Experimental feature in GraalVM

Interoperability

The Polyglot API #

Since GraalVM supports several other programming languages including JavaScript, R, Ruby, and those that compile to LLVM bitcode, it also provides a Python API to interact with them. In fact, GraalVM uses this API internally to execute Python C extensions using the GraalVM LLVM runtime.

You can import the polyglot module to interact with other languages:

import polyglot

You can import a global value from the entire polyglot scope:

imported_polyglot_global = polyglot.import_value("global_name")

This global value should then work as expected:

  • Accessing attributes assumes it reads from the members namespace.
  • Accessing items is supported both with strings and numbers.
  • Calling methods on the result tries to do a straight invoke and falls back to reading the member and trying to execute it.

You can evaluate some inlined code from another language:

polyglot.eval(string="1 + 1", language="ruby")

You can evaluate some code from a file, by passing the path to it:

polyglot.eval(path="./my_ruby_file.rb", language="ruby")

If you pass a file, you can also rely on the file-based language detection:

polyglot.eval(path="./my_ruby_file.rb")

You can export some oblect from Python to other supported languages so they can import it:

foo = object()
polyglot.export_value(foo, name="python_foo")

The export function can be used as a decorator. In this case the function name is used as the globally exported name:

@polyglot.export_value
def python_method():
    return "Hello from Python!"

Here is an example of how to use the JavaScript regular expression engine to match Python strings. Save this code to the polyglot_example.py file:

import polyglot

re = polyglot.eval(string="RegExp()", language="js")

pattern = re.compile(".*(?:we have (?:a )?matching strings?(?:[!\\?] )?)(.*)")

if pattern.exec("This string does not match"):
    raise SystemError("that shouldn't happen")

md = pattern.exec("Look, we have matching strings! This string was matched by Graal.js")
if not md:
    raise SystemError("this should have matched")

print("Here is what we found: '%s'" % md[1])

To run it, pass the --jvm --polyglot option to the graalpython launcher:

graalpython --jvm --polyglot polyglot_example.py

This program matches Python strings using the JavaScript regular expression object. Python reads the captured group from the JavaScript result and prints: Here is what we found: ‘This string was matched by Graal.js’.

As a more complex example, see how you can read a file using R, process the data in Python, and use R again to display the resulting data image, using both the R and Python libraries in conjunction. To run this example, first install the required R library:

R -e 'install.packages("https://www.rforge.net/src/contrib/jpeg_0.1-8.tar.gz", repos=NULL)'

This example also uses image_magix.py and works on a JPEG image input (you can try with this image). These files have to be in the same folder that the script below is located in and executed from.

import polyglot
import sys
import time
sys.path.insert(0, ".")
from image_magix import Image

load_jpeg = polyglot.eval(string="""function(file.name) {
    library(jpeg)
    jimg <- readJPEG(file.name)
    jimg <- jimg*255
    jimg
}""", language="R")

raw_data = load_jpeg("python_demo_picture.jpg")

# the dimensions are R attributes; define function to access them
getDim = polyglot.eval(string="function(v, pos) dim(v)[[pos]]", language="R")

# Create object of Python class 'Image' with loaded JPEG data
image = Image(getDim(raw_data, 2), getDim(raw_data, 1), raw_data)

# Run Sobel filter
result = image.sobel()

draw = polyglot.eval(string="""function(processedImgObj) {
    require(grDevices)
    require(grid)
    mx <- matrix(processedImgObj$`@data`/255, nrow=processedImgObj$`@height`, ncol=processedImgObj$`@width`)
    grDevices:::awt()
    grid.raster(mx, height=unit(nrow(mx),"points"))
}""", language="R")

draw(result)
time.sleep(10)

The Java Host Interop API #

Finally, to interoperate with Java (only when running on the JVM), you can use the java module:

import java
BigInteger = java.type("java.math.BigInteger")
myBigInt = BigInteger(42)
myBigInt.shiftLeft(128)
# public Java methods can just be called
myBigInt["not"]()
# Java method names that are keywords in Python can be accessed using "[]"
byteArray = myBigInt.toByteArray()
# Java arrays can act like Python lists
print(list(byteArray))

For packages under the java package, you can also use the normal Python import syntax:

import java.util.ArrayList
from java.util import ArrayList

java.util.ArrayList == ArrayList

al = ArrayList()
al.add(1)
al.add(12)
print(al)
# prints [1, 12]

In addition to the type builtin method, the java module exposes the following methods as well:

Builtin Specification
instanceof(obj, class) returns True if obj is an instance of class (class must be a foreign object class)
is_function(obj) returns True if obj is a Java host language function wrapped using Truffle interop
is_object(obj) returns True if obj if the argument is Java host language object wrapped using Truffle interop
is_symbol(obj) returns True if obj if the argument is a Java host symbol, representing the constructor and static members of a Java class, as obtained by java.type
import java
ArrayList = java.type('java.util.ArrayList')
my_list = ArrayList()
print(java.is_symbol(ArrayList))
# prints True
print(java.is_symbol(my_list))
# prints False, my_list is not a Java host symbol
print(java.is_object(ArrayList))
# prints True, symbols are also host objects
print(java.is_function(my_list.add))
# prints True, the add method of ArrayList
print(java.instanceof(my_list, ArrayList))
# prints True

See Polyglot Programming and Embed Languages for more information about interoperability with other programming languages.

The Behaviour of Types #

The interop protocol defines different “types” which can overlap in all kinds of ways and have restrictions on how they can interact with Python.

Interop Types to Python #

Most importantly and upfront - all foreign objects passing into Python have the Python type foreign. There is no emulation of i.e., objects that are interop booleans to have the Python type bool. This is because interop types can overlap in ways that the Python builtin types cannot, and it would not be clear what should take precendence. Instead, the foreign type defines all of the Python special methods for type conversion that are used throughout the interpreter (methods like __add__, __int__, __str__, __getitem__ etc) and these try to do the right thing based on the interop type (or raise an exception.)

Types not listed in the below table have no special interpretation in Python right now.

Interop type Python interpretation
Null It is like None. Important to know: interop null values are equal, but not identical! This was done because JavaScript defines two “null-like” values; undefined and null, which are not identical
Boolean Behaves like Python booleans, including the fact that in Python, all booleans are also integers (1 and 0 for true and false, respectively)
Number Behaves like Python numbers. Python only has one integral and one floating point type, but it cares about the ranges in some places such as typed arrays.
String Behaves like Python strings.
Buffer Buffers are also a concept in Python’s native API (albeit a bit different). Interop buffers are treated like Python buffers in some places (like memoryview) to avoid copies of data.
Array Arrays can be used with subscript access like Python lists, with integers and slices as indices.
Hash Hashes can be used with subscript access like Python dicts, with any hashable kind of object as key. “Hashable” follows Python semantics, generally all interop types with identity are deemed “hashable”. Note that if an interop object is both Array and Hash, the behavior of the subscript access is undefined.
Members Members can be read using normal Python ~.~ notation or the getattr etc functions.
Iterable Iterables are treated like Python objects with an __iter__ method, that is, they can be used in loops and other places that accept Python iterables.
Iterator Iterators are treated like Python objects with a __next__ method.
Exception Interop exceptions can be caught in generic except clauses.
MetaObject Interop meta objects can be used in subtype and isinstance checks
Executable Executable objects can be executed as functions, but never with keyword arguments.
Instantiable Instantiable objects behave like executable objects (similar to how Python treats this)

Python to Interop Types #

Interop type Python interpretation
Null Only None.
Boolean Only subtypes of Python bool. Note that in contrast to Python semantics, Python bool is never also an interop number.
Number Only subtypes of int and float.
String Only subtypes of str.
Array Any object with a __getitem__ and a __len__, but not if it also has keys, values, and items (like dict does.)
Hash Only subtypes of dict.
Members Any Python object. Note that the rules for readable/writable are a bit ad-hoc, since checking that is not part of the Python MOP.
Iterable Anything that has an __iter__ method or a __getitem__ method.
Iterator Anything with a __next__ method.
Exception Any Python BaseException subtype.
MetaObject Any Python type.
Executable Anything with a __call__ method.
Instantiable Any Python type.