Re: hurd/term users.c

[Roland McGrath]

I see your point.  (Note that the behavior you had implemented in term was
no better on this mark.  There has never been a check on the peropen
modes.)  From what I can see, on Linux and FreeBSD select and poll will
tell you all about the states you don't have access to.  So they don't give
us much advice.  I am inclined to simply treat the disallowed accesses as
never happening, i.e. block forever if you select for read on an fd you
can't read.  I think this is the most consistent way to best serve the
programs that e.g. request both readability and writability on all their
fds both read-only and write-only.  Those programs want nothing select to
wake up only when some fd is ready for an allowed access.