Subject: Re: increasing FD_SETSIZE to 1024 or 2048?
To: John Kohl <jtk@kolvir.arlington.ma.us>
From: Todd Vierling <tv@pobox.com>
List: current-users
Date: 07/04/2000 18:48:51
On Tue, 4 Jul 2000, John Kohl wrote:

: This causes problems if the calling program increases FD_SETSIZE and its
: file descriptor process limit, and then expects high-numbered file
: descriptors to work properly with the RPC runtime's built-in small
: fd_sets.  To make it worse, the RPC runtime exports its fd_set as a
: global symbol the program can tweak.  All callers had better use the
: same set size!

As a follow-up, it seems that this is providing a *read-only* fd_set as an
export.  However, the type is "fd_set", not "fd_set *", which introduces yet
another problem--if you increase its size, a.out shared objects will lose on
binary compatibility (the COPY reloc problem already noted with
sys_errlist).  That's really broken!

This can be fixed, but it still doesn't negate prior arguments; the rpc
system could easily use a larger FD_SETSIZE or even use a dynamically sized
array of bytes, if necessary.

-- 
-- Todd Vierling (tv@pobox.com)