Yeah, but -1 isn't represented in IEEE 754 floating point as all bits one, is it? The ECMAScript standard specifies that the bitwise operators work by first converting their arguments to 32-bit signed integers, but at least how I read it, it doesn't actually say anything about how these 32-bit signed integers are represented (which implies that they continue to be represented as IEEE 754, which isn't how browsers behave, and wouldn't be very useful).
It's not. And you're right, the spec isn't clear. On the other hand, the current behavior of pretending the bits are arranged in 32-bit two's-complement fashion for appropriate integer values when applying bitwise operators is really the only sensible interpretation (especially since a 64-bit float can represent all 32-bit ints).
