signed ifields

[Ben Elliston]

I'm struggling to understand some aspects of CGEN's design with
respect to {signed,unsigned} quantities for instruction fields.  I'm
unsure about why it makes sense to attribute "signedness" to what is
effectively a sequence of bits.  Shouldn't the signedness be
attributed to operands (which it already can be)?  It seems redundant
that both objects can be attributed as such.

I bring this up because some internal users of CGEN have been caught
by unintuitive behavior when using (dnf ..) to define their ifields
and then flagging their operands as signed, only to find that the
extraction code treats the underlying ifield as unsigned.

Comments?

Ben