jgo 0.3.0

Launch Java code from Python and the CLI, installation-free.

jgo: painless Java component execution

Summary

Maven is a great tool. It manages dependencies so that Java projects become reusable "building blocks" in a much more robust way than many other languages offer. And the Maven Central repository contains a tremendous wealth of code, ripe for reuse in your own projects.

But shockingly, Maven provides no easy way to actually launch code from the beautifully managed dependencies stored so lovingly into ~/.m2/repository.

This project fills that gap: jgo launches Java code. You do not need to download or install any JARs; you just specify an "endpoint" consisting of a Maven artifact identifier, plus a main class if needed/desired, and jgo uses Maven to obtain and run it.

Installation

There are two implementations from which to choose! Each has pros and cons.

Prerequisites

jgo uses mvn and java for the heavy lifting. The shell script version needs some common utilities (e.g., cat). If you are missing anything, the script will tell you.

The shell script

The jgo.sh shell script requires a POSIX-friendly system. It is known to work on Linux, macOS, Cygwin, Microsoft's Windows Subsystem for Linux, and MinGW via the Git for Windows project.

Installing the shell script

Just clone this repo and symlink jgo.sh into your favorite bin directory.

For example, assuming ~/bin is on your PATH:

cd
git clone https://github.com/scijava/jgo
cd bin
ln -s ../jgo/jgo.sh jgo
jgo --help

The Python module

The jgo/jgo.py module requires Python. It offers a jgo console script, as well as a jgo module for programmatically creating endpoints.

Installing with pip

pip install jgo

Installing with conda

conda install -c conda-forge jgo

Installing from source

git clone https://github.com/scijava/jgo
cd jgo

# install globally (not recommended unless using conda or other virtual environment)
pip install .

# install into $HOME/.local (see pip install --help for details)
pip install --user .

# install into $PREFIX
pip install --prefix=$PREFIX .

Usage

Usage: jgo [-v] [-u] [-U] [-m] <jvm-args> <endpoint> <main-args>

  -v          : verbose mode flag
  -u          : update/regenerate cached environment
  -U          : force update from remote Maven repositories (implies -u)
  -m          : use endpoints for dependency management (see "Details" below)
  <jvm-args>  : any list of arguments to the JVM
  <endpoint>  : the artifact(s) + main class to execute
  <main-args> : any list of arguments to the main class

The endpoint should have one of the following formats:

- groupId:artifactId
- groupId:artifactId:version
- groupId:artifactId:mainClass
- groupId:artifactId:version:mainClass
- groupId:artifactId:version:classifier:mainClass

If version is omitted, then RELEASE is used.
If mainClass is omitted, it is auto-detected.
You can also write part of a class beginning with an @ sign,
and it will be auto-completed.

Multiple artifacts can be concatenated with pluses,
and all of them will be included on the classpath.
However, you should not specify multiple main classes.

Examples

Program	Command
Jython REPL	jgo org.python:jython-standalone
JRuby eval	echo "puts 'Hello Ruby'" | jgo org.jruby:jruby-complete:@jruby.Main
Groovy REPL	jgo org.codehaus.groovy:groovy-groovysh:@shell.Main+commons-cli:commons-cli:1.3.1
SciJava REPL with JRuby	jgo org.scijava:scijava-common:@ScriptREPL+org.scijava:scripting-jruby
SciJava REPL with Jython	jgo org.scijava:scijava-common:@ScriptREPL+org.scijava:scripting-jython
SciJava REPL with Groovy	jgo org.scijava:scijava-common:@ScriptREPL+org.scijava:scripting-groovy
SciJava REPL with Clojure	jgo org.scijava:scijava-common:@ScriptREPL+org.scijava:scripting-clojure
SciJava REPL with JavaScript	jgo org.scijava:scijava-common:@ScriptREPL+org.scijava:scripting-javascript

Note the usage of the + syntax as needed to append elements to the classpath.

FAQ

• Is it fast? Endpoints are synthesized in a local cache under ~/.jgo. So invoking the same endpoint a second time is really quick.
• What does "no installation" mean? Classpath elements are hard-linked into ~/.jgo from ~/.m2/repository rather than copied, so the ~/.jgo folder has a tiny footprint even if you execute lots of different endpoints.
• What if an endpoint has a new version? Pass the -U flag to jgo to rebuild the endpoint. Note that unlike mvn, though, jgo does not check for updates otherwise.

Configuration

You can configure the behavior of jgo using the $HOME/.jgorc file.

Repositories

You can define additional remote Maven repositories, from which artifacts will be retrieved. E.g.:

[repositories]
imagej.public = https://maven.imagej.net/content/groups/public

If you need more control over where artifacts come from—for example, if you want to use your own remote Maven repository as a mirror of Maven Central—you can do it using Maven's usual ~/.m2/settings.xml; see Using Mirrors for Repositories.

Shortcuts

You can define shortcuts for launching commonly used programs:

[shortcuts]
repl = imagej:org.scijava.script.ScriptREPL
imagej = net.imagej:imagej
fiji = sc.fiji:fiji:LATEST
scifio = io.scif:scifio-cli

Shortcuts are substituted verbatim from the beginning of the endpoint, single-pass in the order they are defined. So e.g. now you can run:

jgo repl

Note that with the repl shortcut above, the main class (org.scijava.script.ScriptREPL) comes from a different artifact than the toplevel artifact (net.imagej:imagej). This is intentional, so that all of ImageJ, including all of the various SciJava scripting-<foo> plugins, is included in the classpath of the REPL.

Settings

There are a few configurable settings:

[settings]
m2Repo = /path/to/.m2Repo (default ~/.m2/repository)
cacheDir = /path/to/.jgo (default ~/.jgo)
links = soft (options: hard, soft, none; default hard)

Note that the jgo cache dir can also be set via the JGO_CACHE_DIR environment variable when using Python jgo. The precedence of reading the cache dir, from highest to lowest:

• JGO_CACHE_DIR environment variable
• cacheDir in settings sections in ~/.jgorc
• default to ~/.jgo

Details

Dependency management

Maven has a feature whereby a project can override the versions of transitive (a.k.a. inherited) dependencies, via a <dependencyManagement> configuration. The problem is: a library may then believe it depends on components at particular versions as defined by its <dependencyManagement>, but downstream projects which depend on that library will resolve to different versions. See this SO thread and this gist for full details.

To work around this issue, you can pass -m to jgo, which causes it to add all endpoints to the synthesized POM's <dependencyManagement> section using import scope. By doing this, the versions of transitive dependencies used in the synthesized project should more precisely match those of each endpoint itself—although in the case of multiple endpoints concatenated via the + operator with conflicting dependency management, the earlier endpoints will win because they will be declared earlier in the POM. See also issue #9 in the jgo issue tracker.

Alternatives

There is JPM4J, but it did not work too well for me:

• It wants to maintain its own local repository of JARs outside of Maven—why? Everyone should use Maven repositories, a thoroughly established standard.

• For each artifact, you have to choose a single main class as its sole command which gets linked into a shell command that runs it.

• It does not seem well-synced with Maven Central, and/or does not seem to deal with dependencies in the expected way; e.g.:

  $ jpm install -l -f -m org.scijava.script.ScriptREPL org.scijava:scijava-common
  Errors
    0. Target specifies Class-Path in JAR but the indicated file .../repo/scijava-expression-parser-3.0.0.jar is not found
    1. Target specifies Class-Path in JAR but the indicated file .../repo/gentyref-1.1.0.jar is not found
    2. Target specifies Class-Path in JAR but the indicated file .../repo/eventbus-1.4.jar is not found
  

• The source is, oddly, part of bnd rather than in the jpm4j organization anywhere, which is not a good sign, modularity-wise.

• Since April 2017, the web site is offline.

There is also Mop, a more general tool with similar features, but it has not been developed since 2010, and its website is also offline.