Hi,

this time I don't have a bug report on Oolong but just simple questions
on two opcodes of the JVM: fcmpg and fcmpl.

In fact, I'm not sure I understand them right.
When I consult the VM Specs (on Sun's site), it says that these two
instructions provide a way to compare two float and that they have an
inverse behaviour with NaN because NaN is unordered.

Until now, everything seems clear. But, I have two questions about that:

1/ Does it mean, that each time I want to translate an expression written
in Java, I have to choose which opcode is appropriate ?

In the VMSpecs, they give this example:
if (d < 100.0) -- use dcmpg
if (d > 100.0) -- use dcmpl

So, if I had used dcmpg in the previous example, the code would be false
when compared to NaN. Right ?

But then, which instruction should I use if I compare two variables ?
(and not a variable with a constant) like this: if (d1 > d2)

That was the easy question ;) Now an even easier question:

2/ What about comparing two NaN numbers ?
Which opcode should I use ?

Comparing two NaN numbers should also always return false, right ?

Thanks in advance for your answers.