If a client is already running and has a mapped window when windowlab
starts, windowlab doesn't put a WM_STATE property on that client window.
I expected it would do so, the same as it does for clients which start
after windowlab.

This can be seen by having .xsession start an xterm, and then in that
xterm run windowlab.  Then

    xprop -id $WINDOWID

in the xterm shows no WM_STATE.

Usually this would be if a client program starts up faster than the
window manager.  It affects the click-to-select of xwininfo, xkill, and
similar.  They end up getting the windowlab frame rather than the
client, so xkill doesn't work, and xwininfo describes the frame whereas
for subsequently started clients it describes the client window.