Github user mgaido91 commented on the issue:
https://github.com/apache/spark/pull/19916
I am reporting the result of my tests. I run:
```
Object a = new Integer(1);
long start = System.currentTimeMillis();
for(long i = 0; i < 1000000000; ++ i){
Integer b = ((Integer)a) + 1;
}
long end = System.currentTimeMillis();
System.out.println("Time for casting: " + (end-start));
long start1 = System.currentTimeMillis();
Integer ai = (Integer) a;
for(long i = 0; i < 1000000000; ++ i){
Integer b1 = ai + 1;
}
long end1 = System.currentTimeMillis();
System.out.println("Time without casting: " + (end1-start1));
```
In the compiled code there is an additional operation `checkcast` (both
with janinoc and javac), but the performance result of running this seems
telling that there in not a significant impact:
```
â janino_tests java A
Time for casting: 305
Time without casting: 311
â janino_tests java A
Time for casting: 305
Time without casting: 302
â janino_tests java A
Time for casting: 309
Time without casting: 293
â janino_tests java A
Time for casting: 291
Time without casting: 312
â janino_tests java A
Time for casting: 297
Time without casting: 294
```
What do you think given these results? Should I go on with the PR and
remove also the remaining `addReferenceObj`?
Thanks.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
