.jinit initializes the Java Virtual Machine (JVM). This
function must be called before any rJava functions can be used.
.jvmState() returns the state of the current JVM.
.jinit(classpath = NULL, parameters = getOption("java.parameters"), ...,
silent = FALSE, force.init = FALSE)
.jvmState()The return value is an integer specifying whether and how the VM was initialized. Negative values indicate failure, zero denotes successful initialization and positive values signify partially successful initilization (i.e. the VM is up, but parameters or class path could not be set due to an existing or incompatible VM).
.jvmState returns a named list with at least the following
elements:
TRUE if rJava is initialized and has a
runing JVM, FALSE otherwise.
string representing the current state of the JVM. One of
the following values:
"none" if there is no JVM, "created" if the current
JVM has been created by rJava, "attached" if rJava attached
into an existing JVM (typically when R is embedded into a running
JVM via JRI), "detached" if there is a JVM (such as embedded
R), but rJava has not been initialized to use it, "dead" if
the process is about to die due to the JVM forcing en exit or
"destroyed" if a JVM existed before, but was destroyed.
Any additional classes to include in the Java class
paths (i.e. locations of Java classes to use). This path will be
prepended to paths specified in the CLASSPATH environment
variable. Do NOT set this system class path initializing a package,
use .jpackage instead, see details.
character vector of parameters to be passed to the virtual machine. They are implementation dependent and apply to JDK version 1.2 or higher only. Please note that each parameter must be in a separate element of the array, you cannot use a space-separated string with multiple parameters.
Other optional Java initialization parameters (implementation-dependent).
If set to TRUE no warnings are issued.
If set to TRUE JVM is re-initialized even if
it is already running.
Starting with version 0.5 rJava provides a custom class loader that can
automatically track classes and native libraries that are provided in
R packages. Therefore R packages should NOT use .jinit, but
call .jpackage instead. In addition this allows the use
of class path modifying function .jaddClassPath.
Important note: if a class is found on the system class path (i.e. on
the classpath specified to .jinit) then the system class
loader is used instead of the rJava loader, which can lead to problems
with reflection and native library support is not enabled. Therefore
it is highly recommended to use .jpackage or
.jaddClassPath instead of classpath (save for system
classes).
Stating with version 0.3-8 rJava is now capable of modifying the class path on the fly for certain Sun-based Java virtual machines, even when attaching to an existing VM. However, this is done by exploiting the way ClassLoader is implemented and may fail in the future. In general it is officially not possible to change the class path of a running VM.
At any rate, it is impossible to change any other VM parameters of a
running VM, so when using .jinit in a package, be generous with
limits and don't use VM parameters to unnecessarily restrict
resources (or preferably use .jpackage instead). JVM
parameters can only be set if the initial state of the JVM is
"none".
There is a subtle difference between "initialized" and the JVM
state. It is in theory possible for "initialized" to be
FALSE and still "state" to be "created" or
"attached" in case where JVM was created but rJava has not been
able to initialize for other reasons, although such state should be
rare and problematic in either case. Behavior of rJava functions other
than .jinit and .jvmState is undefined unless
.jvmState()$initialized is TRUE.
.jpackage
if (FALSE) {
## set heap size limit to 512MB (see java -X) and
## use "myClasses.jar" as the class path
.jinit(classpath="myClasses.jar", parameters="-Xmx512m")
.jvmState()
}
Run the code above in your browser using DataLab