I'm fairly sure that's what he meant -- at least that's what I
read out of it too. (Though I would also pass along a void* for
the callback routine so that it can avoid searching for some struct,
and retval would be -ERROR as usual.)
The thing that has always bothered me about the threaded userland
approach is that the new thread requires a stack. With this
approach, we consume minimal resources as well as already being in
the right place to issue the signal for the i/o completion.
Anyway, it got me thinking enough to actually start on posix.4
signals (yeah, my book is a bit behind ;-), so that those would
be ready for whichever method is used for aio.
r~