home *** CD-ROM | disk | FTP | other *** search
- From: peter@ficc.ferranti.com (Peter da Silva)
-
- I find myself in violent agreement with most of the points raised here,
- but there are two areas I don't think have been adequately thought through,
- and I have one alternative suggestion.
-
- (mmap):
- > SVID 3 (System V Release 4.0), OSF AES (OSF/1), Berkeley BSD 4.4, MACH
- > 2.5, ULTRIX, SunOS, HP/UX, etc, all include an (almost) identical
- > definition of mmap() based on the current Sun/Berkeley specification.
-
- V.3.2 does not include this interface, and V.3.2 is going to be around,
- and in fact very common, for at least a couple of years to come. A standard
- that locks out V.3.2 is going to hurt portability.
-
- I believe that implementing shared memory in terms of memory mapped files
- is a good thing, and that every effort should be made to encourage it...
- within reason. Preventing the implementation of POSIX shared memory on
- the largest installed base of UNIX systems in existence is not within
- reason.
-
- If the interface is "mmap" with a set of constant parameters, this will
- serve to hurt portability.
-
- a) People with "real mmap" will end up using more than the
- standard allows. This is a pretty simple prediction... look
- at existing extended implementations.
-
- b) People with the "fake mmap" will end up blowing the extra
- parameters, "safely". Then when run on a system with the real
- thing their programs will fail.
-
- In practice, it would be best for both interfaces to (for portable
- programs, anyway) have a #define that hides the extra parameters. Which
- is exactly what the draft does.
-
- (IPC in general):
- I don't believe the interface requires a shared file system. "mkmq()"
- can broadcast a message to all machines to create the named file as a
- normal file and place in it a special token. This token can be read from
- the file descriptor returned by "open()", and used by the "mq*()" routines
- to indicate the real name of the message queue.
-
- I think it fairly important to put as many of the names of objects in the
- UNIX file system name space as possible. See pp 256,257.
-
- And one minor point (a wish-list entry):
-
- (sender_t in IPC):
- > The only information about the sender which would be useful
- > would be an indication about how to reply to the sender. In
- > this framework that would be a pathname;
-
- Or a file descriptor, perhaps? I had some questions about the utility of
- this feild, too, since it can't be used for any portable purpose. I'd rather
- see some meaning defined for it, though, than have it kicked out. Since
- returning an fd here is probably not a portable usage, this is a bit of a
- wish-list item...
- --
- `-_-' Peter da Silva. +1 713 274 5180. <peter@ficc.ferranti.com>
- 'U` Have you hugged your wolf today? <peter@sugar.hackercorp.com>
- @FIN Dirty words: Zhghnyyl erphefvir vayvar shapgvbaf.
-
- Volume-Number: Volume 20, Number 2
-
-