/pom-scijava-base

Low-level base POM for all SciJava-based software

Primary LanguageShell

The pom-scijava-base project is a Maven POM that serves as the base for all Maven-based SciJava software, including:

Fiji ImageJ ImgLib2 KNIME LOCI SCIFIO SciJava SLIM Curve Virtual Cell
Fiji ImageJ ImgLib2 KNIME LOCI SCIFIO SciJava SLIM Curve Virtual Cell

pom-scijava-base vs. pom-scijava

It is encouraged that you not extend this POM directly, and instead use pom-scijava as parent.

pom-scijava-base pom-scijava
"Low level" base POM, without dependency version management. Extend pom-scijava-base only if you are a Maven expert, and have good reasons for doing so. Friendly base POM for SciJava software, including dependency version management. Extend pom-scijava to inherit the unified SciJava Bill of Materials: component versions which have been tested to work together.

See these examples for guidance:

Enforcer rules declared in this parent

The pom-scijava-base parent POM declares several enforcer rules which we believe make SciJava-based projects more reproducible and more consistent:

  • Plugin versions. Out of the box, Maven does not require plugins used to declare a version. But plugin versions must be declared to ensure reproducible builds. Otherwise, the version of Maven core you use at build time will determine which plugin versions are used, and the behavior might differ between builds.

  • No duplicate classes. If two dependencies define the same class, then it introduces the possibility of serious class-loading issues. Which version of the class should be chosen? In some scenarios, classes from one part of a certain library may be loaded from dependency foo, but classes from a different part of that same library at a different version may be loaded from dependency bar. When this happens, difficult-to-understand compiler errors, or even runtime errors, may occur. Best practice is to ensure that all classes come onto the classpath from exactly one source.

  • No too-new dependencies. When a project is compiled for Java version X, then it may not use any dependencies which require a version newer than X. Otherwise, your project is lying about needing only version X.

  • No circular dependencies. Actually, this is a central rule of Maven. But we configure the Enforcer to explicitly check for it, just to be safe. It is always possible to avoid circular dependencies; if you feel like you need one, you should instead solve it in one of the following ways, depending on how much code is co-dependent:

    1. Combine the co-dependent artifacts into a single artifact.
    2. Introduce a third artifact to house the co-dependent code, which depends on the other two artifacts.
    3. If all else fails, write to the SciJava list for help.
  • Reproducible builds. This rule means no SNAPSHOT dependencies, no SNAPSHOT parents, and no SNAPSHOT plugin versions. A snapshot version is not immutable, which means that code which depends on a snapshot may build today, but not build tomorrow, if the snapshot is later changed. The best way to avoid this conundrum is to never depend on SNAPSHOT versions. Snapshot are best used for testing only; they can be used transiently, but their use should never make it onto the main integration branch (i.e., master) of a project. See also Using snapshot couplings during development.

  • Developer roles. SciJava-based projects define developers and contributors with roles matching the SciJava team roles. Doing this is vital for consistency, and for communicating expectations to the community. By being careful about which developers are pledging which sorts of responsibility, the social status of each project becomes much clearer, and which social actions to take in various circumstances becomes a more tractable problem. We have automated tooling which generates MediaWiki sidebar templates for all components of the ImageJ software stack; this tooling requires SciJava developer roles to be present for sensible results.

  • Required metadata. Every SciJava-based project must override key pieces of metadata, including the name, description, url, inceptionYear, organization, licenses, developers, contributors, mailingLists, scm, issueManagement and ciManagement elements, as well as the license.licenseName and license.copyrightOwners properties. There are several reasons for requiring these overrides:

    • Avoid inadvertent inheritance. The pom-scijava-base POM itself declares all of this metadata for itself (e.g., its <scm> block defines where the pom-scijava-base source code is managed, and this information is necessary for the tooling which cuts releases of the pom-scijava-base POM itself). For better and worse, when extending pom-scijava-base, the child POM inherits all of these elements (except for <name>, but that is the sole exception in the above list). If the child POM does not override each and every one of these elements, then it will inadvertently inherit the incorrect values from the pom-scijava-base parent. Furthermore, due to a quirk/limitation in Maven, if you specify an empty block (e.g., <contributors />), then the non-empty value from the ancestor will take precedence in the interpolated POM. Hence, we enforce that all of these fields are overridden with non-empty values. See below for advice on how to best override specific metadata fields.
    • Present the project's metadata simply and clearly. For humans, being able to look at a project POM and clearly see the metadata is very helpful for understanding the project. Whereas when inheritance is involved, the human must be patient enough to dig through the ancestor POMs manually looking for the information, or else knowledgeable enough to know that they should actually use mvn help:effective-pom and check the metadata there instead, to know the actual values.
    • Make the metadata easier for tooling to consume. Maven-based tooling which uses the interpolated POM will be able to extract the correct inherited metadata from a POM, sure. But in many cases, it is much simpler and more natural, especially for Maven non-experts, to code tooling using shell scripts and similar approaches. In those cases, it is much easier if the tooling can simply extract the metadata from the child POM itself and be guaranteed that the values there are the correct ones, without needing to recurse into parent POMs whose contents may be less trivial to inspect.
    • Make it easier to maintain license headers in the sources. Putting a license blurb at the top of each source file is legal best practice, but it is undeniably a hassle, which is one reason many projects do not bother. But with the license-maven-plugin, generating and maintaining these license headers becomes very easy—as long as the inceptionYear, license.projectLicense and license.copyrightOwners values are provided in the POM. At that point, you can just invoke mvn license:update-file-header license:update-project-license and your work is done.
    • Encourage responsible metadata curation. As a project maintainer, you are responsible for your project's metadata. Yes, it is a hassle to specify it. But regardless, you need to understand where (if anywhere) your project lives in SCM, which (if any) system is being used to automatically build it, and so on. You have legal and social obligations to clearly communicate the project license, to clearly document community expectations, to give credit where credit is due, etc.

The full set of Enforcer rules as of pom-scijava-base version 3.0.0 can be seen here.

How to override a field with an "empty" value

For some projects, you may have "empty" metadata fields, and you may be unsure how best to override those values accordingly. The most common scenarios are:

  • If your project has no contributors, write:
    <contributors>
      <!--
      NB: Need at least one element to override the parent.
      See: https://issues.apache.org/jira/browse/MNG-5220
      -->
      <contributor>
        <name>None</name>
      </contributor>
    </contributors>
  • If your project has no discussion forum or mailing list, write:
    <mailingLists>
      <mailingList>
        <name>None</name>
      </mailingList>
    </mailingLists>
    But actually, you are warmly welcome to use the ImageJ Forum for discussing your project, so instead it is better to write:
    <mailingLists>
      <mailingList>
        <name>ImageJ Forum</name>
        <archive>http://forum.imagej.net/</archive>
      </mailingList>
    </mailingLists>
  • If your project has no CI, write:
    <ciManagement>
      <system>None</system>
    </ciManagement>
  • If your project has no issue tracker, write:
    <issueManagement>
      <system>None</system>
    </issueManagement>
  • If your project does not live in SCM, then write:
    <scm>
      <system>None</system>
    </scm>
    But as an aside, in this case, we strongly encourage you to adopt an SCM; check yourself before you wreck yourself.

Getting help with Maven

For more information about Maven, see: