The tests can be run without starting the JVM, which enables testing of JVM data structures that would be fragile to play with in a running JVM.

static int demo_comparator(int a, int b) {
  if (a == b) {
    return 0;
  }
  if (a < b) {
    return -1;
  }
  return 1;
}

TEST(Demo, quicksort) {
  int test_array[] = {7,1,5,3,6,9,8,2,4,0};
  int expected_array[] = {0,1,2,3,4,5,6,7,8,9};

  QuickSort::sort(test_array, 10, demo_comparator, false);
  for (int i = 0; i < 10; i++) {
    ASSERT_EQ(expected_array[i], test_array[i]);
  }
}

ASSERT_EQ is one example of an assertion that can be used in the test. Below are a few other examples. A full list is found in the Google Test Documentation.

ASSERT_TRUE(condition);
ASSERT_FALSE(condition);
EXPECT_EQ(expected, actual);
EXPECT_LT(val1, val2);
EXPECT_STREQ(expected_str, actual_str);

ASSERT is a fatal assertion and will interrupt execution of the current sub-routine. EXPECT is a nonfatal assertion and will report the error but continues to run the test. All assertions have both an ASSERT and an EXPECT variant.

For more information on how to write good GTests in HotSpot, see doc/hotspot-unit-tests.md.

Running OpenJDK GTests

When configuring the OpenJDK build you can tell it where your GTest installation is located. Once configured, use make to run GTests.

sh ./configure --with-gtest=/path/to/gtest
make test TEST=gtest

You can also use a regular expression to filter which tests to run:

make test TEST=gtest:code.*:os.*
make test TEST=gtest:$X/$variant

The second example above runs tests which match the regexp $X.* on a specific variant of the JVM. The variant is one of client, server, etc.

GitHub actions

GitHub has a feature called GitHub actions (GHA) that can be used to automate testing. The GHA is executed whenever a push is made to a branch in your repository. The bots will show the results of the GHA in your PR when you create or update it. The GHA in the JDK project is configured to run a set of tests that is commonly known as tier 1. This is a relatively fast, small set of tests that tries to verify that your change didn’t break the JDK completely. In tier 1 the JDK is built on a small set of platforms including (but not necessarily limited to) Linux, Windows, and MacOS, and a few tests are executed using these builds. There’s also a set of other platforms for which GHA does cross-complilation builds.

In addition to the testing you run manually before publishing your changes, it’s recommended that you take advantage of this automated testing that the GHA offers. This will for instance allow you to run tests on platforms that you may not otherwise have access to. To enable this on your personal fork of the JDK on GitHub go to the “Actions” tab and click the big green button saying “I understand my workflows, go ahead and enable them”. If you don’t understand these workflows there’s a link to the actual file that describes them right below the green button.

In case of failures in the GHA it’s always a good start to try to reproduce the failure locally on a machine where you have better control and easier access to a debug environment. There have been cases where the GHA has failed due to issues unrelated to the change being tested, e.g. because the GHA environment was updated and changes were needed to the JDK GHA configuration. The configuration is in general updated fairly quickly, so in cases were you can’t reproduce the failure locally, consider re-running the GHA later.

Please keep in mind that the tier 1 tests run by the GHA should only be seen as a smoke test that finds the most critical breakages, like build errors or if the JDK is DOA. These tests can never replace the targeted testing that you always must do on your changes. There are several areas of the JDK that aren’t part of tier 1 at all. To see exactly what tier 1 includes, please see the various TEST.groups files that you will find in the subdirectories of jdk/test/.

Excluding a test

Sometimes tests break. It could be e.g. due to bugs in the test itself, due to changed functionality in the code that the test is testing, or changes in the environment where the test is executed. While working on a fix, it can be useful to stop the test from being executed in everyone else’s testing to reduce noise, especially if the test is expected to fail for more than a day. There are two ways to stop a test from being run in standard test runs: ProblemListing and using the @ignore keyword. Removing tests isn’t the standard way to remove a failure. A failing test is often a regression and should ideally be handled with high urgency.

Remember to remove the JBS id from the ProblemList or the test when the bug is closed. This is especially easy to forget if a bug is closed as Duplicate or Won’t Fix. jcheck will report an error and prohibit a push if a PR is created for an issue that is found in a ProblemList if the fix doesn’t remove the bug ID from the ProblemList.

ProblemListing jtreg tests

ProblemListing should be used for a short term exclusion while a test is being fixed, and for the exclusion of intermittently failing tests that cause too much noise, but can still be useful to run on an ad-hoc basis. ProblemListing is done in the file ProblemList.txt. For more details about the ProblemList.txt file see the jtreg FAQ.

There are actually several ProblemList files to choose from in the JDK. Their location and name hint about what area or feature each file belongs to. Each file has sections for different components. All ProblemList files complement each other to build the total set of tests to exclude in jtreg runs.

test/hotspot/jtreg/ProblemList.txt
test/hotspot/jtreg/ProblemList-aot.txt
test/hotspot/jtreg/ProblemList-graal.txt
test/hotspot/jtreg/ProblemList-non-cds-mode.txt
test/hotspot/jtreg/ProblemList-Xcomp.txt
test/hotspot/jtreg/ProblemList-zgc.txt
test/jaxp/ProblemList.txt
test/jdk/ProblemList.txt
test/jdk/ProblemList-aot.txt
test/jdk/ProblemList-graal.txt
test/jdk/ProblemList-Xcomp.txt
test/langtools/ProblemList.txt
test/langtools/ProblemList-graal.txt
test/lib-test/ProblemList.txt

Use the suitable ProblemList and add a line like this in the proper section:

foo/bar/MyTest.java                        4711   windows-all

In this example, MyTest.java is ProblemListed on windows, tracked by bug JDK-4711.

Currently there’s no support for multiple lines for the same test. For this reason it’s important to always make sure there’s no existing entry for the test before adding a new one, as multiple entries might lead to unexpected results, e.g.

foo/bar/MyTest.java                        4710   generic-all
...
foo/bar/MyTest.java                        4711   windows-all

This would lead to sun.tools.jcmd.MyTest.java being ProblemListed only on windows-all. The proper way to write this is:

foo/bar/MyTest.java                        4710,4711   generic-all,windows-all

Although windows-all isn’t strictly required in this example, it’s preferable to specify platforms for each bugid (unless they are all generic-all), this makes it easier to remove one of the bugs from the list.

A common way to handle the JBS issue used to problemlist a test is to create a Sub-task of the bug that needs to be fixed to be able to remove the test from the problem list again.

Remember to always add a problemlist label in the JBS issue referenced in the ProblemList entry.

ProblemListing some, but not all, test cases in a file

Some tests contain several test cases and there may be a need to ProblemList only a few of them. To do this use the full test name, i.e. <filename> + # + <test case id>, where test case id can be specified in the test header. If no id is specified each test case can be referenced with id + ordinary number of the test case in the test file.

Let’s assume we have four test cases in foo/bar/MyTest.java:

/* @test */
/* @test id=fancy_name */
/* @test */
/* @test */

A ProblemList entry that excludes the first, second, and third test case would look like this:

foo/bar/MyTest.java#id0          4720  generic-all
foo/bar/MyTest.java#fancy_name   4721  generic-all
foo/bar/MyTest.java#id2          4722  generic-all

Due to an issue described in CODETOOLS-7902712 tests that contains more than one @test must actually use this way to specify all test cases if all of them should be ProblemListed. Specifying just the test name will not work.

Running ProblemListed tests

To run ad-hoc runs of ProblemListed tests use RUN_PROBLEM_LISTS=true.

make test TEST=... JTREG=RUN_PROBLEM_LISTS=true

Exclude jtreg tests using @ignore

The @ignore keyword is used in the test source code. This is mainly used for tests that are so broken that they may be harmful or useless, and is less common than ProblemListing. Examples can be tests that don’t compile because something changed in the platform; or a test which might remove your /etc/shadow. Use @ignore with a bug reference in the test case to prevent the test from being run.

/**
 *  @test
 *  @ignore 4711
 */

In this example, MyTest.java is excluded, tracked by bug JDK-4711. @ignore should generally be placed directly before the first “action” tag (e.g. @compile, @run) if any in the test.

Dealing with JBS bugs for test exclusion

ProblemListing and @ignore-ing are done in the JDK source tree, that means a check-in into the repository is needed. This in turn means that a unique JBS issue and a code review are needed. This is a good thing since it makes test problems visible.

• Code review: ProblemListing a test is considered a [trivial] change.
• JBS issue: A JBS issue is obviously created for the bug that caused the failure, we call this the main issue. To exclude the test, create a subtask to the main issue. Also add the label problemlist to the main issue.

The fix for the main issue should remove the test from the ProblemList or remove the @ignore keyword from the test.

Triage excluded issues

After a failure is handled by excluding a test, the main JBS issue should be re-triaged and possibly given a new priority. This should be handled by the standard triage process. A test exclusion results in an outage in our testing. This outage should be taken into consideration when triaging, in addition to the impact of the bug itself.

Backing out a change

If a change causes a regression that can’t be fixed within reasonable time, the best way to handle the regression can be to back out the change. Backing out means that the inverse (anti-delta) of the change is pushed to effectively undo the change in the repository. There are two parts to this task, how to do the bookkeeping in JBS, and how to do the actual backout in git or Mercurial.

The backout is a regular change and will have to go through the standard code review process, but is considered a [trivial] change. The rationale is that a backout is usually urgent in nature and the change itself is automatically generated. In areas where two reviewers are normally required, only one additional Reviewer is required for a backout since the person who is performing the backout also will review the change.

How to work with JBS when a change is backed out

1. Close the original (failed) JBS issue (O).
   • “Verify” the issue and choose “Fix Failed”.
2. If the intention is to fix the change and submit it again, create a redo-issue (R) to track that the work still needs to be done.
   • Clone (O) and add the prefix [REDO] on the summary - the clone becomes the redo-issue (R).
   • Make sure relevant information is brought to (R).
   • Remember that comments aren’t automatically brought over when cloning.
3. Create a backout-issue (B):
   • Alternative 1 - the regression is identified directly.
     • Create a sub-task to (R) with the same summary prefixed with [BACKOUT].
   • Alternative 2 - an investigation issue was created (I), and during the investigation backing out the change is identified as the best solution.
     • Use the investigation issue (I) for the backout.
     • Change summary of (I) to the same as (O) and prefix with [BACKOUT].
     • Move and change type of (I) to become a sub-task of (R).
   • Alternative 3 - no redo issue was created.
     • Create a backout-issue (B) with the same summary as (O), prefix with [BACKOUT].
     • Add a relates to link between (B) and (O).

ProblemList entries and @ignore keywords will continue to point to the original bug (unless updated at back out). This is accepted since there is a clone link to follow.

How to work with git when a change is backed out

To backout a change with git, use git revert. This will apply (commit) the anti-delta of the change.

$ git show aa371b4f02c2f809eb9cd3e52aa12b639bed1ef5
commit aa371b4f02c2f809eb9cd3e52aa12b639bed1ef5 (HEAD -> master)
Author: Jesper Wilhelmsson <jesper.wilhelmsson@oracle.com>
Date:   Wed Jun 23 20:31:32 2021 +0200

    8272373: Make the JDK mine
    Reviewed-by: duke

diff --git a/README.md b/README.md
index 399e7cc311f..4961acb2126 100644
--- a/README.md
+++ b/README.md
@@ -1,4 +1,4 @@
-# Welcome to the JDK!
+# Welcome to my modified JDK!

 For build instructions please see the
 [online documentation](https://openjdk.org/groups/build/doc/building.html),

$ git revert aa371b4f02c2f809eb9cd3e52aa12b639bed1ef5
[master d454489052d] 8280996: [BACKOUT] Make the JDK mine Reviewed-by: duke
 1 file changed, 1 insertion(+), 1 deletion(-)

$ git show d454489052dc6ff69a21ad9c8f56b67fdeb435ee
commit d454489052dc6ff69a21ad9c8f56b67fdeb435ee (HEAD -> master)
Author: Jesper Wilhelmsson <jesper.wilhelmsson@oracle.com>
Date:   Wed Jun 23 20:32:08 2021 +0200

    8280996: [BACKOUT] Make the JDK mine
    Reviewed-by: duke

diff --git a/README.md b/README.md
index 4961acb2126..399e7cc311f 100644
--- a/README.md
+++ b/README.md
@@ -1,4 +1,4 @@
-# Welcome to my modified JDK!
+# Welcome to the JDK!

 For build instructions please see the
 [online documentation](https://openjdk.org/groups/build/doc/building.html),

How to work with Mercurial when a change is backed out

In order to backout a change, the hg backout command is recommended, which essentially applies the anti-delta of the change. Make sure you perform the backout in the most upstream repository the change has escaped to.

hg backout [OPTION]... [-r] REV

reverse effect of earlier changeset

    Prepare a new changeset with the effect of REV undone in the current
    working directory.

    If REV is the parent of the working directory, then this new changeset is
    committed automatically. Otherwise, hg needs to merge the changes and the
    merged result is left uncommitted.

Backing out a backport

In rare cases it may be necessary to back out a backport from an update release without backing out the original fix in mainline. This will require a somewhat different procedure and will result in a small mess in JBS. It’s extremely important to add comments in all relevant issues explaining exactly what’s happened.

The steps to take in order to do this are described below. (M) used below refers to the main bug entry - the first fix that was later backported.

1. Close the original (failed) JBS backport issue (O).
   • “Verify” the issue and choose “Fix Failed”.
2. If the intention is to fix the backport and submit it again, create a redo-issue (R) to track that the work still needs to be done.
   • Clone (M) and add the prefix [REDO BACKPORT] on the summary - the clone becomes the redo-issue (R).
   • Add a relates to link between (R) and (O).
   • Set Fix Version of (R) to the target release for the backport - either the exact release if known, or <N>-pool if it’s not critical which release the fixed backport goes into.
3. Create a backout-issue (B):
   • Alternative 1 - the broken backport is identified directly.
     • Create a sub-task to (R) with the same summary, but prefixed with [BACKOUT BACKPORT].
   • Alternative 2 - an investigation issue was created (I), and during the investigation backing out the backport is identified as the best solution.
     • Use the investigation issue (I) for the backout.
     • Change summary of (I) to the same as (M) and prefix with [BACKOUT BACKPORT].
     • Move and change type of (I) to become a sub-task of (R).
   • Alternative 3 - no redo issue was created.
     • Create a backout-issue (B) with the same summary as (M) and prefix with [BACKOUT BACKPORT].
     • Add a relates to link between (B) and (M).
4. Add comments to (M), (R) and (O) explaining the situation.

The end result in JBS should look like this:

JBS structure after backout and redo of a backport

For this example in JBS see the 15.0.2 backport of JDK-8272373.

Rationale for using this model

The approach described here has both advantages and disadvantages. The key invariants that lead to this model are:

• A backported by link should only refer to issues of type Backport
• A bug id should never be reused for different patches in the same repository

Disadvantages of this model are that the list of backports in JBS will still list the old (failed) backport as the 15.0.2 backport, and the new backport will not be linked to using a backported by link. It is assumed that the advantages above outweighs the disadvantages and that the capital letter prefixes for the backout and the redo will be visible enough in JBS to alert that something fishy is going on.

Working With Pull Requests

Once you have made a change that you want to integrate into an OpenJDK code base you need to create a Pull Request (PR) on GitHub. This section assumes that you have previous experience from using git and GitHub and won’t go into details of how those work. Still, the aim is of course to provide a useful guide, so send an email if more details are needed.

This section also assumes that you have already read I have a patch, what do I do? and followed all the steps there.

Think once more

All code reviews in OpenJDK are done in public. Once you open your PR for public review the internet can see it and comment on it. Make sure your code is ready for it. Look through your comments, make sure that temporary code is gone, and make sure you have sanitized your method and variable names. Also, make sure you understand your code. Why is it working? What are the potential pitfalls? What are the edge-cases? If you haven’t already answered all these questions in the mail conversation that preceded this PR, it’s likely that you will need to answer them during the review.

It’s also worth taking the extra time to see if the change can be split into a few different separate changes. A large change will take more effort and thus attract fewer Reviewers. Smaller changes will get reviewed faster and get better quality reviews. You can compare proposing a single large change to proposing ten individual small unrelated changes. What happens in practice when all these ten changes are presented as one PR is that there’s a focus on say 5-6 of these smaller changes and no one really looks hard at the other 4-5. For complexity, even small changes that are hard to understand and test may be risky.

The timing of your change will also affect the availability of reviewers. The JDK runs on a six-months release cadence. During the months around the start of the ramp down phase most area experts will be busy working on their own changes and reviewing major features that are planned for the release. If you propose a change during this period (basically May-June, or November-December) it may take a long time before you get the required reviews.

Rebase before creating the PR

It’s likely that other people have pushed changes to the code base since you created your branch. Make sure to pull the latest changes and rebase your fix on top of that before creating your PR. This is a courtesy issue. Your reviewers shouldn’t have to read your patch on top of old code that has since changed. This is hopefully obvious in cases where the upstream code has gone through cleanups or refactorings, and your patch may need similar cleanups in order to even compile. But even in cases where only smaller changes have been done, the reviewers shouldn’t have to react to issues like “that line of code was moved last week, why is it back there?”.

git rebase master

After the PR has been published, rebasing, force-pushing, and similar actions are strongly discouraged. Such actions will disrupt the workflow for reviewers who fetch the PR branch. Pushing new changes is fine (and even merging if necessary) for a PR under review. Incremental diffs and other tools will help your reviewers see what you have changed. In the end, all commits will be squashed into a single commit automatically, so there’re actually no drawbacks whatsoever to making commits to a PR branch during review.

Final check before creating the PR

Creating the PR is essentially the same as asking a large group of people to start reviewing your change. Before doing that, you want to make sure your change is done in every detail you have the power to control. These are a few of the things you should think about in order to avoid wasting people’s time on an unfinished change. (You may think that some of these are too obvious to even mention, but all of them are things that in the past have caused actual failures that broke the JDK for all developers out there.)

• Is the copyright statement at the top of each modified source file correct?

• Did you run a full build and all relevant tests on the final version of the change? It’s important to do this on the truly final version, as even an apparently trivial change in a comment can break the build.

• Did you git add all new files?

• Did you add regression tests for your change?

• Did you run those new regression tests?

If you are unsure of any of these things but still want to go ahead and create the PR, don’t!

If you have an actual reason to create a PR before the change is all done, make sure to create it in DRAFT mode. The bot won’t add the rfr label or send emails to labeled groups as long as the PR is in DRAFT mode.

Life of a PR

1. Make sure the PR is reviewable

   There are changes that span across several areas, for example wide spread cleanups or the introduction of a new langauge feature. Accordingly, the number of lines of code touched can be quite large, which makes it harder to review the entire PR. In such cases, it may make sense to split the change into several PRs, most commonly by grouping them by module or area.

2. Set a correctly formatted title

   The title of the PR should be of the form “nnnnnnn: Title of JBS issue” where nnnnnnn is the JBS issue id of the main JBS issue that is being fixed, and the Title of JBS issue is the exact title of the issue as written in JBS. In fact, the title can be set to only the JBS issue id (nnnnnnn) in which case the bot will fetch the title from JBS automatically. If you are creating a backport PR, see Using the Skara tooling to help with backports for more details on the title requirements.

3. Write a useful description

   The description of the PR should state what problem is being solved and shortly describe how it’s solved. Reviewers and other interested readers are referred to the text in the JBS issue for details, but the description should be enough to give an overview. This assumes there’s useful information in the JBS issue, like an evaluation etc. If not, add it.

   Remember that the description is included in many emails sent to lists with many receivers, so a too long description can cause a lot of noise, while of course a too short description won’t give the reader enough information to perform the review. If you have a lot of information you wish to add to your PR, like performance evaluations, you can put that in a separate comment in the PR.

4. Finish the change before publishing it

   Each update to a published PR will result in emails being sent to all relevant lists. This is per design and it’s how we want it to be, but it also mean that if you publish a PR before you have gone through the final check mentioned above, and later find that a few more updates are necessary, a lot of people will get a lot of emails.

5. Make sure all relevant groups are included

   The bot will make an attempt to include the groups that need to review your change based on the location of the source code you have changed. There may be aspects of your change that are relevant to other groups as well, and the mapping from source to groups isn’t always perfect, so make sure all relevant groups have been included, and add new labels using /label if needed. Consult the Code Owners section if you are unsure of who owns the code you are changing.

6. Allow enough time for review

   In general all PRs should be open for at least 24 hours to allow for reviewers in all time zones to get a chance to see it. It may actually happen that even 24 hours isn’t enough. Take into account weekends, holidays, and vacation times throughout the world and you’ll realize that a change that requires more than just a trivial review may have to be open for a while. In some areas [trivial] changes are allowed to be pushed without the 24 hour delay. Ask your reviewers if you think this applies to your change.

7. Get the required reviews

   At least one Reviewer knowledgeable in each area being changed must approve every change. Some areas (e.g. Client and HotSpot) require two reviewers in most cases, so be sure to read the relevant OpenJDK group pages for advice or ask your sponsor.

   Be open to comments and polite in replies. Remember that the reviewer wants to improve the world just as much as you do, only in a slightly different way. If you don’t understand some comment, ask the reviewer to clarify. Accept authority when applicable. If you’re making changes in an area where you’re not the area expert, acknowledge that your reviewers may be. Take their advice seriously, even if it is to not make the change. There are many reasons why a change may get rejected. And you did read the section [Things to consider before changing OpenJDK code], right?

8. Updating the PR

   You may need to change the code in response to review comments. To do this, simply commit new changes and push them onto the PR branch. The PR will be updated automatically. Multiple commits to the branch will be squashed into a single commit when the PR is eventually integrated.

   If the set of files in the PR has changed, this may affect the groups that need to review the PR. Make sure to adjust the PR labels accordingly.

   If you want to update the title of the PR, remember that the one and only truth is the JBS issue, so the title must first be changed there.

9. Merge the latest changes

   If your PR is out for review for a longer time it’s a good habit to pull from the target repository regularly to keep the change up to date. This will make it easier to review the change and it will help you find issues caused by other changes sooner. Typically this involves fetching changes from the master branch of the main JDK repo, merging them into your local branch, resolving conflicts if necessary, and then pushing these changes to the PR branch. Pushing additional commits and merges into the PR branch is fine; they will be squashed into a single commit when the PR is integrated. Avoid rebasing changes, and prefer merging instead.

   If there are upstream changes that might affect your change, it’s likely a good idea to rerun relevant testing as well. The GHA testing that’s done automatically by GitHub should only be seen as a smoke test that finds the most severe problems with your change. It’s highly unlikely that it will test your actual change in any greater detail - or even at all execute the code that you have changed in most cases.

10. Integrate your change

    When you have the required reviews and have made sure all relevant areas have had a chance to look at your change, integrate by entering the command /integrate in a comment on the PR. If you are not yet a Committer in the project, ask your sponsor to enter the command /sponsor in the PR as well in order for your change to be allowed to be integrated.

11. After integration

    After you have integrated your change you are expected to stay around in case there are any issues with it. As mentioned above, you are expected to have run all relevant testing on your change before creating your PR, but regardless of how thorough you test it, things might slip through. After your change has been integrated an automatic pipeline of tests is triggered and your change will be tested on a variety of platforms and in a variety of different modes that the JDK can be executed in. A change that causes failures in this testing may be backed out if a fix can’t be provided fast enough, or if the developer isn’t responsive when noticed about the failure. Note that this directive should be interpreted as “it’s a really bad idea to push a change the last thing you do before bedtime, or the day before going on vacation”.

Webrevs

As all OpenJDK projects are hosted on GitHub, most code reviews takes place there. When you publish a PR to an OpenJDK repository the corresponding JBS issue will get a link to the code review in the PR, making this the natural place to go for review. OpenJDK do however provide other tools and infrastructure for code reviews as well: The webrevs.

Webrev refers to both the tool used to create them and its output. The script, webrev.ksh, is maintained in the Code Tools project. Please note that this version of webrev is for use with mercurial and won’t work with the git based repositories. You don’t actually need tools like this anymore unless you want to host code reviews in other locations than GitHub.

On the GitHub reviews you will find links to webrevs. These are automatically generated by the bot and are provided as a complement for those who prefer this style of code review. Many OpenJDK developers are used to the webrevs as this was the default way to provide code for review before OpenJDK moved to GitHub. Though webrevs are mainly deprecated today, they used to be a central part of OpenJDK development and you may still see people use the word as a synonym for code review, so they do deserve to be mentioned here as well.

File storage for OpenJDK artifacts - cr.openjdk.java.net

The cr.openjdk.java.net server provides storage and display of code review materials such as webrevs and other artifacts related to the OpenJDK Community. This area can also be used to store design documents and other documentation related OpenJDK but not related to any specific project that have an OpenJDK wiki space for such purposes.

Any OpenJDK Author can publish materials on the cr.openjdk.java.net server. Users can upload files to temporary storage using secure methods (rsync, scp, and sftp).

This site is for open source materials related to the OpenJDK Community only. Users uploading inappropriate materials will lose access and the material will be deleted. Please review the Terms of Use before using this server.

Backporting

Development of the latest version of the JDK often results in bug fixes that might be interesting to include in some of the JDK update releases still being maintained or in a feature release stabilization repository. Moving a fix from a more recent release train (e.g. JDK 21) to an older release train (e.g. JDK 17) is called backporting.

The guideline for what to backport into a specific release will vary over the lifetime of that release. Initially more fixes are expected to be backported as new features and large changes introduced in a mainline release stabilize. Over time the focus will shift from stabilization to keeping it stable - the release will go into maintenance mode. This means that bug fixes that require larger disruptive changes are more likely to be made in mainline and backported to more recent release trains only, and not to older release trains.

Over time it’s likely that the code base will diverge between mainline and any given update release, and the cost of backporting will increase. The cost in this case is not only the effort needed to perform the actual backport, but also the cost inferred by bugs introduced by the backport. This should be taken into consideration when deciding if a change should be backported or not. For more details on how to reason around what to backport, this email from JDK 8 Updates Project lead Andrew Haley has some guidelines for JDK 8u. The reasoning in this mail is specific to JDK 8u, but will in general apply to any JDK release in maintenance mode.

Any change that originally required a CSR will require a new CSR to be backported unless the backport was covered by the initial CSR. Changes to Java SE specifications cannot be made in an update release without a Java SE Maintenance Release. CSR-related issues affect interfaces and behavior and must be very carefully scrutinized.

Backporting to a feature release stabilization repository

During rampdown of a feature release there are two repositories in play, the release stabilization repository for the outgoing release, and the mainline repository where the next release is being developed. Any change going into the release stabilization repository is likely to be desired in mainline as well. When making a change intended both for the stabilization release and the mainline repository, you should always create your pull request targeting the mainline repository first, and then, once the pull request is integrated, backport the resulting commit to the release stabilization repository. For bugfixes that are only applicable to the release stabilization repository, regular pull requests targeting the stabilization fork should be created.

Please note that special rules applies during rampdown regarding what can and can’t be included into the rampdown repository. See the The JDK Release Process for more information.

Working with backports in JBS

Example of backport hierarchy

In general there’s no need to create backport issues in JBS manually. All work that’s done in JBS in preparation for a backport (requesting approvals etc) is done in the main issue. The backport issue will be created automatically by the bots when you integrate the change to the source code repository.

There can be cases where it’s desirable to create a backport issue before the fix is done, e.g. if a CSR needs to be filed. In these cases set the Fix Version/s of the backport to N or N-pool, where N is the release train the backport is targeting. E.g. 17-pool. Please note that even if a backport issue is created ahead of time, all work done in JBS for approvals and similar is still done in the main issue.

Obviously it’s possible to set the Fix Version/s to the exact release the backport is targeting, but in general this isn’t recommended unless you are targeting a feature release in rampdown. When a change is pushed to an update release repository, the bots will look at the main issue as indicated in the PR title, and look for backports with the current N.0.x release version as Fix Version/s, if no such backport is found they will look for N-pool, and if that isn’t found either, a new backport issue will be created. This means that if the backport has an exact Fix Version/s set, but is delayed and misses the release indicated by this Fix Version/s, a new backport issue is created with a small mess as the result. (See How to fix an incorrect backport creation in JBS.)

Setting the Fix Version/s of a backport targeted to an update release to N is always wrong. JDK N has already been released (or it wouldn’t be an update release) and can’t get any more fixes.

Requesting approvals for backports

In order to be allowed to push a change to one of the OpenJDK update development repositories (e.g. jdk17u-dev), an approval is required. The official process for how to request push approval for a backport describes in detail how to work with JBS when requesting approvals. In short, there’s a label jdk<release>u-fix-request that should be added to the main JBS issue. Also put a motivation as to why the issue needs to be backported as a comment in the main issue. Once the label and motivation has been added, wait for the maintainers of the release to approve your request. The approval will be indicated with a label, jdk<release>u-fix-yes, added to the main issue.

If the update release is in rampdown, changes are pushed to the release repository (e.g. jdk17u). During rampdown the bar to get changes in are significantly higher and fixes need to be approved with jdk<release>u-critical-request / jdk<release>u-critical-yes.

If your request to backport a change is denied, but you for some reason already created the backport issue in JBS (why?!), the backport issue should be closed as Won’t Fix.

Using the Skara tooling to help with backports

The Skara tooling includes support for backports. The official Skara documentation describes in detail how to work with the tooling to create backport PRs on GitHub or using the CLI tools. As described in the documentation, the /backport command can be used on a commit or a PR to create the backport PR. If a backport PR is manually created, set the PR title to Backport <original commit hash>. This ensures that the bots will recognize it as a backport as opposed to a main fix specifically targeting an older release. One can tell whether or not the bots recognized a PR as a backport by the backport label being added if it’s recognized.

How to fix an incorrect backport creation in JBS

If an issue is targeted to a release and a fix referring to that issue is pushed to a different release repository, then a backport issue is automatically created in JBS. Usually this is a “good thing”, e.g., when you are backporting a fix to an earlier release, but not always… If the main issue is targeted to a later release (due to schedule planning) but someone finds the time to fix that issue in the current release, or if the main issue is targeted to a feature release in rampdown and the fix is pushed to the mainline repository, then the issue should be retargeted to the correct release before pushing the fix. However, sometimes we forget.

Here’s how to fix that:

1. Reopen the backport issue that was created automatically

   • Use a comment like the following (in the reopen dialog):

   Fix was pushed while main issue was targeted to 'N'. Reset the main issue to fixed in 'N+1', reset this issue to fix in 'na' and closed as [Not An Issue]{.jbs-value} to avoid confusion.

   • Change the Fix Version/s from ‘N+1’ to ‘na’.
   • Close the backport issue as Not an Issue. Note: Closed, not Resolved

   Even if you intend to backport the issue from ‘N+1’ to ‘N’ you shouldn’t reuse this backport issue. The existing (bad) push notification comment and the later to be added correct push notification comment will look very similar and it’s just a disaster waiting to happen to deliberately add these to the same issue.

2. Clean up the main issue

   • Copy the push notification comment from the backport issue to the main issue, e.g.:

   Changeset: 12345678
   Author: Duke <duke@openjdk.org>
   Date: 2020-10-23 15:37:46 +0000
   URL: https://git.openjdk.org/jdk/commit/12345678

   • Add a comment like the following to the main issue:

   Fix was pushed while main issue was targeted to 'N'. Reset the main issue to fixed in 'N+1' and copied the Robo Duke entry here.

   • Reset the main issue Fix Version/s from ‘N’ to ‘N+1’.
   • Resolve the main issue as Fixed in build “team” or in build “master” depending on where the fix was pushed - or to an actual build number if the change has already made it to a promoted build (look in the backport issue if you are unsure). Pushes to ‘openjdk/jdk’ are fixed in build “master” and pushes to project repositories are fixed in build “team”.

Release Notes

Release notes for a product such as the JDK are part of the release deliverables providing a way to highlight information about a fix, such as when it may have changed behavior, or when it’s decided not to fix something. While what should go into a release note isn’t something that can be precisely defined, it should describe changes that are important for a user to take into account when they are upgrading to the specific version. While release notes should not duplicate information in other documents, they can serve to highlight that a change has been made.

Release notes are associated with a JBS issue that has been fixed (or in some cases not been fixed) in a release and are generated with each build of a release. Any note should be considered as an integral part of the fix process, rather than waiting until the end of the release to determine what to write. In OpenJDK, release notes are currently being generated for the JDK and JDK Updates projects.

Writing a release note

Writing the release note is the responsibility of the engineer who owns the issue. The note should be generated before the fix is reviewed, or in the case of known issues, when it’s determined that a fix won’t be possible in the release the issue was found in.

When writing a release note, be prepared for rather picky review comments about grammar, typos, and wording. This is for the sake of the Java community as a whole, as the language of the release note sets the tone for many blogs and news articles. For a widely used product like the JDK, the release notes are often copied verbatim (including typos) and published to highlight news in the release. This means that we need to take extra care to make sure the text in the release note is correct and follows a similar style.

The release note itself is written in a JBS sub-task of the issue that is used to push the change. There are a few steps to follow for the release note to find its way from JBS to the actual release note document.

1. Create a sub-task (More → Create Sub-Task) for the issue that requires a release note - the main issue, that is, the JBS issue that is used to push the original change, not for backports or the CSR (if there is one).

2. For the newly created sub-task, follow these steps:

   • The Summary should be a one sentence synopsis that is informative (and concise) enough to attract the attention of users, developers, and maintainers who might be impacted by the change. It should succinctly describe what has actually changed, not be the original bug title, nor describe the problem that was being solved. It should read well as a sub-section heading in a document.
   • Prefix the Summary with “Release Note:”.
   • Add the release-note label. This is required for the release note to be included in the release notes.
   • Add the proper RN-label if applicable to indicate what section of the release notes it should be included in (see RN-labels below).
   • Set the Assignee to the same person who owns the main issue.
   • Set Affects Version/s to the release versions for which the release note should be published.
   • Set the Fix Version/s to the same value that the main issue - in almost all cases this will be the version of mainline.
   • Enter the text of the release note in the Description field using markdown formatting, following the CommonMark specification. While the markdown won’t be rendered in JBS, you can use dingus to see what the release note will look like. Note that Github stlye ascii table formatting is supported but will not display correctly in the dingus page. For more information see General Conventions for Release Notes below.
   • While the Priority of the sub-task is set by default to be the same as the priority of the issue itself, it can be changed to adjust in what order the release note is listed compared to other release notes in the same build or release note section.

3. Have the release note ready to be reviewed at the same time as the code is reviewed. If it’s later determined that a release note is necessary, then go back to the same engineers who reviewed the fix to review the release note. Special care should be taken when writing a release note that will cover changes related to a vulnerability fix in order to avoid describing technical details of how it could have been exploited.

4. When you are done, Resolve the release note sub-task as Delivered. Only release notes where the sub-task has been resolved as Delivered is considered to be part of the EA/GA release notes. To avoid mixing up the release notes with the code fixes that have gone into a particular release or build, we don’t use Resolved/Fixed.

If you see an issue you feel should have a release note but you are not the assignee of the bug, then add the label release-note=yes to the main bug (not on a backport nor a sub-task). This acts as a flag to make sure that the release note is considered. This can be done even with fixes that have been shipped already if it’s noticed that there is confusion around the change. If, after discussion, it’s decided that a release note isn’t required either remove the label, or change it to release-note=no if it makes sense to have a clear indication that a release note isn’t required for the fix. The label release-note=yes can be removed once the release note sub-task has been created.

For examples of well written release note issues in JBS, see JDK-8276929 or JDK-8278458.

General conventions for release notes

The following are general practices that should be followed when creating release notes.

• Release notes should be no longer than 2-3 paragraphs.

• Don’t repeat information that will be included in updates to the docs, keep it to a high level summary or key changes.

• Note that where the changes are more fully documented in the JDK documentation, then refer to that document for details. When covering a change in behavior provide some idea to what can be done if a developer or user encounters problems from the change.

• Don’t include graphics etc. Refer to the main docs if there are more details that need explaining.

• Don’t include your name or affiliation, make sure however, you are the assignee of the release note sub-task.

• If you have a < in the Summary then use <. For <’s in the Description surround them by back-ticks.

• Avoid using Latin and abbreviations in the release note.

  • Use “also known as” instead of “aka”
  • Use “that is” or “to be specific” instead of “i.e.”
  • Use “for example” instead of “e.g.”

• The Summary should be in title case instead of sentence case.

  • Example: Decode Error with Tomcat Version 7.x

• The Description should be standardized to follow this pattern:

  • Sentence stating the change that was made
  • Background info/context
  • Example: A new system property, jdk.disableLastUsageTracking, has been introduced to disable JRE last usage tracking for a running VM.

Advanced options

• JEP release notes
  • Summary - If the change is an actual JEP, use the JEP title.
  • Description - the JEP Summary text have already been heavily reviewed and also approved by the project lead. It should be the first sentence in the release note description. That would be analogous to the “change that was made” sentence in other release note descriptions. The remaining text would be composed of the background info from the JEP.
  • Description - The JEP release note description should contain the link to the JEP.
• Single release note for multiple changes
  • A link to the parent issue that the note is a sub-task of, will be placed alongside the summary in the releasenotes. If note relates to additional changes, then add them as Relates links to the note and add the label RN-MultipleLinks - see JDK-8284975 as an example.
• Multiple release notes for the same change
  • If more than one release note is required for the same set of fixes, then open additional sub-tasks with the same Affects Version - see JDK-8073861 as an example.
• Release notes across backports
  • If an issue is backported to earlier releases the same note will be used - just add the new release version in the Affects Version field of the relese note.
  • Where a different release note is required, then create a separate note with the Affects Version for the new release - see JDK-8254631 as an example.

RN-labels

Unless labeled otherwise it will be assumed that the release note documents a change in behavior (will have likely required a CSR) or other item which should be included in the release notes. If the note covers a more specific type of change, then one of the following labels can be included (notes of a similar type will be listed together).

RN-NewFeature

A new feature or enhancement in the release. The Summary must be the item/API or new functionality. The Description must contain the name of the new feature, its intended function, and how a user can utilize it. Example: JDK-8315443

RN-IssueFixed

A significant issue which has been fixed. This would normally be a regression or an issue which was unknowingly released in a new feature. The Summary must be a summary of the error that was fixed. The Description must contain a statement about what was fixed, how the fix effects the user, and any special conditions that a user should be aware of regarding the fix. Example: JDK-8184172

RN-KnownIssue

An issue that wasn’t possible to fix by the time the release was GA’d. The Summary must be a summary of the error that the user sees. The Description must contain details about the error, how it effects the user, and workarounds if any exist. Example: JDK-8191040

RN-Removed

Only for major releases. The release note covers an API, feature, tool etc. which has been removed from the JDK. The Summary must be of the form “Removal of” Item/API. The Description must contain the list or name of the removed items/API with (optional) the reason for its removal. Include any special conditions that a user should be aware of regarding the removal. Example: JDK-8185066

RN-Deprecated

Only for major releases. The release notes cover an API, feature, tool etc. that has been marked as deprecated in the release. The Summary must be of the form “Deprecated” Item/API. The Description must contain the name of the item that has been deprecated, the reason for its deprecation, and (optional) any special conditions that a user should be aware of regarding the possible future removal. Example: JDK-8179909

RN-Important

Used to indicate that the release note should be highlighted in some fashion, such as listing it at the beginning of the release notes.

RN-(distro)

Used to indicate that the release note is only relevant for a specific JDK distribution. E.g. RN-Oracle

RN-MultipleLinks

Used to indicate that the release note should refer to multiple changes - see Advanced options section.

RN-Change

Deprecated. This is the default and no label is needed to indicate this.

Querying the release notes

The Release Notes for a particular release can be found using the JBS query

affectedversion = <version> and type = sub-task and labels = release-note

where <version> is the appropriate release value, e.g. 17.

The JDK Release Process

The JDK project has a well defined release process. JEP 3 describes this process in detail. This section intends to clarify some topics that often cause questions.

Release cycle

The release cycle starts when development of a new release begins, and ends when that release is delivered to the public. The current release cadence is six months. This means that every six months we start development of a new release, and every six months a new release is delivered. However, this doesn’t mean that each release cycle is six months. As described below, the total development time for a release (the release cycle) is actually nine months. Obviously this in turn doesn’t mean that all features are developed in nine months. Most features are developed for a much longer time than that, and goes through long time development in other project repositories, and through a series of preview and experimental stages. But any feature that is to be included in a specific release has a specific window of nine months to integrate the code into mainline and fix all the remaining bugs.

It may be tempting to integrate a new feature near the end of a release cycle, to get more time to fix all those last bugs before integration. Please don’t. If you are getting close to the end of a release and you still just have one more bug to fix, please defer your feature to the next release. It’s only six months out. Not only will this vouch for your new feature to be more stable on release, you will also help keeping the JDK as a whole more stable by allowing others to find and fix bugs in their new code that might come as a result of your changes.

Integrating early in a release is preferable, but all new features can’t be integrated at the same time. If many large changes enters the repository at the same time it will be more difficult to determine which change that caused all the new bugs. If you’re about to integrate a larger change you must therefore communicate this on the relevant mailing lists to synchronize with other projects that may also be planning to integrate something soon.

Milestones and phases

Throughout the release there are a number of milestones and phases that define where in the release cycle we are.

The start of a release

Since development is always ongoing in the mainline repository (openjdk/jdk), the start of a new release can be said to be when the former release is forked from mainline. After the start of the release follows six months of development to implement and integrate all the cool stuff that will go into the next release. After these six months ramp down begins.

Ramp Down Phase 1 (RDP1)

The ramp down of a release starts with a fork of the mainline repository. A clone of the entire code base is made and hosted in a separate ramp down repository (e.g. openjdk/jdk17). During the ramp down of a release we focus on bug fixing and stabilization in order to get the JDK ready for release. In RDP1 you may continue to fix P1-P3 product bugs (and some other issues) in the stabilization repo. For detailed information on what can be fixed when, see Push or defer during rampdown below. The start of RDP1 is essentially the deadline for integrating JEPs and enhancements into this particular release.

All Tests Run (ATR)

ATR is not a milestone described in JEP 3, but it’s still a concept that might be mentioned in discussions on this topic and is therefore good to know about. ATR (a.k.a. ATR Start) is the start of an approximately six week long test period where all tests in the test plan for the given release is ran. ATR usually starts at the same time as RDP1.

Ramp Down Phase 2 (RPD2)

In RDP2 the bar is higher to get changes into the release. For product bugs, only P1:s and P2:s are supposed to be fixed here, and to do so an approval is needed. See the Fix-Request Process for details on how to obtain one. All other product bugs should be deferred. See Push or defer during rampdown below for more details.

Release Candidate (RC)

Towards the end of the release cycle, when there are no more open product bugs targeted to the release, a stable build is selected to be the release candidate. This build will go through additional testing and if no more issues are found it will be the build released. If new bugs are found these are investigated and hopefully fixed, and a new build becomes the release candidate. The RC phase has a few milestones with a deadline for finding a candidate build, and another for making sure the build is ready to go live.

General Availability (GA)

This is the end of the release cycle. The last release candidate build is made available to the public.

Push or defer during rampdown

JEP 3 contains the exact definitions for what can be done when. This is a visualization of those definitions that may or may not clarify things.

Push and defer guidelines during rampdown

Deferring P1 and P2 bugs

Even though there’s nothing explicitly written in the process about deferring P1 and P2 bugs during the initial development phase, the assumption is that these aren’t deferred unless time runs out at the end of the release cycle.

Please note that the priority of a bug doesn’t change just because you want to get your fix in late in the release, or if you want to be able to defer it. The priority is based on the severity of the bug and if it was deemed to be a P2 before, you better have a really good explanation to why that conveniently has changed by the end of the release. Being hard to fix is not a reason to lower the priority of a bug.

Project Maintenance

Many OpenJDK projects build on top of the JDK source code for instance to produce new language features, like projects Amber and Valhalla. When doing this there are a number of common workflows that are dealt with by most project maintainers. For instance, updating the codebase (merging) to bring in the latest changes from the upstream JDK project.

Merging JDK mainline into a project repository

Merging changes from one git repository to another is basically the same thing as getting your own changes merged into the project repository, with the slight twist that you don’t write all the changes yourself, you just pull them from somewhere else.

In this example we’ll use a separate clone of the project repository to perform the merge in. This can be done using branches as well, but let’s keep it simple for now.

Init - done once

First set up your personal fork of the project repository, in this example called my-project. If you already are a contributor to the project you most likely have this set up. If not, see Cloning the JDK for details on how to do that.

git clone git@github.com:OpenDuke/my-project.git project-merge
cd project-merge
git remote add upstream git@github.com:openjdk/my-project.git
git remote add mainline git@github.com:openjdk/jdk.git

We clone the personal fork (in this case we clone OpenDuke’s personal fork) into a local directory, here called project-merge. We then set up two remotes, upstream and mainline.

Performing the merge

The clone we set up above is used each time you want to bring changes from mainline in to your project. This is done by first pulling the changes from mainline and then pushing to your personal fork. A regular PR will then be created which you can integrate into your main project repository. It sounds easy, and it is, but there are a few details below to keep in mind.

cd project-merge
git pull upstream master
git push
git switch -c Merge_mainline
git fetch mainline

We start by updating the local fork with the latest changes from the main project repository. Note that we then create a new branch “Merge_mainline” in which the merge will happen. Finally we fetch all new changes from mainline.

Merging from what ever is latest isn’t usually a good idea, mainline code is not “clean” for any given commit. Merging JDK tags ensures you have a known quality, those tagged commits are known to compile and pass tests. Therefore, next we check which tags have not been merged yet.

git tag -l "jdk-*" --no-merged

Or if you just want to see the latest tag you haven’t merged,

git tag -l "jdk-*" --no-merged | tail --lines 1

Before merging, you may want to check what’s incoming, to get an idea of the size of the merge and look for any incoming changes that you suspect may cause issues.

git log --topo-order --pretty=oneline --reverse ..$TAG

And finally we initiate the actual merge.

git merge $TAG

The commands above will likely run without a hitch up until the final git merge. This is where you need to combine the changes that were made in mainline with the changes that have been made in your project repository. If there are no conflicts you’re in luck, then the merge will be completely automated and you will end up with a committed merge. If there are conflicts however you’ll need to manually go through the files where the conflicts are and make sure you select the correct version for each change. Using git status you can see what files that need to be merged. Depending on how much code your project has touched, this can be quite a bit of work.

For complicated merges, see Sharing the work below.

Test before push

Regardless of if you encountered conflicts or not, you should always build and test your merge before pushing it to your project repository. Testing needs to be done even when there are no textual conflicts as changes like for instance a rename can result in a compile or test error without any conflict. One could argue that git merge --no-commit could be used and have logical errors fixed in the merge commit. However, a subsequent “Fix logical merge errors” commit, is in fact more useful, as it clearly shows the project specific adjustments needed for incoming changes.

It’s always okay to have further commits to clean up after a merge. Hiding a large amount of reworking project code to fit with upstream changes in a single merge commit will make it hard for further errors post integration to be identified.

The commit, push, and PR

Once you have a working version of your merged code you’re ready to create the merge commit and push. Please note that git commit is only needed if there were conflicts. If the changes were successfully merged by git merge, you already have a committed merge.

git commit -m "Merge"
git push --set-upstream origin Merge_mainline

Now it’s time to create the PR on GitHub. Just opening the PR page in your browser will most often be enough to see a message about new code that was pushed to your personal fork. Click the button to create the PR.

Make sure the PR title starts with “Merge”. You may have noticed that when you integrate a “normal” PR into an OpenJDK repository, all commits that have been done in that PR will be squashed into a single commit. For normal changes this is a good thing as each PR normally corresponds to a single JBS issue, but for a merge it would be highly undesirable to squash all the different commits that you pull in from mainline. A PR with a title that starts with “Merge” won’t be squashed. That means that all the changes that you brought over will remain separate changes.

It’s always a good idea to also include what was merged in the title of the PR. If you for instance is pulling in JDK mainline into your project repository it’s likely (because it’s in general a good idea) that you choose some stable EA tag in mainline to merge. Your PR title could then be something like “Merge jdk-21+2”.

Whether a merge requires a review or not is up to your project lead to decide. Many projects don’t require this so the GitHub bots will allow you to push the merge as soon as the GHAs are done. (They actually allow you to push even before the GHAs are done, but that’s in general not a good idea.)

Once the PR has been integrated, you can clean up your fork and its clone in preparation for the next merge.

git switch master
git branch -d Merge_mainline
git push -d origin Merge_mainline

These commands will remove the temporary branch that we created to perform the merge. There’s a button in the GitHub GUI to delete the branch after you have integrated the PR. This can be used instead of the last of the three commands above (git push -d...).

Sharing the work

When conflicts take place in areas requiring specialized knowledge you may need help from other contributors. Backing up the original conflicts will help if you find yourself “in too deep”, and need assistance from other contributors. You can add and later remove these backups, along with a readme describing the merge status, to the actual merge branch to aid communication (i.e. you may not be able to compile certain components).

Something like the following shell one-liner can be used to perform the backup.

git status | grep "both modified:" | while read B M FILE; do cp -v $FILE $DEST ; done

Below are two different methods of collaborating on a merge described. Please note that extra commits are fine. The merge PR itself will describe any special actions that were taken in case further failures turn up after merge integration. Ultimately these commits will be squashed when integrating the project back into mainline.

1. Parking a merge with conflicts in place

“Park” the conflicts, unresolved, in a personal fork, and let others do the further work (by sending you a patch, or opening your personal fork up to push from other contributors). Do this by keeping a list of unresolved conflicts (perhaps checking in said list to describe the merge state), and then marking them as resolved in git, committing, and pushing them to your personal fork. E.g. git add $UNRESOLVED_FILES; git commit; git push

Pros: All unresolved conflicts are stated and can be worked on by multiple parties, all at once.

Cons: Broken branch in terms of compile and test, may require temporary workaround patches to be passed around to complete work on specific unresolved components.

2. Incremental merging

An alternative to parking a merge with conflicts in place, is to incrementally merge up to the troublesome point. For example:

• Perform the initial merge: git merge $TAG
• Find yourself in trouble, identify which change is causing the issue.
• Abort: git merge --abort
• Find the troublesome change: git log --topo-order --pretty=oneline --reverse $(current_branch)..$TAG
• Merge up to the previous change, commit and push.
• Ask others to continue the merge from the troubled change forward, how far forward is up you of course, either just that troublesome change, or the rest of the merge up to the $TAG.
• Rinse and repeat: There may appear further conflicts requiring other contributors help.

Pros: All commits in the merge branch compile and test, you always have a working branch.

Cons: There is an unknown extra amount of merge work, multiple iterations create more work. For instance you may find yourself resolving the same files multiple times (e.g. back-out commits).

HotSpot Development

See Working With Pull Requests for generic guidance and requirements around pushing changes. For the HotSpot codebase there are a few additional requirements:

• Your change must have been approved by two reviewers out of which at least one is also a Reviewer
• Your change must have passed through HotSpot tier 1 testing with zero failures (See tier1 definition in test/hotspot/jtreg/TEST.groups.)

Logging

While developing your fix, you might want your code to output some diagnostic information. You might even want to leave some logging in the code you check in, to facilitate future diagnostics. The appropriate way to print logging output from HotSpot is through the Unified Logging Framework (JEP 158). It gives you a lot of nice features and enables common command-line options for all logging.

A basic log message can be output like this:

log_info(gc, marking)("Mark Stack Usage: " SIZE_FORMAT "M", _mark_stack_usage / M);

Where ‘gc’ and ‘marking’ are tags, and ‘info’ is the log level. Tags associate log messages with certain subsystems or features and the log level determines the importance and verbosity of the message. The most verbose output is trace, and the least is error. The full list of tags and levels are available via -Xlog:help.

The basic log API looks as follows:

log_<level>(Tag1[,...])(fmtstr, ...)

Sometimes single line printf-style logging isn’t enough. For example, it can be useful to group several log lines together or to use HotSpot’s outputstream API. UL supports both of these use cases using LogMessage and LogStream, respectively.

LogMessage(gc, marking) lm;
if (lm.is_info()) {
  lm.info("We are guaranteed to be");
  lm.info(" grouped together");
}

LogMessage will submit its output when it goes out of scope.

LogStream is typically used when a single printf-style format string becomes unwieldy.

LogStream st(Log(gc, marking)::info());
if (st.is_enabled()) {
  // Print without newline
  st.print("I'm printing a lot of %s ", "arguments");
  st.print("With a lot of extra info %d ", 3);
  // Print with newline (cr stands for carriage return)
  st.print_cr("and so it's useful to use a stream");
}

If you need to print multiple lines grouped together with complex formatting requirements then NonInterleavingLogStream is probably what you want.

LogMessage(gc) lm;
NonInterleavingLogStream st{LogLevelType::Info, lm};
if (st.is_enabled()) {
  st.print_cr("Line one: %d %d %d ", 1, 2, 3);
  st.print("Line two: %d %d %d", 4, 5, 6);
  st.print_cr(" still line two: %d %d %d", 7, 8, 9);
}

Enabling logging

You enable logging in the JVM by using the -Xlog command line option specified. For example, the messages from the examples would be visible if the JVM were run with any of the following options:

-Xlog:gc+marking=info
-Xlog:gc+marking
-Xlog:gc*

You can have multiple -Xlog options, these are applied in an additive manner. Consider this example:

-Xlog:gc+marking=info:stdout -Xlog:alloc=warning:stderr -Xlog:breakpoint=error:breakpoint.txt:level

This specifies that:

1. Log messages with info level and up, with tags gc and marking, to stdout.
2. Log messages with warning level and up, with tag alloc, to stderr.
3. Log messages with error level and up, with tag breakpoint, to file breakpoint.txt, with the decorator level.

UL automatically applies a default argument of -Xlog:all=warning:stdout:uptime,level,tags when logging is enabled. This can be disabled by prepending -Xlog:disable to your arguments.

-Xlog:disable -Xlog:gc+marking=info -Xlog:alloc=warning

Starting the JVM with the option -Xlog:help outputs more information and more examples.

A full description of the syntax of -Xlog is available in JEP 158.

Working With the Legacy Mercurial Servers

After the initial release of the JDK source code into OpenJDK in 2007 the OpenJDK project moved from TeamWare to using Mercurial. Starting in 2019 the source revision control has been moved to Git and GitHub. Even though most large projects have moved to Git by now, some still use the Mercurial servers. To access these projects some additional setup is required.

This document assumes familiarity with the first two chapters of the free on-line book Mercurial: The Definitive Guide.

Installing and configuring Mercurial

Source bundles and binary packages for Mercurial are available at https://www.selenic.com/mercurial/wiki/index.cgi. The OpenJDK repositories recommend installation of Mercurial 2.6.3 (or later). A Mercurial installation is sufficient to clone a repository. Contributors who wish to submit changes will need some additional configuration as described below.

Once Mercurial is installed, create and edit the ~/.hgrc file to minimally contain the following entry:

[ui]
username = <openjdk_username>

openjdk_username is in general the same as your GitHub user name. (See Contributing to an OpenJDK Project for more information.) If you don’t have a GitHub user name, you choose your OpenJDK user name when you sign the OCA. The user name should be a plain lowercase, alphanumeric token (not an e-mail address) with twelve characters or less. The first character should be alphabetic. This username will be publicly visible in all Mercurial changeset logs. It will be used to verify that the changeset author is at least an Author for the Project and that the person pushing the changeset is at least a Committer. It’s recommended that the openjdk_username be somehow related to the Author’s full name, such as the first character of the Author’s first name followed by the Author’s last name.

Some Projects may recommend additional tools or scripts that help with repository manipulation and code development. For instance, in JDK 8u, the utility script common/bin/hgforest.sh may be used to apply commands to all the repositories in the [forest]. Some useful Mercurial extensions for OpenJDK developers are jcheck, trees, and Mercurial Queues (mq). Note that trees is enabled on the OpenJDK Mercurial server.

Verifying the configuration

After installing and configuring Mercurial, validate the configuration using the following steps.

1. Verify that Mercurial is version 2.6.3 (or newer).

   $ hg version
   Mercurial Distributed SCM (version 2.9)
   (see http://mercurial.selenic.com for more information)
   
   Copyright (C) 2005-2014 Matt Mackall and others
   This is free software; see the source for copying conditions. There is NO
   warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

2. Verify that the list of enabled extensions includes fetch and mq.

   $ hg help

3. Verify that the ~/.hgrc configuration looks correct. Minimally it should contain the following entries:

   $ hg showconfig
   ui.username=iris

At this point, it should be possible to start retrieving source from the repositories.

Cloning a Mercurial repository

Some Projects organized their code into multiple Mercurial repositories. For instance, JDK 8 uses a forest of multiple related repositories which contain components of the entire JDK. If a Project uses a forest, It’s strongly recommended for developers to clone an entire forest, rather than a single repository. This is the only means to ensure consistency in builds. The following examples illustrate two alternatives for cloning the entire jdk8u/jdk8u-dev forest into the directory 8u-dev.

1. To clone the forest using the trees extension just use tclone:

   $ hg tclone http://hg.openjdk.java.net/jdk8u/jdk8u-dev/ 8u-dev

2. To clone the forest using get_source.sh, first clone the main tree:

   $ hg clone http://hg.openjdk.java.net/jdk8u/jdk8u-dev/ 8u-dev
   requesting all changes
   adding changesets
   adding manifests
   adding file changes
   added 997 changesets with 1477 changes to 138 files
   updating to branch default
   82 files updated, 0 files merged, 0 files removed, 0 files unresolved

   Then clone the repositories in the forest:

   $ cd 8u-dev
   $ sh ./get_source.sh

Regardless of how the forest was cloned, this is the resulting populated forest.

$ ls
ASSEMBLY_EXCEPTION  hotspot    LICENSE   README-builds.html
common              jaxp       make      test
configure           jaxws      Makefile  THIRD_PARTY_README
corba               jdk        nashorn
get_source.sh       langtools  README

Cloning a single repository

If the source for the Project is contained within a single repository or reading a limited portion of the source is the only goal, it’s possible to clone a single repository (even if it’s part of a forest). For instance, this example shows how to clone the langtools repository from jdk8u/jdk8u-dev into the default destination directory.

$ hg clone http://hg.openjdk.java.net/jdk8u/jdk8u-dev/langtools
destination directory: langtools
requesting all changes
adding changesets
adding manifests
adding file changes
added 2289 changesets with 21194 changes to 7004 files
updating to branch default
6212 files updated, 0 files merged, 0 files removed, 0 files unresolved

Creating a Mercurial changeset

The timing for creating a changeset is important. Creating the changeset long before it gets pushed into the parent repository may require complex merges. If a changeset is created before sufficient review or testing, a rollback may be required and a new changeset may be required to correct previous mistakes. The mq extension is recommended for managing changes before they become committed to a changeset.

In the examples below, the script common/bin/hgforest.sh can be used to apply the Mercurial command to all the repositories in the forest. So when you see hg, if you are dealing with one repository, just use “hg”, if it’s a forest, use “sh common/bin/hgforest.sh”.

Each repository in the forest is managed independently. After editing files in the individual cloned repositories of the forest, the hg status command may be used to see the changes in a single repository.

$ hg root
/u/iris/sandbox/box
$ hg status
? duke/images/DukeTubbingSmall.png
$ hg add duke/images/DukeTubbingSmall.png
$ hg status
A duke/images/DukeTubbingSmall.png

To see changes made to the repositories use hg status:

$ hg status
[.]
A duke/images/DukeTubbingSmall.png

In this example, a new file DukeTubbingSmall.png was added to a new subdirectory.

Formatting a changeset comment

A single change is described by a block of text of the following form:

<bugid>: <synopsis-of-symptom>
Summary: <summary-of-code-change>
Reviewed-by: <reviewer>+
Contributed-by: <contributor-email>

There may be more than one bugid line, but there must be at least one.

The summary line is optional, but authors are strongly encouraged to include one if the nature of the change isn’t obvious from the synopsis. It’s just one line, meant to give the reader a clue as to how the code changed. A more complete description of the change belongs in the bug report.

A reviewed-by line is required. Reviewers must have the ability to deal with any adverse consequences of the change, and so must themselves be authors. They are therefore identified by their OpenJDK usernames rather than full e-mail addresses.

The contributed-by line is optional. If present, it’s a list of comma-separated email addresses. It should be included only when the author of the change doesn’t have commit rights to the target repository and thus would not otherwise receive acknowledgment, or when there are multiple authors.

There will be exceptions for merge changesets, tag changesets, etc.

Example:

1234567: NPE thrown on FileInputStream("")
Summary: Rewrite precondition-checking code in io.c
Reviewed-by: mr
Contributed-by: Ben Bitdiddle <ben at bits.org>

If a changeset contains multiple unrelated changes (this is frowned upon, but may happen from time to time) then its comment will contain multiple blocks of the above form, separated by blank lines.

The required format of the comments will be enforced whenever the changeset is pushed into the JDK forests. Other Projects may copy these conventions, adopt some other conventions, or have no conventions, depending upon their goals.

Committing a changeset

The following commands commit all of the changes in a repository to a changeset.

$ cat ../message
1111111: Missing Duke gif
Summary:  Add missing file
Reviewed-by: iag
$ hg commit -l ../message
$ hg toutgoing
[.]
comparing with http://hg.openjdk.java.net/sandbox/box
searching for changes
changeset:   23:fb12953f3a35
tag:         tip
user:        iris
date:        Wed Dec 12 21:05:59 2007 -0800
summary:     1111111: Missing Duke gif

Merging Mercurial changesets

It’s often necessary to merge local changes with those made in the parent repositories. The first step in a merge process is to retrieve (or pull) the collection of changesets which have been pushed since the last merge or initial clone. If there if there are merge conflicts, then they must be resolved. Chapter 3 of the Mercurial book contains detailed information on the merging process.

There are two basic ways to update the working set files in the repositories:

Option 1: hg pull

One way to merge the parent repository with the working set of files is to use hg pull all by itself. This option allows merging off-line or at a later time.

$ hg pull
[.]
pulling from http://hg.openjdk.java.net/jdk8u/jdk8u-dev
searching for changes
no changes found

In Mercurial, pulling changesets will not update or merge into the working set of files. To update the clone, run hg update. If the update reports conflicts, run hg merge to resolve them.

Option 2: hg fetch

Alternatively, use hg fetch to pull the changes, update the working set files, and create simple merge changesets as necessary. The fetch extension is distributed with Mercurial but needs to be enabled. Edit the .hgrc to include the following entries:

[extensions]
fetch=

Once the fetch extension has been enabled, hg fetch may be invoked as follows:

$ hg fetch
[.]
pulling from http://hg.openjdk.java.net/jdk8u/jdk8u-dev
searching for changes
no changes found

---

Actual file merging will be done with the selected Mercurial merging tool see MergeProgram for the details on how to define the selected merge tool in ~/.hgrc.

---

Pushing Mercurial changesets

In order to push changesets into the parent repository, some additional configuration is required. The following sections describe the operations that will be performed by users with push access.

Get your SSH key installed

First you should create a new SSH key. See Generating an SSH key for guidance on how to do that. Your public key (~/.ssh/id_rsa.pub) should be mailed as an attachment along with your JDK username to keys@openjdk.org. An administrator will install your key on the server and notify you on completion. This process may take a couple of days.

Host *.openjdk.org
ProxyCommand /usr/lib/ssh/ssh-socks5-proxy-connect -h [socks_proxy_address] %h %p

Setting the default-push path to the server repositories

This is the typical development model:

Diagram of server repos and user's clone

Changesets need to be pushed via ssh to the read/write repository which resides on the OpenJDK Mercurial server. The easiest way to do this is to have each repository define the “default-push” path in every repository’s .hg/hgrc file. The .hg/hgrc file isn’t a managed file - it’s private to the repository. The following example defines the “default” and “default-push” paths for clones of the jdk8u/jdk8u-dev repository.

[paths]
default = http://hg.openjdk.java.net/jdk8u/jdk8u-dev
default-push = ssh://<JDK_username>@hg.openjdk.java.net/jdk8u/jdk8u-dev

Given a JDK_username this simple script will attempt to do this for all the repositories:

#!/bin/sh
username=$1
hgdirs="`find . -type d -name .hg`"
for i in ${hgdirs}; do
  d="`dirname ${i}`"
  defpush="`(cd ${d} && hg paths default-push 2> /dev/null)`"
  if [ "${defpush}" = "" ] ; then
    defpath="`(cd ${d} && hg paths default 2> /dev/null)`"
    if [ "${defpath}" != "" ] ; then
      defpush="`echo ${defpath} | sed -e 's@http://\([^/]*/[^/]*/[^/]*\)/\(.*\)@ssh://$username\@\1/\2@'`"
      cp ${i}/hgrc ${i}/hgrc.orig
      echo "default-push = ${defpush}" >> ${i}/hgrc
      echo "Added default-push: ${defpush}"
    fi
  fi
done
for i in ${hgdirs}; do
  d="`dirname ${i}`"
  echo "(cd ${d} && hg paths)"
  (cd ${d} && hg paths)
done
exit 0

Performing the push

Committers can use the hg push command to propagate changesets into the repositories.

Most developers will only find a need to create changesets in one or two repositories. However, it’s important that before any changesets are pushed, the corresponding forest pull and merge with the destination forest be performed; otherwise there is a risk of breaking the build.

$ hg push

After the push has been accepted, an automatic e-mail notification will be sent to the mailing list associated with the repository. In most cases notifications are sent to the Project’s -dev mailing list. Some Projects with high traffic -dev mailing lists use a dedicated -changes list for notifications.

In general all of a Project’s Committers can push to all of the the Project’s repositories. However, some Projects may chose to restrict the set of Committers with push to key repositories. For instance, JDK Release Projects restrict push access to MASTER repositories to Committers who are either integrators or members of the Release Engineering Team. See Becoming a Committer for information about becoming a Project Committer.

Code Owners

This list is intended to make it easier to identify which email list to include in code reviews when making changes in different areas. The list may also help when assigning bugs based on which code they are found in. Please note that some directories may have been created or removed between releases. The intention here is to include directories that exists in mainline, LTS releases and other releases (post JDK 9) commonly being updated.

Area mailing lists

• Generic JDK Development: jdk-dev

• Build: build-dev

• Client Libs: client-libs-dev

• Core Libs: core-libs-dev

  • Net: net-dev
  • NIO: nio-dev
  • I18n: i18n-dev

• HotSpot: hotspot-dev

  • Compiler: hotspot-compiler-dev
  • GC: hotspot-gc-dev
  • Runtime: hotspot-runtime-dev
  • JFR: hotspot-jfr-dev
  • Serviceability: serviceability-dev

• Javac: compiler-dev

• Security: security-dev

• Tools

  • Javadoc: javadoc-dev
  • JShell: kulla-dev
  • Nashorn: nashorn-dev

Directory to area mapping

• .jcheck – Build
• bin – Build
• doc –
• make – Build
• src
  • demo – Client Libs
  • hotspot
    • cpu – Compiler, Runtime
    • os – Runtime
    • os_cpu – Compiler
    • share
      • adlc – Compiler
      • asm – Runtime
      • c1 – Compiler
      • cds – Runtime
      • ci – Compiler
      • classfile – Runtime
      • code – Compiler
      • compiler – Compiler
      • gc – GC
      • include – HotSpot
      • interpreter – Runtime
      • jfr – JFR
      • jvmci – Compiler
      • libadt – Compiler
      • logging – Runtime
      • memory – Runtime, GC
      • metaprogramming – HotSpot
      • oops – GC, Runtime
      • opto – Compiler
      • precompiled – HotSpot
      • prims – Runtime, Serviceability
      • runtime – Runtime
      • services – Runtime
      • utilities – GC, Runtime
  • java.base
    • Core Libs should almost always be included but Java Language, HotSpot, Security and/or I18n may also be involved.
    • */classes
      • */crypto – Security
      • */foreign – Core Libs
      • */invoke – Core Libs
      • */io – NIO
      • */jmod – HotSpot
      • */lang – Core Libs
      • */math – Core Libs
      • */misc – Core Libs, HotSpot
      • */module – Core Libs
      • */net – Net
      • */nio – NIO
      • */ref – Core Libs, GC
      • */reflect – Core Libs
      • */security – Security
      • */text – I18n
      • */time – Core Libs
      • */util – I18n, Core Libs
      • jdk/internal
        • access – Core Libs, Security
        • event – JFR
        • icu – Core Libs
        • javac – Java Language
        • jimage – Client Libs
        • jrtfs – Core Libs
        • loader – Core Libs
        • logger – Core Libs
        • org – Core Libs
        • perf – Runtime
        • platform – HotSpot
        • vm – HotSpot
      • sun/launcher – Generic List
      • META-INF – Core Libs
    • */conf
      • sdp – Net
      • security – Security
    • */legal –
    • */lib/security – Security
    • */native
      • common –
      • include – Runtime, Core Libs
      • jspawnhelper – Generic List
      • launcher – Generic List
      • libjava – Core Libs
      • libjimage – Core Libs
      • libjli – Generic List
      • libjsig – HotSpot
      • libnet – Net
      • libnio – NIO
      • libosxsecurity – Security
      • libsyslookup – Core Libs
      • libverify – Runtime
      • libzip – Core Libs
    • share/data
      • blockedcertsconverter – Security
      • cacerts – Security
      • currency – I18n
      • lsrdata – I18n
      • publicsuffixlist – Client Libs
      • tzdata – I18n
      • unicodedata – I18n
    • share/man
      • java.1 – Generic List
      • keytool.1 – Security
  • java.compiler – Javac
  • java.datatransfer – Client Libs
  • java.desktop – Client Libs
  • java.instrument – Serviceability
  • java.logging – Core Libs
  • java.management – Serviceability
  • java.management.rmi – Serviceability
  • java.naming – Core Libs
  • java.net.http – Net
  • java.prefs – Core Libs
  • java.rmi – Core Libs
  • java.scripting – Generic List
  • java.se – Core Libs
  • java.security.jgss – Security
  • java.security.sasl – Security
  • java.smartcardio – Security
  • java.sql – Core Libs
  • java.sql.rowset – Core Libs
  • java.transaction.xa – Core Libs
  • java.xml – Core Libs
  • java.xml.crypto – Security
  • jdk.accessibility – Client Libs
  • jdk.attach – Serviceability
  • jdk.charsets – I18n, Core Libs
  • jdk.compiler – Javac
  • jdk.crypto.cryptoki – Security
  • jdk.crypto.ec – Security
  • jdk.crypto.mscapi – Security
  • jdk.dynalink – Nashorn
  • jdk.editpad – JShell
  • jdk.hotspot.agent – Serviceability
  • jdk.httpserver – Net
  • jdk.incubator.concurrent – Core Libs
  • jdk.incubator.vector – Compiler
  • jdk.internal.ed – JShell
  • jdk.internal.jvmstat – Serviceability
  • jdk.internal.le – JShell
  • jdk.internal.opt – Generic List
  • jdk.internal.vm.ci – Compiler
  • jdk.internal.vm.compiler – Compiler
  • jdk.internal.vm.compiler.management – Compiler
  • jdk.jartool – Core Libs, Security
  • jdk.javadoc – Javadoc
  • jdk.jcmd – Serviceability
  • jdk.jconsole – Serviceability
  • jdk.jdeps – Core Libs
  • jdk.jdi – Serviceability
  • jdk.jdwp.agent – Serviceability
  • jdk.jfr – JFR
  • jdk.jlink – Generic List
  • jdk.jpackage – Core Libs
  • jdk.jshell – JShell
  • jdk.jsobject – Nashorn
  • jdk.jstatd – Serviceability
  • jdk.localedata – I18n
  • jdk.management – Serviceability
  • jdk.management.agent – Serviceability
  • jdk.management.jfr – Runtime
  • jdk.naming.dns – Core Libs
  • jdk.naming.rmi – Core Libs
  • jdk.net – Net
  • jdk.nio.mapmode – NIO
  • jdk.random – Core Libs
  • jdk.sctp – Net
  • jdk.security.auth – Security
  • jdk.security.jgss – Security
  • jdk.unsupported – Core Libs
  • jdk.unsupported.desktop – Client Libs
  • jdk.xml.dom – Core Libs
  • jdk.zipfs – Core Libs
  • utils –
• test
  • The test directories follow to a large part the same structure as the source code in src. The owners are the same for directories with the same names.

Directories removed

• hotspot
  • *.jdk – Compiler (Removed in 10)
  • share
    • aot – Compiler (Removed in 17)
    • shark – Compiler (Removed in 10)
    • trace – Runtime (Removed in 11)
• java.base
  • man
    • jfr.1 – Runtime (Removed in 16)
• jdk.aot – Compiler (Removed in 17)
• jdk.crypto.ucrypto – (Removed in 12)
  • only available on Solaris
• jdk.pack – Generic List (Removed in 14)
• jdk.rmic – Core Libs (Removed in 15)
• jdk.scripting.nashorn – Nashorn (Removed in 15)
• jdk.scripting.nashorn.shell – Nashorn (Removed in 15)

About This Guide

This guide is being maintained through the OpenJDK Developers’ Guide Project. The source repository is available at GitHub. The revision hash at the bottom of this page refers to the last published commit.

Comments and questions may be sent to guide-dev@openjdk.org. Please let us know if there’s anything in the guide that isn’t clear.

Glossary

provisional (in the CSR)

The stage of the CSR process after “DRAFT”, and “PROPOSED”. At this stage the primary goals are to ensure that the proposed changes are suitable for the release in a general sense.

approved (by the CSR)

The stage of the CSR process after “FINALIZED”. The CSR has approved the final version of the request which permits push into the project repository.

changeset

A collection of changes with respect to the current clone of a repository.

trivial

A change that is small, well contained, and that makes no semantic changes. Typical examples are fixing obvious typos or renaming a local identifier. A trivial change can also be pushing an already-reviewed change that was missed in an earlier push (e.g., forgot to add a file) or generated changes like a git revert. It’s up to the author of a change to claim that the change is trivial in the RFR, and it’s up to the Reviewer whether to approve such a claim. A change is trivial only if the Reviewer agrees that it is. A trivial change doesn’t need to wait 24 hours before being pushed, and it only needs one Reviewer, even in areas where stricter rules for pushing normally apply.