Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.
Introduction
In this part of the primer we discuss creating and using custom .jar archives within our R scripts and packages, handling of Java exceptions from R and a quick look at performance comparison between the low and high-level interfaces provided by rJava.
In the first part we talked about using the rJava package to create objects, call methods and work with arrays, we examined the various ways to call Java methods and calling Java code from R directly via execution of shell commands.
Contents
Using rJava with custom built classes
Preparing a .jar archive for use
Getting back to our example with running the main
method of our HelloWorldDummy
class from the first part of this primer, in practice we most likely want to actually create objects and invoke methods for such classes rather than simply call the main method.
For our resources to be available to rJava, we need to create a .jar archive and add it to the class path. An example of the process can be as follows. Compile our code to create the class file, and jar it:
$ javac DummyJavaClassJustForFun/HelloWorldDummy.java $ cd DummyJavaClassJustForFun/ $ jar cvf HelloWorldDummy.jar HelloWorldDummy.class
Adding the .jar file to the class path
Within R, attach rJava, initialize the JVM and investigate our current class path using .jclassPath
:
library(rJava) .jinit() .jclassPath()
Now, we add our newly created .jar to the class path using .jaddClassPath
:
.jaddClassPath(paste0(jardir, "HelloWorldDummy.jar"))
If this worked, we can see the added jar(s) in the class path if we call .jclassPath()
again.
Creating objects, investigating methods and fields
Now that we have our .jar in the class path, we can create a new Java object from our class:
dummyObj <- .jnew("DummyJavaClassJustForFun/HelloWorldDummy") str(dummyObj) ## Formal class 'jobjRef' [package "rJava"] with 2 slots ## ..@ jobj :<externalptr> ## ..@ jclass: chr "DummyJavaClassJustForFun/HelloWorldDummy"
We can also investigate the available constructors, methods and fields for our class (or provide the object as argument, then its class will be queried):
.jconstructors
returns a character vector with all constructors for a given class or object.jmethods
returns a character vector with all methods for a given class or object.jfields
returns a character vector with all fields (aka attributes) for a given class or object.DollarNames
returns all fields and methods associated with the object. Method names are followed by ( or () depending on arity.
# Requesting vectors of methods, constructors and fields by class .jmethods("DummyJavaClassJustForFun/HelloWorldDummy") ## [1] "public java.lang.String DummyJavaClassJustForFun.HelloWorldDummy.SayMyName()" ## [2] "public static void DummyJavaClassJustForFun.HelloWorldDummy.main(java.lang.String[])" ## [3] "public final void java.lang.Object.wait(long,int) throws java.lang.InterruptedException" ## [4] "public final native void java.lang.Object.wait(long) throws java.lang.InterruptedException" ## [5] "public final void java.lang.Object.wait() throws java.lang.InterruptedException" ## [6] "public boolean java.lang.Object.equals(java.lang.Object)" ## [7] "public java.lang.String java.lang.Object.toString()" ## [8] "public native int java.lang.Object.hashCode()" ## [9] "public final native java.lang.Class java.lang.Object.getClass()" ## [10] "public final native void java.lang.Object.notify()" ## [11] "public final native void java.lang.Object.notifyAll()" .jconstructors("DummyJavaClassJustForFun/HelloWorldDummy") ## [1] "public DummyJavaClassJustForFun.HelloWorldDummy()" .jfields("DummyJavaClassJustForFun/HelloWorldDummy") ## NULL # Requesting vectors of methods, constructors and fields by object .jmethods(dummyObj) ## [1] "public java.lang.String DummyJavaClassJustForFun.HelloWorldDummy.SayMyName()" ## [2] "public static void DummyJavaClassJustForFun.HelloWorldDummy.main(java.lang.String[])" ## [3] "public final void java.lang.Object.wait(long,int) throws java.lang.InterruptedException" ## [4] "public final native void java.lang.Object.wait(long) throws java.lang.InterruptedException" ## [5] "public final void java.lang.Object.wait() throws java.lang.InterruptedException" ## [6] "public boolean java.lang.Object.equals(java.lang.Object)" ## [7] "public java.lang.String java.lang.Object.toString()" ## [8] "public native int java.lang.Object.hashCode()" ## [9] "public final native java.lang.Class java.lang.Object.getClass()" ## [10] "public final native void java.lang.Object.notify()" ## [11] "public final native void java.lang.Object.notifyAll()" .jconstructors(dummyObj) ## [1] "public DummyJavaClassJustForFun.HelloWorldDummy()" .jfields(dummyObj) ## NULL
Calling methods 3 different ways
We can now invoke our SayMyName
method on this object in the three ways as discussed is the first part of this primer:
# low level lres <- .jcall(dummyObj, "Ljava/lang/String;", "SayMyName") # high level hres <- J(dummyObj, method = "SayMyName") # convenient $ shorthand dres <- dummyObj$SayMyName() c(lres, hres, dres) ## [1] "My name is DummyJavaClassJustForFun.HelloWorldDummy" ## [2] "My name is DummyJavaClassJustForFun.HelloWorldDummy" ## [3] "My name is DummyJavaClassJustForFun.HelloWorldDummy"
Very quick look at performance
The low-level is much faster, since J
has to use reflection to find the most suitable method. The $
seems to be the slowest, but also very convenient, as it supports code completion:
microbenchmark::microbenchmark(times = 100 , .jcall(dummyObj, "Ljava/lang/String;", "SayMyName") , J(dummyObj, "SayMyName") , dummyObj$SayMyName() ) ## Unit: microseconds ## expr min lq ## .jcall(dummyObj, "Ljava/lang/String;", "SayMyName") 45.503 65.507 ## J(dummyObj, "SayMyName") 870.890 917.514 ## dummyObj$SayMyName() 1148.603 1217.089 ## mean median uq max neval ## 95.20935 77.6195 84.445 1976.195 100 ## 1091.08645 963.7035 1064.606 7603.580 100 ## 1307.03536 1260.5855 1377.438 1731.829 100
Usage of jars in R packages
To use rJava within an R package, Simon Urbanek, the author of rJava even provides a convenience function for this purpose which initializes the JVM and registers Java classes and native code contained in the package with it. A quick step by step guide to use .jars within a package is as follows:
- place our .jars into
inst/java/
- add
Depends: rJava
andSystemRequirements: Java
into ourNAMESPACE
- add a call to
.jpackage(pkgname, lib.loc=libname)
into our.onLoad.R
or.First.lib
for example like so:
.onLoad <- function(libname, pkgname) { .jpackage(pkgname, lib.loc = libname) }
- if possible, add
.java
source files into/java
folder of our package
If you are interested in more detail than provided in this super-quick overview, Tobias Verbeke created a Hello Java World! package with a vignette providing a verbose step-by-step tutorial for interfacing to Java archives inside R packages.
Setting java.parameters
The .jpackage
function calls .jinit
with the default parameters = getOption("java.parameters")
, so if we want to set some of the java parameters, we can do it for example like so:
.onLoad <- function(libname, pkgname) { options(java.parameters = c("-Xmx1000m")) .jpackage(pkgname, lib.loc = libname) }
Note that the
options
call needs to be done before the call to.jpackage
, as Java parameters can only be used during JVM initialization. Consequently, this will only work if other package did not intialize the JVM already.
Handling Java exceptions in R
rJava maps Java exceptions to R conditions relayed by the stop
function, therefore we can use the standard R mechanisms such as tryCatch
to handle the exceptions.
The R condition object, assume we call it e
for this, is actually an S3 object (a list) that contains:
call
– alanguage
object containing the call resulting in the exceptionjobj
– anS4
object containing the actual exception object, so we can for example investigate investigate it’s class:e[["jobj"]]@jclass
tryCatch( iOne <- .jnew(class = "java/lang/Integer", 1), error = function(e) { message("\nLets look at the condition object:") str(e) message("\nClass of the jobj item:") print(e[["jobj"]]@jclass) message("\nClasses of the condition object: ") class(e) } ) ## ## Lets look at the condition object: ## List of 3 ## $ message: chr "java.lang.NoSuchMethodError: <init>" ## $ call : language .jnew(class = "java/lang/Integer", 1) ## $ jobj :Formal class 'jobjRef' [package "rJava"] with 2 slots ## .. ..@ jobj :<externalptr> ## .. ..@ jclass: chr "java/lang/NoSuchMethodError" ## - attr(*, "class")= chr [1:9] "NoSuchMethodError" "IncompatibleClassChangeError" "LinkageError" "Error" ... ## ## Class of the jobj item: ## [1] "java/lang/NoSuchMethodError" ## ## Classes of the condition object: ## [1] "NoSuchMethodError" "IncompatibleClassChangeError" ## [3] "LinkageError" "Error" ## [5] "Throwable" "Object" ## [7] "Exception" "error" ## [9] "condition"
Since class(e)
is a vector of simple java class names which allows the R code to use direct handlers, we can handle different such classes differently:
withCallingHandlers( iOne <- .jnew(class = "java/lang/Integer", 1) , error = function(e) { message("Meh, just a boring error") } , NoSuchMethodError = function(e) { message("We have a NoSuchMethodError") } , IncompatibleClassChangeError = function(e) { message("We also have a IncompatibleClassChangeError - lets recover") recover() # recovering here and looking at # 2: .jnew(class = "java/lang/Integer", 1) # we see that the issue is in # str(list(...)) # List of 1 # $ : num 1 # We actually passed a numeric, not integer # To fix it, just do # .jnew(class = "java/lang/Integer", 1L) } , LinkageError = function(e) { message("Ok, this is getting a bit overwhelming, lets smile and end here :o)") } ) ## Meh, just a boring error ## We have a NoSuchMethodError ## We also have a IncompatibleClassChangeError - lets recover ## recover called non-interactively; frames dumped, use debugger() to view ## Ok, this is getting a bit overwhelming, ## lets smile and end here ## :o) ## Error in .jnew(class = "java/lang/Integer", 1): java.lang.NoSuchMethodError: <init>
References
- Hello Java World! vignette – a tutorial for interfacing to Java archives inside R packages by Tobias Verbeke
- rJava basic crashcourse – at the rJava site on rforge, scroll down to the Documentation section
- The JNI Type Signatures – at Oracle JNI specs
- rJava documentation on CRAN
- Calling Java code from R by prof. Darren Wilkinson
R-bloggers.com offers daily e-mail updates about R news and tutorials about learning R and many other topics. Click here if you're looking to post or find an R/data-science job.
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.