step_signal sometimes left unset?

[Jim Blandy]

I've been looking at handle_inferior_event, and noticed a path that
seems wrong.  In the switch (ecs->ws.kind) case for
TARGET_WAITKIND_LOADED, control may continue on past the switch
without ever setting step_signal.  The code after the switch seems to
generally assume that step_signal has been initialized; we test it
directly after handling the two software single-step cases
(stepping_past_singlestep_breakpoint and stepping_past_breakpoint).

Is this really the right behavior?  When we proceed out of the
TARGET_WAITKIND_LOADED case, what should step_signal be set to?