tableswitch and Method size

[mayerrobert]

Hi,

it seems like JarScan miscalulates method sizes. Apparently it considers the first occurrence of the "tableswitch" bytecode as the end of the method and only counts bytecodes up to but not including the first "tableswitch". Here's what I tried to prove my assumption:

• compile a simple class with a method that contains "switch"
• decompile using javap. The bytecode size of "f" is 66, there are 9 bytes up to "tableswitch"
• JarScan will report "f" was 9 bytes which is incorrect IMO

C:\robert\jarscanbug>cat Tableswitch.java
public class Tableswitch {
void f(int n) {
System.out.println("Hi!");
switch (n) {
case 0: System.out.println("zero"); break;
case 1: System.out.println("one"); break;
case 2: System.out.println("two"); break;
}
}
}
C:\robert\jarscanbug>javac Tableswitch.java

C:\robert\jarscanbug>javap -c Tableswitch.class
Compiled from "Tableswitch.java"
public class Tableswitch {
public Tableswitch();
Code:
0: aload_0
1: invokespecial #1 // Method java/lang/Object."":()V
4: return

void f(int);
Code:
0: getstatic #2 // Field java/lang/System.out:Ljava/io/PrintStream;
3: ldc #3 // String Hi!
5: invokevirtual #4 // Method java/io/PrintStream.println:(Ljava/lang/String;)V
8: iload_1
9: tableswitch { // 0 to 2
0: 36
1: 47
2: 58
default: 66
}
36: getstatic #2 // Field java/lang/System.out:Ljava/io/PrintStream;
39: ldc #5 // String zero
41: invokevirtual #4 // Method java/io/PrintStream.println:(Ljava/lang/String;)V
44: goto 66
47: getstatic #2 // Field java/lang/System.out:Ljava/io/PrintStream;
50: ldc #6 // String one
52: invokevirtual #4 // Method java/io/PrintStream.println:(Ljava/lang/String;)V
55: goto 66
58: getstatic #2 // Field java/lang/System.out:Ljava/io/PrintStream;
61: ldc #7 // String two
63: invokevirtual #4 // Method java/io/PrintStream.println:(Ljava/lang/String;)V
66: return
}

C:\robert\jarscanbug>java -cp ..\scripts\jitwatch-ui-1.4.7-shaded-win.jar org.adoptopenjdk.jitwatch.jarscan.JarScan --mode=methodLength --length=9 .
"","Tableswitch","f","int",9

[chriswhocodes]

Thanks for reporting! I had no idea anyone was still using Jarscan. I will look into this for you.

[chriswhocodes]

Hi @mayerrobert can you check you are using the latest version of JITWatch please? I believe I fixed this in 8aa1097 and I get the correct result from your test class on the latest code:

java -cp ui/target/jitwatch-ui-shaded.jar org.adoptopenjdk.jitwatch.jarscan.JarScan --mode=methodLength --length=67 . | grep Tableswitch

"","Tableswitch","f","int",67

Kind regards,

Chris

[mayerrobert]

Great, it works now!

I have since written a simple testprogram that displays method sizes using javaassist (but without convenient features of jarscan such as commandline arguments).

Both my simple program as well as the updated jarscan now show the same results.

As far as I'm concerned you could close the issue now.

Cheers, Robert.

PS: if you're courious - I'm currently refactoring a rather old and big project that has methods up to 11k bytecode size, and jarscan is handy to find these offenders which I then split up or otherwise refactor. Number of lines of Java source code is one thing to consider, but jarscan provides useful information, too.

Edit: took me a while to figure out that just updating to HEAD got me your fix that you already did some time ago. Should have thought of that myself, sorry :-D I'll keep that in mind for the future.

[chriswhocodes]

Glad it's working for you and that you're finding it useful 😀