Interface for regression testing & input event capture/replay?

Hi;

I'm looking for some guidance from Linux/input people...

I've looked through the archives, and around the net a bit, and I
haven't been able to find anything to capture events from an arbitrary
set of inputs, and then replay them back to software attached to those
inputs.

This would be useful for:

o Regression testing user space algorithms
o Training software for gesture recognition
o Tuning drivers for new hardware to correct self-censor undesirable events
o Probably lots of other things I haven't thought of

The multiple event sources requirement comes from using things like
keyboard modifier keys to change mouse button behaviours.  The most
obvious example is probably option-click on Apple hardware with a one
button mouse.

The basic problems with just opening up /dev/eventX, /dev/eventY, ...
and listening/injecting are:

o If a piece of software grabs the input, I don't get events
o If I grab the input, the software I'm testing doesn't get events
o If neither of us grabs the input, filters apply
o If the input is grabbed by someone, you can't inject events
o It won't work if the event consuming software lives in kernel space

The basic problems with using the filter mechanism are:

o Filters are applied after grabs, so if it's grabbed the filter is not invoked
    (I read the thread introducing filters; this makes sense)
o You can only have one filter at a time
    (there might already be one)

Has anyone tried to solve this problem before?  Any suggestions?

Thanks,
-- Terry