Re: Problem with locale and keyboard

[Adrian Robert]

On Sep 28, 2004, at 9:40 PM, Rob Burns wrote:
> I'm using cvs from a couple days ago, and recently started having 
> problems with modifier keys. If I would like to press 'command-A' 
> (alt-shift-a on my keyboard), I now have to press alt first, then 
> shift, then a. previously I was pressing shift first, though I think 
> it worked in the reverse order as well.
>
> Currently, after pressing that key combination, my keyboard will be 
> stuck in 'command-lock' mode until I hit the alt key again. meaning, 
> that whenever I hit a character, its interpreted as if the command key 
> was also being pressed. I don't have any of these problems if the 
> shift key is not involved.  ie, pressing alt-a works as expected.

OK, I managed to replicate this on my machine and track it down.  I'm 
using an Apple keyboard with 'alt' in small letters at the top of the 
'option' key.  If I hit option alone, I get Alt_L out of X.  If I hit 
shift, then option, I get Meta_L.  If I hit option then shift, it is 
considered Alt_L with a shift.  Furthermore, once both shift and option 
are depressed, and you lift up on the option key first, you get a 
Meta_L-up.  If you lift up on shift first, you get an Alt_L-up.

You can verify that this is behind your behavior by running "xev" and 
experimenting with the combinations above.  The problem did not occur 
in GNUstep when modifiers were considered keycodes because the keycode 
for 'option' is the same no matter if shift is pressed or not.

This is a situation that Kazunobu Kuriyama pointed out, but I didn't 
fully see the implications of.  I think technically, the new 
implementation follows X most closely, however this is perhaps not 
desirable in this particular case.  There are three solutions that I 
can think of.  One is to revert to the old behavior of basing mapping 
on key codes.  This has disadvantage of ignoring the user's X modifier 
mapping settings as I originally complained of (but maybe the medicine 
is worse than the disease).  Another is to add a new back default 
"IgnoreShiftForModifiers" which sets the behavior to treat the Meta_L 
as Alt_L (unshifted).  This should be easy to implement.  Finally one 
could work around the issue in particular cases by using xmodmap with a 
line like ('keycode 64 = Alt_L Alt_L', perhaps paired with 'keysym 
Alt_L = Alt_L Meta_L' to retain some meta capability) to obliterate the 
shifted mapping.

I don't know which is best.  Can anyone who relies on shifted-modifiers 
on a regular basis state what they are used for and which of the above 
options would interfere least with this usage?

thanks,
Adrian