a minimized set of JARs for rCDK

[zachcp]

Hi Team-CDK,

I am following up on a set of tweets about the performance of some of the code in rcdk (Tweet here), and I am wondering if you can help us figure out a way to reduce the JAR size we include in rcdk.

Note: I think CDK would be a really good substrate for LOTs of chemical work in R. Java is fast and the $ makes it easy to work. There are currently 2 major dev issues: massive JAR size for CDK-libs and the compiled code step for rCDK.

Although @egonw mentioned that the CDK jars for 2.8 are smaller, I see that the 2.8 bundle is now 45Mb. Maximum package size for CRAN is 5MB. rcdk has been granted a size exception but the latest JAR is already >25Mb. 45 is too big. Ideally, we could find the minimal set of non-bundle JARs that would let us retain backwards compatibility with rcdk ( and all of its dependencies in CRAN/Bioconductor). Do you have any suggestions on the best way to do this?

zach cp

(note: @rajarshi if the above works you might be able to rewrite the JAVA compiled code back to R/rJava for a smoother dev cycle)

[egonw]

rcdk should not use the cdk bundle. Also, in the past the full CDK has been larger than this :)

I guess this issue should go to the rcdk repository, but don't mind discussing this here either.

[zachcp]

Agree this could go in rcdk. The main thing that I thought would be relevant to larger CDK users is how to identify JARs representing some subset of functionality.

[johnmay]

Please see the already open #911. Long story short, don’t include cdk-iordf, then will cut it basically in half. Then next target is jna inchi which targets more platforms than the old JNI libs. You can selectively not support 32-bit Linux/windows for example. You can also shade the jna jar in a similar manner.

[johnmay]

If I have time this year I will try and rewrite the CML processing to be more dependency efficient…

oh the other one to remove is builder3d, the cdk 3D code is not that good anyways :).

As Egon said, don’t use the cdk-bundle JAR you should pull in what you need

[zachcp]

Thank you, @johnmay!

[zachcp]

PR in place for cdk-2.8 jars with a size compatible with previous releases. A bit hacky to generate the individual jars but seems to work for now.