Path: utzoo!utgpu!news-server.csri.toronto.edu!rutgers!uwm.edu!cs.utexas.edu!swrinde!zaphod.mps.ohio-state.edu!uakari.primate.wisc.edu!aplcen!uunet!auspex!guy
From: guy@auspex.auspex.com (Guy Harris)
Newsgroups: comp.unix.wizards
Subject: Re: SVID and writev()
Message-ID: <3924@auspex.auspex.com>
Date: 20 Aug 90 21:42:40 GMT
References: <1990Aug19.210351.24632@dde.dk>
Organization: Auspex Systems, Santa Clara
Lines: 27

>I have not been able to find a specification of {IOV_MAX} anywhere. Unless
>I have overlooked something, it is not defined or mentioned elsewhere
>in the SVID.

{OPEN_MAX} isn't defined there either, because there's no official value
for it; it's implementation-dependent.  In that particular case, it's
specified by POSIX, and there's a way to query it with "sysconf()" (or
you can get it with "getrlimit()", but using "sysconf()" is more
portable).  There is, unfortunately, no "sysconf()" parameter in the
SVID to get it.

>Is it supposed to be defined in an include-file, and if so,
>which?

It's just a hardcoded 16 in 4.[23]BSD and 4.3-tahoe (quite literally
hardcoded - there's no #define, although there *is* an /* XXX */ on the
line containing the hardcoded 16), and the S5R4 documentation (as
opposed to the SVID) also says 16.  It may still be a hardcoded 16 in
S5R4, rather than a #defined constant.

>And what should be its minimum value?

I would not recommend building an implementation with a smaller value
than 16.  I don't know that there *is* some minimum value >1 that
somebody writing a program *using* "readv" or "writev", can count on,
but if you have to pick some such arbitrary value you should probably
pick 16.