libtnoodle

An experimental C/C++ replacement for the scrambling guts of cubing/tnoodle.

NOTHING IN HERE IS SANCTIONED BY THE WCA. THIS MUST NOT BE USED FOR OFFICIAL WCA COMPETITIONS.

How to use

See jtnoodle-demo for an example project that uses jtnoodle.

How to develop

Requirements

• Swig is used to produce the Java (JNI) and Python wrappers for our C++ code.
• Clang and the JDK are required to compile for Linux.
  • Make sure you install the python libclang bindings. On Fedora 18 with clang 3.3, I had to run sudo pip install clang==3.3 (Read this if you think you installed the wrong version.)

A properly configured Linux machine can cross-compile binaries for Windows and OS X as well.

• MinGW-w64 is required to compile to compile for Windows.
• https://github.com/tpoechtrager/osxcross needs to be set up to compile for OS X. You must set the MACOSX_DEPLOYMENT_TARGET environment variable to 10.9 (MACOSX_DEPLOYMENT_TARGET=10.9)!

If you have trouble setting up a development environment, it might be worth checking out .travis.yml, which has all the commands to set things up for Travis CI's 64 bit Ubuntu 12.04 environment.

Building

• ./gradlew jar produces build/libs/jtnoodle-VERSION.jar.
• ./gradlew run runs main.cpp (this may be removed in the future).
• ./gradlew js produces build/tnoodle-VERSION.js.

Testing

• ./gradlew test to run python tests.
• If you want to run tests in src/test/ directly, you need to run ./gradlew pytnoodle first to produce the Swig wrapper code in build/pytnoodle.

Releasing

Travis CI is configured to publish to http://oss.jfrog.org/oss-snapshot-local/tk/tnoodle/ on each build.

We could investigate manual deployment to oss-release-local and promoting releases to jcenter.

Components (under src/)

• cnoodle - A pure C/C++ library for generating scrambles. Produces libcnoodle.{so,dll,dylib}
• jtnoodle - A Java Native Interface (JNI) wrapper for cnoodle. It also defines a very basic main() to generate scrambles via the cli. Produces jtnoodle-VERSION.jar.
• pytnoodle - A Python wrapper for cnoodle. Produces pytnoodle.py and _pytnoodle.so in build/pytnoodle. We could investigate publishing this to PyPI.
• test - Python tests for cnoodle. Depends on pytnoodle.

Random musings...

This is an experiment. It remains to be seen how we're going to include a version of libtnoodle for every major operating system + architecture pair in jtnoodle.jar so that it just works (like the current pure Java tnoodle).

If we succeed at producing a Java desktop app that can work everywhere, then we can declare this experiment a success, and start porting the Java scrambling code in tnoodle to libtnoodle. As this port is occurring, we can change tnoodle to depend on jtnoodle. This way we can reuse the existing webserver/pdf/zip/json generation infrastructure that we've already written in Java in tnoodle.

Why?

TNoodle evolved strangly. Generating random turn scrambles for an event like 7x7x7 takes seconds, when it should be instantaneous. This is due to a lot of infrastructure that was added to aid in scramble filtering. Porting the scrambling code to C++ gives us a needed excuse to revisit and rewrite.

In addition to needing a rewrite, C++ code gives us the option to run libtnoodle in as many places as possible.

• Javascript (via Emscripten) - Emscripten does a much better job of producing efficient Javascript than GWT (the Java to Javascript compiler currently used by tnoodle).
• Android (via the Android NDK) - This is harder with C++ code than with Java, but I experimented with the NDK in jfly/fskube, and determined that it does work well.
• iOS (natively via Objective-C++) - Running Java in iOS is impossible. It should be easy to compile and run libtnoodle because it is written entirely in C++.