It works for me in 3.0.19 (Ubuntu) and 3.5.10 (Mac). Can you upgrade to 3.5.10 and see if that helps?
The two pairs of tests are not identical; one uses localvar arguments (@fmod var1 var2), the other uses constant arguments (@fmod M_N0 M_1). I do that a lot, since Quixe sometimes optimizes constant arguments and I want to test both code paths.
That was at uni, so I can’t upgrade. I can however try to confirm it again on some other computers there.
It works fine for me at home on FF 3.6.
I believe this is Firefox bug bugzilla.mozilla.org/show_bug.cgi?id=503595
As it’s an obscure corner case, and fixed in the current Firefox 3.5 and 3.6 streams, I don’t think it’s necessary to patch Quixe to work around it.
Yeah I don’t think so either. Cause well… -0 == 0 in reality 
Hi,
ZMPP now passes all float tests as well. Somehow the glulxercise version in the glk-dev repository (100811) shows 8 errors (which are sign-related and glulxe reports them as well), but the one in quixe (100816) shows zero for glulxe and ZMPP (I might as well have made a mistake in pulling, I took it from the float branch).
Best,
Wei-ju
It occurs to me that it’d be nice if the opcode numbers were listed with the descriptions of the opcodes, instead of (or in addition to) in a separate table.
I agree, maybe somewhat like in the Z-Machine specification.
The float branches are dead now. I fixed some bugs (in glk-dev, quixe, and glulxe) after I merged to master.
(I haven’t merged in the inform6 project yet, because that tracks the official release.)
As for the spec formatting, maybe. 
I’m more worried about getting the numbers wrong myself.