Re: [hatari-devel] Bad Mood freeze

This looks like the problem I have in my current code with the Bad mood keyboard/mouse code you sent me.

Eero, do you have the problem on a real computer ?
I have with my current project.

If bad mood also suffer the same problem, there may be a bug in the keyboard code ; else, it's an hatari problem

It's possible there is a bug in the code - although what occurred in the case we looked at on your machine was the IKBD reporting 'something' which got interpreted as a time/date response packet. We still don't know why that was happening as a result of keypresses.

Even before this, you were getting 'ghost' packet codes with the previous routine, which I think were the same values (?). So whatever is happening there is happening with completely different versions of the handler.

I also saw some pretty strange stuff happening here but not exactly the same problem. I would see 'superimposed' scancodes arriving when too many keys were held at the same time and one of them released. The pattern of keys seemed to play a role. The 'ghost' scancode was generated by a key release but got misinterpreted as a 'wrong' keypress and caused 'stuck' keys. So even with this, something funny is going on inside the Falcon IKBD which is far from clear. It was highly reproducible and I found no cure.

More puzzling is the fact I don't get anything worse than 'stuck keys' on my Falcon (particularly by avoiding certain key combinations). i.e. no jumping to the dead-stop handler as Eero is getting here. That would only occur if the IKBD buffer overflowed e.g. a HOG blit or some other mistake.

However all of these effects may all be related to the same cause, but effects are different if the key pattern is different. i.e. I see 'stuck keys' where someone else with different key mappings sees a dead IKBD handler due to the ghost scancode vectoring to a non-key handler instead of a wrong keypress.

I'll try to reproduce on my Falcon and see if I can break the handler. There have been changes to the project - none I think in the IKBD stuff, but maybe in areas which affect interrupts or blitter use and might be aggrevating things. I'll need to look.