can we pass bwa index pointer directly?

[akiezun]

I see that the index is passed to the JNI laver via the BWAIndex object. Really though I think only the pointer is actually used. Can it be passed directly rather then passing a Java object that needs to be unwrapped in the C layer? I mean changing this API

    private native AlnRgn[] align(BwaIndex bwaIndex,byte bases[])  throws IOException;
    private native String[] align2(BwaIndex bwaIndex,final ShortRead ks1[],final ShortRead ks2[])  throws IOException;

to

    private native AlnRgn[] align(long bwaIndexPtr,byte bases[])  throws IOException;
    private native String[] align2(long bwaIndexPtr,final ShortRead ks1[],final ShortRead ks2[])  throws IOException;

[akiezun]

@SHuang-Broad wdyt?

[SHuang-Broad]

If you look at this core-file, it seems that though in Java it is a long, when it is actually used in the "core" function align2:

/*
 * Class:     com_github_lindenb_jbwa_jni_BwaMem
 * Method:    align2
 * Signature: ([Lcom/github/lindenb/jbwa/jni/ShortRead;[Lcom/github/lindenb/jbwa/jni/ShortRead;)I
 */
JNIEXPORT jobjectArray JNICALL QUALIFIEDMETHOD(BwaMem_align2)(
    JNIEnv *env,
     jobject self,
     jobject bwaIndex,
     jobjectArray ks1,
     jobjectArray ks2
     )
    {
    bseq1_t *seqs=NULL;
    int i;
    int numPairs=0;
    jclass shortReadClass;
    jmethodID shortReadGetName;
    jmethodID shortReadBases;
    jmethodID shortReadQualities;  
    jobjectArray samRetArray=NULL;
    mem_opt_t* opt= _getBwaMem(env,self);
    bwaidx_t* idx= _getBwaIndex(env,bwaIndex);  

It actually calls the C-style template for not just getting reference (see lines 255 and 285), the scheme may not be that simple?

static retType _get##method(JNIEnv * env, jobject self)\
    {\
    jlong ptr;jfieldID field;\
    jclass c= (*env)->GetObjectClass(env,(self));\
    VERIFY_NOT_NULL(c);\
    field = (*env)->GetFieldID(env,c, ref, "J");\
    VERIFY_NOT_NULL(field);\
    ptr= (*env)->GetLongField(env,self,field);\
    return (retType)ptr;\
    }\

There could exist a better way of not using this C-style template (I would guess that because mem_opt_t is simple but contains loads of parameters, doing this template-style makes sense), and not exposing an BwaIndex object, but rather handles index loading inside the binding would be better?

Idea:
In align2, simply check if a "global" bwaidx_t * is not 0, instead of calling _getBwaIndex() (which is the call to the template of converting the long representation to the true C ptr representation).

[lindenb]

@akiezun i you could pass the pointer/long directly, but IMHO you would lose the semantics of the method (my IDE can see that it needs BwaIndex and not any kind of long value).

The C template are a generic (and paranoid) way to safely set/get the internal C pointer stored in the java objects as a long.

@SHuang-Broad @akiezun as I said before in another thread, jbwa is a playground/proof-of-concept for calling bwa from java. I havent' used it since I've published on my blog. I you want to make some major changes feel free to submit a PR or I can even add you as collaborators for this project to speed-up things.

[akiezun]

@SHuang-Broad can you try both and report performance? That's all I care about here.

[SHuang-Broad]

@akiezun Will do.
@lindenb I'll let you know as well, and may submit PR's along the way.