我在使用jni从java类中读取arraylist时遇到了麻烦。在java类中,arraylist的定义如下
public static List XTN_STC_search_result = new ArrayList();
然后,arraylist将填充一个字符串,一个得分值为double和一个整数数组。
所以结构看起来像这样:
[
[label str 1, score value 1, [int 1, int 2, int n]],
[label str 2, score value 2, [int 1, int 2],
...
]
在JNI C方面,我执行以下操作来阅读该arraylist。
在读取静态字段并获取对象后,我进入循环迭代数组以读取所有值。一切顺利,我可以读取每个arraylist条目,标签字符串,double值,并可以使用嵌套循环访问整数数组。在嵌套循环中,如果使用getObjectArrayElement获取整数数组条目,它似乎获得该值。之后我在NULL上测试它并且它不是NULL。为了迂腐,我仔细检查该数组条目(整数数组)是否是带有isInstanceOf的Integer实例。那个叫做SIGSEGV和JVM的程序崩溃了。
下面你会看到问题区域的代码片段......
int
XTN_ce_java_get_cluster(XTN_VM_IDX vm_idx) {
...
/* java result object , the arraylist to read */
jobjectArray result_obj;
/* result object array (element of result_obj) */
jobjectArray jv_result_object_array;
jint jv_result_object_array_size;
/* element to hold phrase, score and nested integer array */
jobject jv_object_array_cluster_elem;
jint jv_object_array_cluster_elem_size;
...
/* get arraylist */
jv_result_object_array = (*(Record.env))->GetObjectArrayElement(
Record.env, result_obj, 1 );
jv_result_object_array_size = (*(Record.env))->GetArrayLength(Record.env,
jv_result_object_array );
...
/* loop over arraylist entries */
for (i = 0; i < jv_result_object_array_size; i++) {
/* read label string */
if ( (*(Record.env))->IsInstanceOf(Record.env,
(jobject)jv_object_array_cluster_elem, classString)) {
...
} else if ( (*(Record.env))->IsInstanceOf(Record.env, jv_object_array_cluster_elem, classObject)) {
/* we are about to read the integer array from the arraylist ... */
/* get size of integer array */
jv_object_array_cluster_elem_size = (*(Record.env))->GetArrayLength(
Record.env, jv_object_array_cluster_elem);
/* this case should not happen, the integer array has a minimum entry of 1 element */
if (jv_object_array_cluster_elem_size <= 0)
continue;
for (j = 0; j < jv_object_array_cluster_elem_size; j++) {
/* get element from integer array */
jv_cluster_data_array_elem = (*(Record.env))->GetObjectArrayElement(
Record.env, jv_object_array_cluster_elem, j);
/* check if null */
if (jv_cluster_data_array_elem == NULL) {
break;
/* now check type */
/* THIS CALL CRASHES JVM */
if ( (*(Record.env))->IsInstanceOf(Record.env, jv_cluster_data_array_elem,
classInteger)) {
fprintf(stdout, "got array entry of type integer */
}
...
} /* inner loop integer array */
} /* outer loop object array */
return 1;
}
我不明白它为什么会在那里崩溃,即使我似乎无法访问该字段进行读取(例如使用GetIntField())。如果我省略了isInstanceOf调用(只是循环遍历整数数组)一切顺利,它会循环遍历完整的arraylist,读取我的字符串和双精度数,循环遍历嵌套的整数数组并成功完成。
。老实说是否有人暗示我要去哪里或要改善什么?我希望提供的arraylist reader函数的代码片段足以理解我的问题。也许有人可以指出我在这个问题上的正确方向,希望: - ))
我用 JRE版本:6.0_22-b04 Java VM:Java HotSpot(TM)64位服务器VM(17.1-b03混合模式linux-amd64) 在Ubuntu上
非常感谢您提前帮助我解决这个问题。
Update 2010/12/06: ================== As Peter suggested, I completely reworked the object reader function using the Java methods of the object through the native interface. Additionally it turned out, that I even do not need that very arraylist I constructed in the Java class, I rather used and read the internal class Results Object List directly and its object methods to read the results I need. So in short, now it is more efficient and I was able to skip that Arraylist construction completey and spare one processing step. First I thought, using the Java methods of that very class through JNI would give me a much higher response time, using Java methods through JNI is surely not very fast (compared to using the Java methods in Java directly). Using JNI as a bridge from C to Java costs performance and computing time. But anyway, I figured out, I am about 200ms faster (using the same data set) than my old reader method with the arraylist, so I even gained some performance. Ok, to put it in short, using JNI and the Object methods of the Java class spared me a lot of headache and the code now is much more readable. Conclusion: I would like to encourage everybody that is facing a similar problem using the Object methods of the class you work with through JNI.
答案 0 :(得分:1)
首先要注意的是ArrayList!=一个数组。用于访问arraylist的方法仅适用于数组。
我建议您通过调用List的方法来访问ArrayList,就像在Java中一样。