Calling Native C-code from Clojure

[John Doe]

Hello Everyone,

Motivation: I want to invoke C-code functions from Clojure: writing Java functions in Clojure.

Plan of Attack: Rather than go with the JNI approach I want to interface with C-Library via JNR-FFI (https://github.com/jnr/jnr-ffi)

Based on some example of using JNR-FFI

https://github.com/jnr/jnr-ffi-examples/blob/master/getpid/src/main/java/getpid/Getpid.java

https://www.programcreek.com/java-api-examples/?api=jnr.ffi.LibraryLoader

The wrapper in Java might look like

```

package blahblah;

public class BlahNative {

private static int[] intDummy;

private static double[] doubleDummy;

public BlahNative() {

}

public static native void native_a(char var0, char var1, int var2, int var3, int var4, double[] var7, int var8, int var9);

static {

lib = LibraryLoader

            .create(LibSodium.class)

            .search("/usr/local/lib")

            .search("/opt/local/lib")

            .search("/usr/lib")

            .search("/lib")

            .load(LIBRARY_NAME);

        initializeLibrary(lib);

    }

}

```

Question: For my package

```

(defproject mypkg

:dependencies [[org.clojure/clojure "1.10.1"]

   [com.github.jnr/jnr-ffi "2.2.0"]])

```

How do I write the wrapper for the native functions in

```

(ns mypkg.native

(:import [jnr.ffi LibraryLoader])

(:import [jnr.ffi.annotations IgnoreError])

(:import [jnr.ffi.provider FFIProvider]))

```

For the Java interop I am assuming the above namespace should be based on gen-class (http://clojure-doc.org/articles/language/interop.html).

What is not clear to me is:

1. How does one create the `native` method (since most (defn foo) are static methods)?

2. How do I access the native functions after loading the Native Library (LibraryLoader.load(LIBRARY_NAME))?

I am neither a C nor a Java expert so any guidance will be much appreciated.

Thanks,

Ngwua

[Laurens Van Houtven]

Hi,

I did a talk on this and conveniently my library caesium wraps the library you're referencing :)

https://www.youtube.com/watch?v=Lf-M1ZH6KME

https://github.com/lvh/caesium/

I still think that approach is great particularly for libraries like libsodium where some care was taken to make them easy to FFI. That is typically less true for more complex, often C++, libraries. In that case I would consider JavaCPP, where you basically do more up front work to tell it how it works and it generates JNI for you (up front, not dynamically like jnr). Which one I'd recommend depends on what you're binding. Like I said, for libsodium I think jnr is probably easier to use.

lvh

--
You received this message because you are subscribed to the Google
Groups "Clojure" group.
To post to this group, send email to clo...@googlegroups.com
Note that posts from new members are moderated - please be patient with your first post.
To unsubscribe from this group, send email to
clojure+u...@googlegroups.com
For more options, visit this group at
http://groups.google.com/group/clojure?hl=en
---
You received this message because you are subscribed to the Google Groups "Clojure" group.
To unsubscribe from this group and stop receiving emails from it, send an email to clojure+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/clojure/3bd05d7a-297e-49cf-86d4-aec4a88c83cbn%40googlegroups.com.

[Chris Nuernberger]

Ngwua,

Looking through the caesium bindings just mentioned, it appears the author dynamically creates an interface and then you can simply ask the native library to instantiate a concrete implementation of that interface.  So you do not always have to create a class with 'native' methods.

Integrating with C libraries is part of the reason I wrote dtype-next which has the ability to efficiently write/read from either java heap or native heap storage.  dtype-next has no dependency on either JNA or JNR.

Hope this helps,

Chris

--

[John Doe]

Thanks so much lvh.

Your Youtube presentation was one of the main reasons I chose JNI-FFI :-)

I did not consider looking into Caesium, let alone the source code: probably, because my current work does not involve cryptography. Thank you so much for the github link to the repo. Looking at the source code I found it very useful. It gave me all sorts of ideas to move forward :-)

[John Doe]

Thank you Chris. It's very helpful.

The approach you took for dtype-next is very interesting. This library is closer to my work (numerical computation). Also, the efficient use of Java heap is very interesting although I admit I don't know enough about it. This reminds me of Rich Hickey's emphasis of understanding data structures.

[Chris Nuernberger]

You are most welcome :-).  For numerical computation, libjulia-clj may interest you :-).

To view this discussion on the web visit https://groups.google.com/d/msgid/clojure/acd1be6c-5156-4316-98e2-e4df9d55566bn%40googlegroups.com.

[John Doe]

Cool! Thanks again Chris.