Developing Calcite

Want to help add a feature or fix a bug?

Source code

You can get the source code by downloading a release or from source control.

Calcite uses git for version control. The canonical source is in Apache, but most people find the Github mirror more user-friendly.

Download source, build, and run tests

Prerequisites are Git, and Java (JDK 8u220 or later, 11 preferred) on your path.

Note: early OpenJDK 1.8 versions (e.g. versions before 1.8u202) are known to have issues with producing bytecode for type annotations (see JDK-8187805, JDK-8187805, JDK-8210273, JDK-8160928, JDK-8144185 ), so make sure you use up to date Java.

Create a local copy of the Git repository, cd to its root directory, then build using Gradle:

$ git clone https://github.com/apache/calcite.git
$ cd calcite
$ ./gradlew build

The HOWTO describes how to build from a source distribution, set up an IDE for contributing, run more or fewer tests and run integration tests.

JIRA accounts

Calcite uses JIRA for issues/case management. You must have a JIRA account in order to log cases and issues.

I already have an ASF JIRA account and want to be added as a contributor

If you already have an ASF JIRA account, you do not need to sign up for a new account. Please email jira-requests@calcite.apache.org using the following template, so that we can add your account to the contributors list in JIRA:

[Open the template in your email client]

Subject: Add me as a contributor to JIRA

Hello,

Please add me as a contributor to JIRA.
My JIRA username is: [INSERT YOUR JIRA USERNAME HERE]

Thanks,
[INSERT YOUR NAME HERE]

I do not have an ASF JIRA account, want to request an account and be added as a contributor

Please request an account using the ASF’s self-serve facility.

Contributing

We welcome contributions.

If you are planning to make a large contribution, talk to us first! It helps to agree on the general approach. Log a JIRA case for your proposed feature or start a discussion on the dev list.

Before opening up a new JIRA case, have a look in the existing issues. The feature or bug that you plan to work on may already be there.

If a new issue needs to be created, it is important to provide a concise and meaningful summary line. It should imply what the end user was trying to do, in which component, and what symptoms were seen. If it’s not clear what the desired behavior is, rephrase: e.g., “Validator closes model file” to “Validator should not close model file”.

Contributors to the case should feel free to rephrase and clarify the summary line. If you remove information while clarifying, put it in the description of the case.

Design discussions may happen in various places (email threads, Github reviews) but the JIRA case is the canonical place for those discussions. Link to them or summarize them in the case.

When implementing a case, especially a new feature, make sure the case includes a functional specification of the change. For instance, “Add a IF NOT EXISTS clause to the CREATE TABLE command; the command is a no-op if the table already exists.” Update the description if the specification changes during design discussions or implementation.

When implementing a feature or fixing a bug, endeavor to create the jira case before you start work on the code. This gives others the opportunity to shape the feature before you have gone too far down (what the reviewer considers to be) the wrong path.

The best place to ask for feedback related to an issue is the developers list. Please avoid tagging specific people in the JIRA case asking for feedback. This discourages other contributors to participate in the discussion and provide valuable feedback.

If there is a regression that seems to be related with a particular commit, feel free to tag the respective contributor(s) in the discussion.

If you are going to take on the issue right away assign it to yourself. To assign issues to yourself you have to be registered in JIRA as a contributor. In order to do that, please follow the instructions outlined in the JIRA Accounts section.

If you are committed to fixing the issue before the upcoming release set the fix version accordingly (e.g., 1.20.0), otherwise leave it as blank.

If you pick up an existing issue, mark it ‘in progress’, and when it’s finished flag it with ‘pull-request-available’.

If for any reason you decide that an issue cannot go into the ongoing release, reset the fix version to blank.

During a release, the release manager will update the issues that were not completed for the current release to the next release.

There are cases where the JIRA issue may be solved in the discussion (or some other reason) without necessitating a change. In such cases, the contributor(s) involved in the discussion should:

  • resolve the issue (do not close it);
  • select the appropriate resolution cause (“Duplicate”, “Invalid”, “Won’t fix”, etc.);
  • add a comment with the reasoning if that’s not obvious.

Fork the GitHub repository, and create a branch for your feature.

Develop your feature and test cases, and make sure that ./gradlew build succeeds. (Run extra tests if your change warrants it.)

Commit your change to your branch, and use a comment that starts with the JIRA case number, like this:

[CALCITE-345] AssertionError in RexToLixTranslator comparing to date literal

If your change had multiple commits, use git rebase -i main to squash them into a single commit, and to bring your code up to date with the latest on the main line.

In order to keep the commit history clean and uniform, you should respect the following guidelines.

  • Read the messages of previous commits, and follow their style.
  • The first line of the commit message must be a concise and useful description of the change.
  • The message is often, but not always, the same as the JIRA subject. If the JIRA subject is not clear, change it (perhaps move the original subject to the description of the JIRA case, if it clarifies).
  • Leave a single space character after the JIRA id.
  • Start with a capital letter.
  • Do not finish with a period.
  • Use imperative mood (“Add a handler …”) rather than past tense (“Added a handler …”) or present tense (“Adds a handler …”).
  • If possible, describe the user-visible behavior that you changed (“FooCommand now creates directory if it does not exist”), rather than the implementation (“Add handler for FileNotFound”).
  • If you are fixing a bug, it is sufficient to describe the bug (“NullPointerException if user is unknown”) and people will correctly surmise that the purpose of your change is to fix the bug.

Then push your commit(s) to GitHub, and create a pull request from your branch to the calcite main branch. Update the JIRA case to reference your pull request, and a committer will review your changes.

The pull request may need to be updated (after its submission) for three main reasons:

  1. you identified a problem after the submission of the pull request;
  2. the reviewer requested further changes;
  3. the CI build failed, and the failure is not caused by your changes.

In order to update the pull request, you need to commit the changes in your branch and then push the commit(s) to GitHub. You are encouraged to use regular (non-rebased) commits on top of previously existing ones.

When pushing the changes to GitHub, you should refrain from using the --force parameter and its alternatives. You may choose to force push your changes under certain conditions:

  • the pull request has been submitted less than 10 minutes ago and there is no pending discussion (in the PR and/or in JIRA) concerning it;
  • a reviewer has explicitly asked you to perform some modifications that require the use of the --force option.

In the special case, that the CI build failed, and the failure is not caused by your changes create an empty commit (git commit --allow-empty) and push it.

Null safety

Apache Calcite uses the Checker Framework to avoid unexpected NullPointerExceptions. You might find a detailed documentation at https://checkerframework.org/

Note: only main code is verified for now, so nullness annotation is not enforced in test code.

To execute the Checker Framework locally please use the following command:

./gradlew -PenableCheckerframework :linq4j:classes :core:classes

Here’s a small introduction to null-safe programming:

  • By default, parameters, return values and fields are non-nullable, so refrain from using @NonNull
  • Local variables infer nullness from the expression, so you can write Object v = ... instead of @Nullable Object v = ...
  • Avoid the use of javax.annotation.* annotations. The annotations from jsr305 do not support cases like List<@Nullable String> so it is better to stick with org.checkerframework.checker.nullness.qual.Nullable. Unfortunately, Guava (as of 29-jre) has both jsr305 and checker-qual dependencies at the same time, so you might want to configure your IDE to exclude javax.annotation.* annotations from code completion.

  • The Checker Framework verifies code method by method. That means, it can’t account for method execution order. That is why @Nullable fields should be verified in each method where they are used. If you split logic into multiple methods, you might want verify null once, then pass it via non-nullable parameters. For fields that start as null and become non-null later, use @MonotonicNonNull. For fields that have already been checked against null, use @RequiresNonNull.

  • If you are absolutely sure the value is non-null, you might use org.apache.calcite.linq4j.Nullness.castNonNull(T). The intention behind castNonNull is like trustMeThisIsNeverNullHoweverTheVerifierCantTellYet(...)

  • If the expression is nullable, however, you need to pass it to a non-null method, use Objects.requireNonNull. It allows to have a better error message that includes context information.

  • The Checker Framework comes with an annotated JDK, however, there might be invalid annotations. In that cases, stub files can be placed to /src/main/config/checkerframework to override the annotations. It is important the files have .astub extension otherwise they will be ignored.

  • In array types, a type annotation appears immediately before the type component (either the array or the array component) it refers to. This is explained in the Java Language Specification.

      String nonNullable;
      @Nullable String nullable;
    
      java.lang.@Nullable String fullyQualifiedNullable;
    
      // array and elements: non-nullable
      String[] x;
    
      // array: nullable, elements: non-nullable
      String @Nullable [] x;
    
      // array: non-nullable, elements: nullable
      @Nullable String[] x;
    
      // array: nullable, elements: nullable
      @Nullable String @Nullable [] x;
    
      // arrays: nullable, elements: nullable
      // x: non-nullable
      // x[0]: non-nullable
      // x[0][0]: nullable
      @Nullable String[][] x;
    
      // x: nullable
      // x[0]: non-nullable
      // x[0][0]: non-nullable
      String @Nullable [][] x;
    
      // x: non-nullable
      // x[0]: nullable
      // x[0][0]: non-nullable
      String[] @Nullable [] x;
    
  • By default, generic parameters can be both nullable and non-nullable:

      class Holder<T> { // can be both nullable
          final T value;
          T get() {
              return value; // works
          }
          int hashCode() {
              return value.hashCode(); // error here since T can be nullable
          }
    
  • However, default bounds are non-nullable, so if you write <T extends Number>, then it is the same as <T extends @NonNull Number>.

      class Holder<T extends Number> { // note how this T never permits nulls
          final T value;
          Holder(T value) {
              this.value = value;
          }
          static <T> Holder<T> empty() {
              return new Holder<>(null); // fails since T must be non-nullable
          }
    
  • If you need “either nullable or non-nullable Number”, then use <T extends @Nullable Number>,

  • If you need to ensure the type is always nullable, then use <@Nullable T> as follows:

      class Holder<@Nullable T> { // note how this requires T to always be nullable
          protected T get() { // Default implementation.
              // Default implementation returns null, so it requires that T must always be nullable
              return null;
          }
          static void useHolder() {
              // T is declared as <@Nullable T>, so Holder<String> would not compile
              Holder<@Nullable String> holder = ...;
              String value = holder.get();
          }
    

Continuous integration testing

Calcite exploits GitHub actions for continuous integration testing.

Getting started

Calcite is a community, so the first step to joining the project is to introduce yourself. Join the developers list and send an email.

If you have the chance to attend a meetup, or meet members of the community at a conference, that’s also great.

Choose an initial task to work on. It should be something really simple, such as a bug fix or a Jira task that we have labeled “newbie”. Follow the contributing guidelines to get your change committed.

We value all contributions that help to build a vibrant community, not just code. You can contribute by testing the code, helping verify a release, writing documentation, improving the web site, or just by answering questions on the list.

After you have made several useful contributions we may invite you to become a committer.