Hiroshi Inoue <Inoue(at)tpf(dot)co(dot)jp> writes:
> Oh I see. But this seems to change the behabior significantly
> at least for die signals.
Well, it considerably reduces the number of places at which either
signal will be accepted, but that's exactly the point. The code
as written was accepting the signals in many more places than we
envisioned in the original discussion, and I'm unconvinced that
that's safe.
AFAIK this should at worst increase the interrupt response time
from order-of-microseconds to order-of-milliseconds, so I'm not
especially worried. Sub-second response time is plenty good enough
for either kind of interrupt, IMHO.
regards, tom lane