So I built tensorflow JNI and now have a directory with the tensorflow JAR file and dylib files. I can't for the life of me figure out how to add this directory as part of the classpath with boot. I can't build a POM (long story, but pretty much impossible). Anyone know how to add a one-off jar to the -cp with boot?

java -cp libtensorflow.jar:clojure-1.8.0.jar clojure.main

kanwei boot.pod/add-classpath should do it (https://github.com/boot-clj/boot/blob/master/doc/boot.pod.md#add-classpath), you can do that in your build.boot

note that if you're producing a jar or something from your build, the jar won't be in there unless you mvn install it and then make it a proper dependency

@alandipert

(set-env! :resource-paths #{"/Users/kanwei/Src/tensorflow-java/libtensorflow.jar"})
(add-classpath "/Users/kanwei/src/tensorflow-java/libtensorflow.jar")

java.net.MalformedURLException: no protocol: /Users/kanwei/src/tensorflow-java/libtensorflow.jar
    clojure.lang.ExceptionInfo: no protocol: /Users/kanwei/src/tensorflow-java/libtensorflow.jar
    line: 6

changing it to file:///Users gives

java.lang.IllegalAccessError: Context classloader is not a DynamicClassLoader
  clojure.lang.ExceptionInfo: Context classloader is not a DynamicClassLoader
    line: 5

kanwei no need for the set-env! piece, just: (require '[boot.pod :as pod]) (pod/add-classpath "/Users/kanwei/Src/tensorflow-java/libtensorflow.jar")

nice, that worked

any reason why there's both add-classpath and pod/add-classpath?

they seem to have totally different behavior despite the same name

one is in clojure.core, one is in boot.pod

the boot one isn't deprecated and does boot stuff

ah, had no idea clojure had that in core. Thanks for the help!

@alandipert another issue: I need to add

(System/setProperty "java.library.path" "/Users/kanwei/src/tensorflow-java")

-Djava.library.path

how do I setProperty on the new pod?

Hi, is there a way to spawn a repl in cider for a single file boot script to execute it without a context of a project?

Can't find the info in the wiki

Ah, ok, I can just do boot repl and jack-in

kanwei you can set BOOT_JVM_OPTIONS, eg BOOT_JVM_OPTIONS="-Djava.library.path=/Users/kanwei/src/tensorflow-java boot ...

setting it from clj is trickier, and i'm not sure works in java 9. but might still: https://github.com/boot-clj/boot/issues/185

@alandipert moving the dylib files into the root of the project folder is working

alright time to do some actual machine learning 🙂

yup, it's gross

Anyone here using tools.deps yet? Anyone looking at Boot tasks to work with that library to support external deps.edn files?

(I'm asking to see whether there's someone to work with on this or whether I'm going to be a trail blazer 🙂 )

Hi all - I'm trying to deploy my boot project to run headless but would like to retain the opportunity to connect via remote repl to adjust config while it's running. I tried playing with the repl task :server true option - but it doesn't seem to be properly fitting the bill. Any guidance out there on this type of deployment? Thanks!

mattyulrich how do you deploy, as standalone uberjar? as .war?

... or is boot the entrypoint?