in our Tcl/Tk application, we are sending key events ( and ) over a socket to a host application which handles the actual key events.

This usually works quite well, as we can relate any events to the corresponding event. For example on my German keyboard I can type the glyph € by pressing AltGr+e. This results in the following events:

As you can see, %A, %K and %N might expand to different values when pressing resp. releasing the e key, but %k is stable, so we can use this to detect which release event corresponds to which press event.

Unfortunately, we are hitting a hard wall with keyboard layouts that have dead keys: in this case we are unable to relate events to the event(s).

E.g. on a German keyboard (with dead keys), I can type the letter á by pressing ´-a.

When I bind all resp bind all to some logging function, I get the following events (in order):

the first thing to notice is that the dead key ´ does not produce a KeyPress event (only a KeyRelease event). We can probably live with that, but what really baffles me is that when pressing the a none of the bind-variables have an obvious relation between the two events.

How can I know that the key that was producing the letter á was released? In a way that works across platforms (Linux, Darwin, Windows) and for unknown keyboard layouts (we've had complaints from French, Spanish and Portuguese people, but the community is really worldwide).