Original keychord patch by Hai Nguyen <hhai2105@gmail.com>, updated for
dwm 6.4+ grabkeys() by Aaron Zueger <contact@azureorange.xyz>. This
revision rebases onto dwm 6.8 and fixes an event-loss bug in the chord
wait loop.
The wait loop used to discard every event that was not a KeyPress:
while (running && !XNextEvent(dpy, &event) && !ran)
if(event.type == KeyPress)
break;
Two consequences, both reproduced on Xephyr:
1. Unrecoverable input freeze. grabbuttons() grabs unfocused clients
GrabModeSync/GrabModeSync, so clicking one freezes pointer AND
keyboard until the grabbing client calls XAllowEvents. dwm does that
in buttonpress() -- but if the ButtonPress is discarded here,
buttonpress() never runs. The grab stays frozen forever and the loop
can never exit, because a frozen keyboard cannot deliver the KeyPress
it is waiting for. Only killing dwm ends it; XSendEvent is discarded
by this same loop and XTEST queues behind the frozen device.
Repro: press a chord prefix, then click any unfocused client.
Observed: XGrabKeyboard -> GrabFrozen, XQueryPointer -> Button1 held.
2. Orphaned windows. A MapRequest arriving mid-chord is dropped, so a
window opening during a chord is never managed. DestroyNotify,
Expose and ConfigureRequest are lost the same way.
Dispatch through handler[] as run() does, breaking on KeyPress before
dispatch so keypress() cannot recurse. The `&& !ran` term is dropped:
ran is always 0 on entry and nothing in the body can set it.
By default, when dwm is recompiled-restarted all clients will
lose it's current tag and collapse to first tag. This patch preserves
clients on old tags, however note that layout order is not preserved.
lupan
Lukasz Kasprzak