That is the teory explanation copy paste from grimware site:
When no device is driving the bus, it's lines are left like disconnected, floating in the air, therefore no real value is fixed on the bus. Usually, no device should try to read the bus in such a state because there's actually nothing to read on it. Nevertheless, if you try do so, for example from BASIC with a PRINT INP(&FFFF), a value is returned. So where this value comes from?
In order to register a 0 on a line, current has to be actively sunk out. If the bus is read while all devices are tri-stated, nothing sinks the current, thus &FF is read. But the more stuff is connected to the bus, the more stray capacitance gets added to the lines, the slower the lines will be able to change their state. For example, if the bus contains &00 just before being left un-driven, it will take some time before the stray capacitance rise to 1 again. If the bus is read in the meantime, the voltage won't cross the TTL high level and a zero will be read.
On the Amstrad, there's only once case where you have to deal with this floating bus value, which is when you want to use the Z80 vectorized interrupts mode (See the Z80 IM2 documentation), and that's all.
The floating value on the databus of a bare CPC is usually &FF, but may also even changes periodically (especially, for some unknown reason at the moment, on CPCs fitted with CRTCtype 1). On a bare Amstrad Plus, this is usually &78 (no idea why it's not &FF, I bet there's some ASIC mystery involved here).
Any program relying on a floating bus value prediction WILL FAIL. Any programmer doing such crap will be flamed all over the intarweb and fucked in mighty hardware scrollers! I warned you!
