[core] Default switch case lost in endless loop

[eybisi]

Issue details

if switch is inside endless while loop, (while(true)) default switch branch is always discarded and lost. Following java code:

    public int jadxEndlessSwitch(int r){
        int r0;
        while(true){
            switch(r){
                case 42:
                    r0 = 32;
                    break;
                case 52:
                    r0 = 42;
                    break;
                default:
                    System.out.println("Default switch case");
                    return 1;
            }
            r = r0;
        }
    }

results in:

    /* JADX WARN: Code restructure failed: missing block: B:8:0x0004, code lost:
   
       java.lang.System.out.println("Default switch case");
    */
   /* JADX WARN: Code restructure failed: missing block: B:9:0x000c, code lost:
   
       return 1;
    */
   /*
       Code decompiled incorrectly, please refer to instructions dump.
   */
   public int jadxEndlessSwitch(int r) {
       int r0;
       while (true) {
           switch (r) {
               case 42:
                   r0 = 32;
                   break;
               case 52:
                   r0 = 42;
                   break;
           }
           r = r0;
       }
   }

I've attached sample app.

Relevant log output or stacktrace

WARN : Code restructure failed: missing block: B:8:0x0004, code lost:

    java.lang.System.out.println("Default switch case");
 in method: org.abc.obfusactiontests.MainActivity.jadxEndlessSwitch(int):int, file: classes3.dex
WARN : Code restructure failed: missing block: B:9:0x000c, code lost:

    return 1;
 in method: org.abc.obfusactiontests.MainActivity.jadxEndlessSwitch(int):int, file: classes3.dex

Provide sample and class/method full name

app-debug.zip

class : org.abc.obfusactiontests.MainActivity
method : jadxEndlessSwitch

Jadx version

dev

[skylot]

@eybisi thanks for test case!
Actually, right now I am working on a new approach for break/continue handling (related to #2339), so it affects loops and switches. And I will add your test case into that change, it may take some time though 🙂

[MrIkso]

vineflower or fernflower handles this case, maybe it's worth looking at how they perform code generation for such cases...

[skylot]

Fixed.
Cause of the issue was incorrect checks for empty default case, hope fix will cover all cases.
Also, rework mentioned above, not going well and will be postponed, so I made a fix just for this case.