BadFloatTest fails with 0.10.1

Godmar Back gback at
Tue Jul 14 16:13:16 PDT 1998


given that 1.0.0b1 was just released, and that support for Sun's
is likely to vanish in the future, you might be wasting your energy on 
fixing these kinds of bugs in a pre 1.0 based system.  Then again, I don't
know your individual needs.

Anyway: I saw that message when porting Kaffe to the ARM, and I think
it indicates problems with the way floating point numbers are represented and 

That is, the problem might lie in LongtoDoubleBits or DoubleToLongBits
in Double.c.  If you enable DEBUG, you can use the test cases 
TestNative.c/ which I wrote for just that purpose.

It is also possible that the system dependent syscall macro in i386/common.h
doesn't work on NT.  You can double check that by adding printf to it
and printing the float and double values that are passed and received.

Hope that helps,


> I'm using the snapshop from the last day or so running on NT4.0. 
> I'm running in interpreter mode, not JIT.
> Most of the regression tests run but a few don't. I'm currently stuck on
> BadFloatTest which produces the following. Any suggestions would be
> appreciated.
> java.lang.IllegalArgumentException: disparate values
> 	at java/lang/Throwable.<init>(line unknown, pc 0x5)
> 	at java/lang/Exception.<init>(line unknown, pc 0x2)
> 	at java/lang/RuntimeException.<init>(line unknown, pc 0x2)
> 	at java/lang/IllegalArgumentException.<init>(line unknown, pc 0x2)
> 	at java/lang/FDBigInt.quoRemIteration(line unknown, pc 0x11)
> 	at java/lang/FloatingDecimal.dtoa(line unknown, pc 0x4d2)
> 	at java/lang/FloatingDecimal.<init>(line unknown, pc 0xb5)
> 	at java/lang/String.valueOf(line unknown, pc 0x5)
> 	at java/io/PrintStream.print(line unknown, pc 0x2)
> 	at java/io/PrintStream.println(line unknown, pc 0x6)
> 	at BadFloatTest.main(12)

More information about the kaffe mailing list