Having to work with code bases running on old JVMs is unpleasant at the best of times, but running into old compiler bugs that are not reproducible on current JDKs is just nasty. Case in point is this code which compiles fine under Java 5:
Using the Java 1.4 compiler causes an "incompatible types for ?: neither is a subtype of the other" error, though. To fix it, we have to help out the compiler with a bit of DWIM:
I myI = true ? ((I) new A()) : ((I) new B());
I really have to make a push to get this code running on a newer JVM...
Lester Burnham wrote:Yep, that's the one I found, too (see my first post :-)
BTW, it didn't make a difference if I was a class that both A and B extended - same error.
Please ignore my post - once again I am guilty of not reading right to the bottom of a posting.
I'm surprised it does not work. Looks like you are going to have to use an explicit if-then-else construct. If that fails to compile then maybe a static method to hide the casting. If that fails then ...
The first and only time I encountered this bug I must have read the code about 10.000 times and when I finally found the bug I tested it on 2 different machines just to be sure. Later that month I read an article about it I believe it was an example of the Java Puzzlers book.
Wouter Oet wrote:It's highly discouraged to use the ternary operator in combination with different classes as result. For example this still fails on java version "1.6.0_17" (the latest version on my pc):
I think that has to do with auto-(un)boxing. It seems like, for some reason, the Integer and Double are unboxed again, then boxed to the highest type (Double). If I replace the Integer.valueOf(5) with new BigDecimal("5") then I get 5. That's because BigDecimal cannot be unboxed.